CN103186650A - Searching method and device - Google Patents

Abstract

The invention provides a search method and device. The search method includes: setting key attributes and key attribute weights for the same type of documents, calculating the key attribute scores of each document; building an index posting list; each record in the document list in the index posting list includes a The document number and key attribute score of the document; the document list is composed of an ordered list and an unordered list, and the ordered list includes the n key attribute scores with the largest value, and the order of the key attribute scores is from large to small Arranged records; wherein, the n is a predetermined value; generate a corresponding etymology according to the search character string input by the user, retrieve the index posting list according to the generated etymology, and search according to the search string entered by the user Result scopes preferentially fetch records from the generated etymologically-corresponding ordered list to obtain the desired total number of search results and related results. The application of the invention can increase the search speed and reduce the occupation of system resources.

Description

A search method and device

technical field

The invention relates to the technical field of data services, in particular to a search method and device.

Background technique

Existing search engines generally sort based on text similarity unilaterally, and index storage is generally based on the form of key-value pairs <KEY, DocList>. Among them, KEY represents a keyword, and DocList represents a list of documents containing the keyword KEY. Each element in the DocList is a document object, which is used to store the basic information of a document, such as: the ID of the document, the number of occurrences of the keyword KEY in the document, and the position where it appears. When the user enters a keyword, it will first search based on the keyword; when the corresponding keyword is retrieved, then the overall calculation of the scores of all documents in the DocList corresponding to the keyword under this keyword; and then , sort all the search results as a whole according to the above scores, and then read the search results of the number required by the user from the sorted search results, and return the read search results to the user.

For example, there is a sorting method for comprehensive search results in the prior art. In this method, a sorting algorithm will be used to calculate the comprehensive value of vertical search engines, and the vertical search engines will be sorted according to the comprehensive value, and all vertical search engines will be summarized. Sort the results to generate the final search results.

However, the above sorting method puts the calculation process of the search result score in the search process, and performs full sorting each time, thus occupying a large amount of CPU and memory resources, high search complexity, and slow search speed. It is slow, and does not consider the value-added attribute of information, so it cannot guarantee that the information with high value will be ranked first. At the same time, since the method needs to store the factors for calculating the score and the factors for calculating the comprehensive score, it also requires a relatively high storage space.

In addition, there is also a method for sorting search results based on a search engine in the prior art. In this method, the score of each relevant resource is comprehensively calculated based on the pre-configured network resource weight and the text weight of the keyword input by the user, and then the full ranking is performed to generate search results. However, in this method, although the additional value attribute of information is taken into account, the scores of each document need to be calculated during the search process, and the search results also need to be fully sorted, thus occupying a large amount of CPU and memory resources. , the search complexity is also higher, and the search speed is also slower. At the same time, because this method also needs to store the factors for calculating the score and the factors for calculating the comprehensive score, it also has high requirements for storage space.

To sum up, in the search methods in the prior art, since the calculation and sorting of the document scores are done in the search process, and the sorting is done for the whole volume; moreover, the structure of the index organization does not consider the user's search habits , resulting in a large amount of stored information; in addition, each search needs to be performed on the full amount, which greatly increases the burden on scarce system resources such as memory and CPU. At the same time, the search methods in the prior art generally have high search complexity and slow search response speed. In addition, the search method in the prior art is based on text relevance, without considering the key attributes that can express the value of information, thus resulting in single sorting methods and insufficient user-friendliness.

Contents of the invention

In view of this, the present invention provides a search method and device, thereby improving the search speed and reducing the occupation of system resources.

The technical scheme that the present invention adopts is specifically realized like this:

A method of searching, the method comprising:

A. Set at least one key attribute and the corresponding key attribute weight for the same type of document, and calculate the key attribute score KFScore of each document according to the key attribute and key attribute weight;

B, all documents to be retrieved are indexed with etymology Term as a keyword, set up an index posting table with Term as a keyword index, with the total TotalCount of documents containing the Term and the list of documents DocList containing the Term as values; Each record in the document list includes a document number and a key attribute score; the document list is composed of an ordered list and an unordered list, and the ordered list includes n key attribute scores The records that are the largest and are arranged in descending order of key attribute scores; wherein, the n is a predetermined value;

C. Generate the corresponding etymology according to the search character string input by the user, search the index posting list according to the generated etymology, and prioritize the sequence corresponding to the generated etymology according to the range of search results input by the user Get the records in the list to get the desired total number of search results and related results.

The present invention also provides a search device, which includes: an inverted list generation module, a storage module, an etymology generation module and a retrieval module;

The inverted table generation module is used to index all documents to be retrieved with etymology Term as a keyword, and establishes Term as a keyword index, with the total TotalCount of documents containing the Term and the list of documents DocList containing the Term is an index posting list of values; each record in the document list includes a document number and a key attribute score; the document list is composed of an ordered list and an unordered list, and the ordered list Including n key attribute scores maximum, and according to the records of key attribute scores arranged in order from large to small; wherein, the n is a predetermined value; the index posting list is sent to the storage module;

The storage module is used to store the index posting list;

The etymology generation module is used to generate a corresponding etymology according to the search character string input by the user; the etymology and the range of search results input by the user are sent to the retrieval module;

The retrieval module is used to retrieve the index posting list stored in the storage module according to the generated etymology, and obtain records from the ordered list corresponding to the generated etymology according to the range of search results input by the user, To obtain the desired total number of search results and related results; sending the search results and the total number of related results to the user.

It can be seen from the above technical scheme that in the present invention, since the key attribute and the key attribute weight are set for the document, the KFScore of each document can be calculated, and an index posting list is established according to the KFScore of each document, so that the index posting list in the index posting list Document lists consist of ordered and unordered lists. When the user enters a search string, the corresponding etymology can be generated, and the index inverted list can be searched according to the generated etymology, and the ordered list corresponding to the generated etymology can be prioritized according to the range of search results entered by the user In order to obtain the required search results and the total number of relevant results, the value of various information can be measured in many ways under the premise of ensuring the relevance (that is, validity) of the search results, so that the value of valuable information The top ranking can greatly increase the search speed and the search response time of the search engine; at the same time, it can also reduce the occupation of system resources such as CPU and memory, thereby saving a lot of hardware and software resources.

Description of drawings

Fig. 1 is a flowchart of the search method in the present invention.

FIG. 2 is an information structure diagram of a document in the present invention.

Fig. 3 is a schematic structural diagram of an indexed inverted list in the present invention.

Fig. 4 is a flow chart of the single Term search method in the present invention.

Fig. 5 is a schematic structural diagram of an index posting list in a specific example of the present invention.

Fig. 6 is a flow chart of the multi-term search method in the present invention.

Fig. 7 is a flow chart of the multi-layer search method in the present invention.

FIG. 8 is a flowchart of an implementation method of step 702 in the present invention.

Fig. 9 is a schematic structural diagram of the search device in the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention more clearly, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

Fig. 1 is a flowchart of the search method in the present invention.

As shown in Figure 1, the method includes:

Step 101, setting at least one key attribute (KeyField) for the same type of document and pre-setting the key attribute weight of each key attribute.

In the field of data business, a collection of information describing a thing can be called a document (Document). FIG. 2 is an information structure diagram of a document in the present invention. As shown in FIG. 2 , in the technical solution of the present invention, each document includes multiple etymologies (Term) and multiple attributes (Field); wherein, the attributes may be composed of one or more Term. For example, a document may include attributes such as "title" and "content", wherein the "title" attribute may be composed of one or more Term. For another example, attribute 1 in FIG. 2 is composed of N etymologies, attribute J is composed of M etymologies, and so on.

In addition, in the technical solution of the present invention, one or more key attributes for identifying important information of the document will be set for each document, and documents of the same type have the same key attributes. Table 1 below is an illustration of the key attributes of the present invention.

Document category Key attribute 1 Key attribute 2  … Commodity documents commodity price Number of purchases  … Thesis documents Citations download times  … Music Documentation Trial times download times  … ... ... ...  …

Table 1

As described in Table 1, for commodity documents, commodity price and/or purchase times can be set as key attributes; for paper documents, citation times and/or download times can be used as key attributes; for music documents, you can set Trials and/or downloads are set as key attributes.

After the key attributes are set, corresponding weights Weight (may be referred to as key attribute weights) will be predetermined for each set key attribute according to actual application conditions.

Step 102, calculating the key attribute score (KFScore) of each document according to the key attribute and the key attribute weight.

After the above key attributes and key attribute weights are set, the key attribute score (KFScore) of each document can be calculated according to the set key attributes and key attribute weights.

For example, the KFScore of a document can be calculated by formula (1) as follows:

KFScore＝KeyField ₁ *W ₁ +KeyField ₂ *W ₂ +......+KeyField _X *W _X

Wherein, W ₁ +W ₂ +...+W _X ＝1 (1)

The KeyField ₁ represents the value of the first KeyField of the document, the W ₁ represents the value of the Weight corresponding to the first KeyField of the document, and the KeyField _x represents the value of the x-th KeyField of the document , the W _x represents the value of Weight corresponding to the xth KeyField of the document, and so on.

Step 103, all the documents to be retrieved are indexed with Term as the keyword (Key), and the index is set up with Term as the keyword, with the total number (TotalCount) of the documents containing the Term and the document list (DocList) containing the Term as the value (Value) index inverted list InvertIndexList.

In this step, an index posting list will be created. Specifically, an index inversion list with Term as the keyword index and key-value pair <TotalCount, DocList> as the value will be established. The index inversion list can be called InvertIndexList.

Fig. 3 is a schematic structural diagram of an indexed inverted list in the present invention. As shown in Figure 3, the index inverted table InvertIndexList among the present invention includes one or more records, and each record can be stored in a key-value pair <Key, Value> mode, and can include two fields in each record: the key Word (Key) field and key value (Value) field. Wherein, the Key field is used to store the Term corresponding to the document, and the Value field is used to store the key-value pair <TotalCount, DocList> corresponding to the Term in the Key field. Among them, DocList is a list of documents containing the Term, and TotalCount is the total number of documents in DocList.

As shown in FIG. 3 , in a specific embodiment of the present invention, the DocList may be a linked list, and TotalCount is the total number of elements in the linked list. The linked list includes multiple records, and each record is stored as a key-value pair <KFScore, Did>, and each key-value pair corresponds to a document containing the corresponding Term. Wherein, the KFScore represents the key attribute score of the document, which can be calculated by the formula (1) in step 102; the Did represents the identification of the document, for example, the serial number of the document.

In addition, in a specific embodiment of the present invention, for each DocList in the index inverted list InvertIndexList, the records in the same DocList can be partially sorted in advance according to the size of the KFScore, that is, the first n KFScore maximum records, and the first n KFScore maximum records are arranged in order of KFScore size to form an ordered list (OrderedList); other records in the DocList form an unordered list (DisorderedList). Therefore, in the specific embodiment of the present invention, each DocList is composed of an ordered list and an unordered list, and the ordered list includes n key attribute scores that are the largest, and are ordered from large to small according to the key attribute scores. sorted records. Wherein, the n is a predetermined value, therefore, the n is the length value of the OrderedList.

It can be seen that, in a specific embodiment of the present invention, it is not necessary to sort all the records in the DocList by KFScore size, but only need to perform partial sorting, that is, only need to determine the record with the largest KFScore of the first n, and sort the first n The record with the largest KFScore (that is, the record in the ordered list) can be sorted according to the size of the KFScore, and the other records in the DocList (that is, the records in the unordered list) can not be sorted.

Step 104, generate the corresponding etymology according to the search character string input by the user, search the index posting list according to the generated etymology, and prioritize the corresponding etymology from the generated etymology according to the range of search results input by the user. Records in the sequence table to get the desired search results and the total number of related results.

In the technical solution of the present invention, after the index posting list is established through the above step 103, the index posting list can be retrieved according to the search character string and the search result range input by the user to obtain the desired search result and the total number of related results.

Specifically, firstly, a corresponding etymology can be generated according to the search character string input by the user.

For example: when the search string entered by the user is "Andy Lau", the search string "Andy Lau" can be converted into the corresponding etymology: "Andy Lau" or "liudehua" according to different conversion rules.

When the search string entered by the user is "Andy Lau's will", the search string "Andy Lau's will" can be converted into the corresponding etymology according to different conversion rules: "Andy Lau's will" or "liudehuatianyi"; The search string "Andy Lau God's Will" is converted into two corresponding etymologies: "Andy Lau" and "God's Will".

After the corresponding etymology is generated, the index postings can be retrieved according to the etymology, and it is judged whether the above etymology is stored in the index postings.

When the etymology is stored in the index posting table, the DocList corresponding to the etymology can be read, and the record is preferentially obtained from the ordered list of the DocList according to the search result range input by the user, to obtain The total number of desired search and related results.

Wherein, the range of the search result is the range where the search result desired by the user is located. Therefore, the search result range includes: an initial position (InitialPositon) and the number of items (Count), and the maximum value of the search result range is: (InitialPositon+Count)-1. Wherein, the starting position is the starting position of the search result, and the number of items is the number of the search result.

For example, after a user enters a search string to perform a search, the search result will be viewed in the form of a web page. Take each webpage displaying Count=30 search results as an example, when the user wants to view the search results on the first page, InitialPositon=1, Count=30, the range of the search results is: the first to 30 search results; And when the user clicks "next page" after viewing the search results of the first page, and wishes to view the search results of the second page, then InitialPositon=31, Count=30, the range of the search results is: the 31st～ 60 search results. And so on.

Therefore, in a specific embodiment of the present invention, said obtaining records from the ordered list of DocList preferentially according to the range of search results input by the user, so as to obtain the required search results may include:

When the maximum value of the search result range is less than or equal to the length n of the ordered list, the search result range at this time is included in the ordered list, so it can directly start from the InitialPositon record in the ordered list , to read Count records as search results without reading the records in the unordered list. Since the records in the ordered list have been arranged in the order of KFScore size, the Count records read are the search results arranged in order of KFScore size.

When the starting position of the search result range is in the ordered list (that is, InitialPositon≤n), and the maximum value of the search result range is greater than the length n of the ordered list, a part of the search result range at this time is located in In the sequence list, the other part is located in the unordered list. At this time, the (n-InitialPositon+1) record can be read from the InitialPositon record in the ordered list, and then the records in the unordered list Arrange according to the size order of KFScore, and read [Count-(n-InitialPositon+1)] records starting from the first record in the unordered list; use the read Count records as the search result. At this time, the Count records read are also the search results arranged in order of KFScore size.

When the starting position of the search result range is in the unordered list (that is, InitialPositon>n), the search result range at this time is all located in the unordered list, so the records in the unordered list can be Arrange according to the size of KFScore, and then start from the (InitialPositon-n)th record in the sorted unordered list, read Count records as the search results. At this time, the Count records read are also the search results arranged in order of KFScore size.

Through the above methods, the desired search results can be obtained.

Considering the user's search habits, generally the user only cares about the top dozens of search results. According to the analysis of statistical results of users' search habits by authoritative organizations, more than 90% of users generally only view the first two pages of search results (generally about 20 search results). Therefore, when the amount of search data is hundreds of millions of levels, if the value of the length n of the ordered list exceeds a certain value (for example, n=100), then in most cases, the search results entered by the user The ranges are included in the ordered list, so that the required number of search results that have been arranged in order of the size of the KFScore can be obtained directly from the ordered list, without the need to sort the search results, and the required memory space is relatively small. The number of CPU operations is less, the response time is higher, and the system resources occupied are less, which can greatly improve the search response time of the search engine, and at the same time save a lot of hardware and software resources.

Furthermore, in a specific embodiment of the present invention, when the etymology generated according to the search string input by the user is not stored in the index posting table, an empty set may be returned to the user as the search result, indicating that there is no Find related documents.

In addition, in the technical solution of the present invention, the above step 104 can be implemented in various manners, that is, various search methods can be used to implement the above step 104. In the following, various search methods will be introduced respectively in the form of multiple specific embodiments.

Embodiment 1: Single Term search method.

In this embodiment, the single-term search method refers to a method of searching when only one corresponding etymology is generated according to the search character string input by the user. Fig. 4 is a flow chart of the single Term search method in the present invention. As shown in Figure 4, the single Term search method may include the steps as follows:

Step 401, generate a corresponding etymology according to the search character string input by the user.

In this step, the search character string input by the user may be used as a word, and the word may be directly used as the corresponding etymology of the search character string. For example, when the search character string input by the user is "Andy Lau", an etymology corresponding to the search character string: "Andy Lau" is generated.

Step 402 , according to the generated etymology retrieval index posting table, judge whether the etymology is stored in the index posting table; if yes, execute step 403 ; otherwise, execute step 415 .

Step 403, read the document list corresponding to the etymology from the index posting list.

Because all include a list of documents (DocList) in the corresponding value (Value) of each etymology in the index posting list, therefore, can read from the described index posting list according to the above-mentioned etymology generated in the step Get a list of documents corresponding to said etymology.

Step 404, analyzing the search result range input by the user to obtain the starting position and the number of items.

In this step, the initial position InitialPositon and the number of items Count can be obtained by analyzing the search result range input by the user.

Step 405, judging whether (InitialPositon+Count)-1>n holds true, if yes, then execute step 406; otherwise, execute step 412;

Because the maximum value of the search result range is (InitialPositon+Count)-1, so in this step, it will first be judged whether the maximum value of the search result range is greater than the length n of the ordered list of the document list, that is, to judge (InitialPositon+Count) Count)-1>n is established.

When the maximum value of the search result range is greater than the length of the ordered list of the document list, it means that the search result range has exceeded the range of the ordered list. At this time, it will not be possible to read all the required items from the ordered list Number of search results, therefore, will continue to perform step 406;

When the maximum value of the search result range is less than or equal to the length of the ordered list of the document list, it means that the search result range does not exceed the range of the ordered list. At this time, all documents can be read only from the ordered list. required search results, therefore step 412 can be executed to directly obtain the required number of search results from the ordered list.

Step 406, judging whether InitialPositon>n holds true, if yes, execute step 410; otherwise, execute step 407;

In this step, it will be judged whether the InitialPositon in the search result range is greater than the length n of the ordered list.

When the InitialPositon in the search result range is greater than the length of the ordered list, it means that the InitialPositon in the search result range is already in the unordered list, the search results are all in the unordered list, and the ordered list does not Include search results, so step 410 can be performed;

When the InitialPositon in the search result range is less than or equal to the length of the ordered list, it means that the InitialPositon in the search result range is still in the ordered list, that is, some search results are in the ordered list, but some search results are in the in an unordered list. At this time, step 407 can be executed.

Step 407, starting from the InitialPositon'th record in the ordered list, read (n-InitialPositon+1) records and add them to the search result list (ResultList).

Step 408, arrange the records in the unordered list according to the size of the KFScore.

Step 409, read [Count-(n-InitialPositon+1)] records from the unordered list and add them to ResultList, and execute step 413.

Step 410, arrange the records in the unordered list according to the size of the KFScore.

Step 411, starting from the (InitialPositon-n)th record in the sorted unordered list, read Count records and add them to the ResultList, and execute step 413.

Step 412, starting from the InitialPositon record in the ordered list, read Count records and add them to the ResultList, and execute step 413.

In step 413, the ResultList is used as the search result, and the total number of documents TotalCount of the document list corresponding to the etymology read from the index posting list is used as the total number of related results SumCount.

Step 414, returning the search result and the total number of related results to the user. End the process.

Step 415, take the empty set as the search result, set SumCount=0, and return the empty set and the total number of related results to the user. End the process.

Through the above steps 401-415, a single-term search can be realized.

The following will take a specific example in the field of music search as an example to further introduce the above-mentioned single-term search method.

Specific example one:

Suppose there are 10 million music files to be retrieved in the field of music search, and the music information of each file includes: song name (SongName), singer name (SingerName), album name (AlbumName), lyrics, song trial listening volume ( ListenCount) and business volume (BusinessCount).

In the example of the present invention, described ListenCount and BusinessCount can be set as the key attribute of above-mentioned music file, and key attribute weight is set to 0.3 and 0.7 respectively, then the key attribute score value (KFScore) of each music file can be passed as follows Formula calculation:

KFScore＝ListenCount*0.3+BusinessCount*0.7

According to the above music files to be retrieved, key attributes and key attribute scores, an index inverted list InvertIndexList can be established, and the length n of the ordered list in the index inverted list can be set to 100.

Fig. 5 is a schematic structural diagram of an index posting list in a specific example of the present invention. As shown in Figure 5, the value corresponding to the etymology "Andy Lau" is: <10000, DocList>, therefore, the total number of documents TotalCount=10000 in the DocList corresponding to the etymology "Andy Lau", the ordered list in this DocList The length n=100, that is, the ordered list includes 100 records; the length of the unordered list is 9900, that is, the unordered list includes 9900 records; similarly, the corresponding value of the etymology "I and you" is: <9870, DocList>, therefore, the total number of documents TotalCount=9870 in the DocList corresponding to the etymology "I and you"; ...; The corresponding value of the etymology "God's Will" is: <60, DocList >, therefore, the total number of documents TotalCount=60 in the DocList corresponding to the etymology "God's Will"; wherein, it can be further assumed that among the above-mentioned 60 documents, the singer in 20 documents is named "Andy Lau".

When the search string entered by the user is "Andy Lau", and the search result range entered by the user is: the 31st to 50th search results, the following single-term search method can be used to search:

Generate the corresponding etymology "Andy Lau" according to the search character string input by the user; analyze the search result range input by the user, and learn that the starting position InitialPositon=31, and the number of entries Count=20;

Search the index posting table shown in Figure 5 according to the generated etymology "Andy Lau", and find the etymology "Andy Lau" corresponding to the above-mentioned generated etymology in the index posting table; the value corresponding to the etymology is: <10000, DocList>, therefore, the total number of documents in the DocList corresponding to the etymology "Andy Lau" TotalCount=10000, the length of the ordered list in the DocList is n=100, that is, 100 records are included in the ordered list ; The length of the unordered list is 9900, that is, the unordered list includes 9900 records.

Since (InitialPositon+Count)-1=31+20-1<n=100, the search result range is included in the ordered list, so you can directly start from the 31st record in the ordered list and read 20 records as search results; meanwhile, the total number of relevant results is SumCount=TotalCount=10000.

Since the records in the ordered list have been arranged in order of size of KFScore, the read 20 records are the search results arranged in order of size of KFScore.

Embodiment 2: Multi-Term search method.

In actual application scenarios, users may enter multiple keywords to search, hoping to locate the content they want to find more accurately. At the same time, some users may also have specific requirements for the search results. Information that includes keyword A and keyword B, or information that includes keyword A or keyword B needs to be found, or information that includes keyword A but does not include keyword B needs to be found, and so on. Therefore, there may be three logical operations of AND (AND), OR (OR) and difference (SUB) among the multiple keywords input by the user.

In order to meet the needs of the above users, in the technical solution of the present invention, a multi-term search method is also proposed. In this embodiment, the multi-term search method refers to a method of searching when multiple corresponding etymologies are generated according to the search character string input by the user. Fig. 6 is a flow chart of the multi-term search method in the present invention. As shown in Figure 6, the multi-Term search method may include the steps as follows:

Step 600, setting the Y-segment reading range in the document list of the index posting list.

In a specific embodiment of the present invention, the segmented read strategy is used for retrieval, so as to avoid operating on the full amount of data for each retrieval as much as possible. Therefore, in this step, the reading range of the Y segment will be preset in the document list of the index posting list. Among them, Y≥2.

For example, when Y=2, a 2-segment reading range will be preset in the document list of the index posting list. At this time, the ordered list of the document list can be used as the first reading range, and the entire document list can be used as the second reading range.

For another example, when Y=3, the reading range of 3 segments will be preset in the document list of the index posting list. At this time, the ordered list of the document list can be used as the reading range of the first paragraph, the unordered list of the document list can be used as the reading range of the second paragraph, and the entire document list can be used as the reading range of the third paragraph.

In a specific embodiment of the present invention, other methods can also be used to set the reading range according to the actual application situation, and the specific setting method will not be repeated here.

Step 601, generate an etymology queue and query logic sequence according to the search character string input by the user.

In this step, the search string entered by the user will be parsed to generate an etymology queue TermArray and a query logic sequence SetOperators. Wherein, the etymology queue TermArray may include x etymologies, which may be expressed as TermArray{Term ₁ , Term ₂ , . . . , Term _x }. The query logic sequence SetOperators includes (x-1) query logic symbols, which can be expressed as SetOperators {symbol ₁ , symbol ₂ , . . . , symbol _x-1 }. Wherein, the logical relationship among the various etymologies in the etymology queue is represented by the corresponding query logic in the query logic sequence. For example, symbol ₁ in the query logic sequence represents the logical relationship between Term ₁ and Term ₂ , symbol ₂ represents the logical relationship between Term ₃ and the previous Term ₁ and Term ₂ , and so on. In addition, the value of the query logic symbol is: AND, OR or SUB.

For example, when the search string entered by the user is "Andy Lau OR God's Will", the above search string can be parsed to generate an etymology queue TermArray{Andy Lau, God's Will} and a query logic sequence SetOperators{symbol ₁ }, wherein, The value of symbol ₁ is OR.

Step 602: Analyze the search result range input by the user to obtain the starting position and the number of items.

In this step, the initial position InitialPositon and the number of items Count can be obtained by analyzing the search result range input by the user.

Step 603, judge whether the length of TermArray is greater than 1; if yes, execute step 604; otherwise, execute step 616.

In this step, at first will judge whether the length of TermArray is greater than 1; If yes, then explain that TermArray includes more than two etymology at least, will continue to perform step 604; If not, only include an etymology in the explanation TermArray, therefore It is possible to search only by using a single-term search method, instead of using a multi-term search method, so step 616 is executed.

It can be seen from the above step 603 that, in a specific embodiment of the present invention, the single-term search method is a special case of the multi-term search method.

Step 604, read the first two etymologies in TermArray, and set the variable i with an initial value of 2 and the variable i with an initial value of 1.

In this step, the first two etymologies, Term ₁ and Term ₂ , will be read directly from the TermArray.

In addition, in this step, two variables may also be preset: a variable i with an initial value of 2 and a variable j with an initial value of 1. Among them, i can be used to indicate that the current reading is the i-th etymology, and j can be used to indicate the number of paragraphs in the current reading range, that is, it means that the current reading range is the reading range of the jth paragraph.

Step 605: Retrieve the index posting list according to the two etymologies read above, and respectively read the reading range of the jth paragraph in the document list corresponding to the two etymologies from the index posting list Records in , store the read records in the first search result list (ResultList1) and the second search result list (ResultList2) respectively.

Specifically, in this step, the two etymologies read above can be regarded as two independent etymologies, and the index posting list is retrieved according to each etymology. If the above two etymologies are stored in the index posting list, the records within the read range of the jth paragraph in the document list corresponding to the two etymologies are respectively read from the index posting list. Then, the records in the reading scope of the first section j in the document list corresponding to the first etymology (i.e. Term ₁ ) that are read are stored in ResultList1, and the second etymology (i.e. Term 1) that is read is stored in ResultList1. ₂ ) Records within the reading range of the jth paragraph in the corresponding document list are stored in ResultList2.

In this step, j=1 at this time, so all the records read are within the reading range of the first paragraph in the document list. If in step 600 the reading scope of the first segment is set as an ordered list, all the records read at this time are the records in the ordered list of the document list.

Furthermore, if any one of the above two etymologies is not stored in the index posting list, the retrieval result list corresponding to the unstored etymology can be set as an empty set. For example, if the first etymology (i.e. Term ₁ ) in the above two etymologies is not stored in the index posting list, then the retrieval result list ResultList1 corresponding to the etymology is set to an empty set; if the index posting list If the second etymology (ie Term ₂ ) of the above two etymologies is not stored in , then the retrieval result list ResultList2 corresponding to the etymology is set as an empty set.

Step 606, read the (i-1)th logic symbol symbol _i-1 in the query logic sequence SetOperators.

Step 607, judge the value of the logical symbol read; when the value of the logical symbol is AND, execute step 608; when the value of the logical symbol is OR, execute step 609; When the value of the symbol is SUB, step 610 is executed.

In step 608, ResultList1 and ResultList2 are merged using AND (AND) logic, and the merged result is added to the search result list (ResultList); step 611 is executed.

In this step, ResultList1 and ResultList2 will be merged using AND logic, and the merged result will be added to ResultList.

Since each record in ResultList1 and ResultList2 includes two attributes: key attribute score (KFScore) and document identifier (Did), the purpose of logical merging is to find out records with the same Did from the two retrieval result lists. Due to the particularity of KFScore, if the KFScores of two documents are the same, the two documents may belong to the same document, and if the KFScores of the two documents are different, they must not be the same document.

Based on the above reasons, in a specific embodiment of the present invention, the use and logic of merging ResultList1 and ResultList2, and adding the merged result to ResultList may include:

Compare the records in ResultList1 and ResultList2 one by one, and add the two records with the same KFScore and Did in the two search result lists as a search result to the search result list (ResultList).

Step 609, use OR logic to merge ResultList1 and ResultList2, and add the merged result to the search result list (ResultList); execute step 611.

In this step, the OR logic will be used to merge ResultList1 and ResultList2, and the merged result will be added to ResultList.

For example, in a specific embodiment of the present invention, the use or logic to merge ResultList1 and ResultList2, and adding the merged result to ResultList may include:

Add each record in the ResultList1 and ResultList2 to the search result list (ResultList), if there are two records with the same KFScore and Did, delete one of the records in the ResultList.

Step 610, use difference (SUB) logic to merge ResultList1 and ResultList2, and add the merged result to the search result list (ResultList); execute step 611.

In this step, ResultList1 and ResultList2 will be merged using difference logic, and the merged result will be added to ResultList.

For example, in a specific embodiment of the present invention, the use of difference logic to merge ResultList1 and ResultList2, and adding the merged result to ResultList may include:

Each record stored in ResultList2 is removed from the ResultList1, and the removed records in ResultList1 are added to the search result list (ResultList).

Step 611 , judging whether there are unused Term in TermArray; if yes, go to step 612 ; otherwise, go to step 613 .

Step 612, set i=i+1, read the i-th etymology in TermArray; Clear ResultList1 and ResultList2, and copy the records in ResultList to ResultList2; Retrieve the index according to the etymology read table, read the records in the jth paragraph reading range in the document list corresponding to the etymology read from the index posting list, store the read records in ResultList1; return to execute step 606 .

Specifically, in this step, i=i+1 can be set earlier to read the next etymology (i.e. the i-th etymology) from TermArray; The results (that is, the records in ResultList) are copied to ResultList2, so that the previous search results and the search results corresponding to the next etymology read can be logically operated in the subsequent search process.

Then, the index posting list is retrieved according to the etymology read. If the above-mentioned etymology read is stored in the index posting list, then read the record in the jth paragraph reading range in the document list corresponding to the etymology from the index posting list, and The read records are stored in ResultList1, and then return to step 606.

In this step, if j=1 at this time, what is read is the record within the reading range of the first paragraph in the document list. However, if the reading scope of the first paragraph is set as an ordered list in step 600, then the records in the ordered list of the document list will be read at this time. Similarly, if j=2 at this time, what is read is the record within the reading range of the second paragraph in the document list. However, if the reading range of the second paragraph is set to the entire document list in step 600, then the records in the entire document list will be read at this time. And so on, no more details here.

Furthermore, if the read i-th etymology is not stored in the index posting list, the ResultList1 may be set as an empty set, and then return to step 606 .

Step 613, judge whether the end condition is satisfied; if yes, execute step 617; otherwise, execute step 614.

Wherein, in a specific embodiment of the present invention, the end condition may be:

The number of records stored in ResultList is greater than the maximum value of the search result range: (InitialPositon+Count)-1; or, j is equal to Y.

In the technical scheme of the present invention, when the number of records stored in the ResultList is greater than the maximum value (InitialPositon+Count)-1 of the search result range, it means that a sufficient number of search results have been retrieved, thus satisfying the requirement for ending the search condition. And when j is greater than Y, it means that the last segment of the search has been carried out, and the search can also be ended at this time, thereby satisfying the end condition.

Step 614, empty ResultList1 and ResultList2; Read the first two etymologies in the TermArray, and set i=2 and j=j+1; Retrieve the index inverted list respectively according to the two etymologies read above, from the In the above-mentioned index posting list, read respectively the record in the jth paragraph reading range in the document list corresponding to the two etymology, store in the first retrieval result list (ResultList1) and the second retrieval result list ( ResultList2).

Since the judgment result of step 613 is that the end condition is not met, it means that the number of records stored in the ResultList is less than the required number of bars, that is, in the retrieval process of the previous stage, a sufficient number of search results has not been retrieved, Therefore, it is necessary to carry out the retrieval process of this stage (that is, the jth stage).

Therefore, in this step, the index posting list will be searched again according to the two etymologies read above. If the above two etymologies are stored in the index posting list, the records within the read range of the jth paragraph in the document list corresponding to the two etymologies are respectively read from the index posting list. Then, the records in the reading scope of the first section j in the document list corresponding to the first etymology (i.e. Term ₁ ) that are read are stored in ResultList1, and the second etymology (i.e. Term 1) that is read is stored in ResultList1. ₂ ) Records within the reading range of the jth paragraph in the corresponding document list are stored in ResultList2.

Step 615, respectively arrange the records in the ResultList1 and ResultList2 according to the size of the KFScore; return to execute step 606.

Step 616, using the above-mentioned single-term search method, according to the etymology in the TermArray, retrieve the index posting list to obtain the search results and the total number of related results; execute step 618.

Step 617, read the required records from the ResultList as search results, and use the number of records stored in the ResultList as the total number of relevant results (SumCount).

For example, in a specific embodiment of the present invention, reading the required records from the ResultList as search results includes:

When SumCount≥(InitialPositon+Count)-1, start from the first InitialPositon record in the ResultList, and read the Count record as the search result;

When InitialPositon≤SumCount<(InitialPositon+Count)-1, it indicates that the number of records in the ResultList is insufficient. At this time, start from the InitialPositon record in the ResultList and read (SumCount-InitialPositon+1) records As a search result,

When SumCount<InitialPositon, it indicates that the number of records in the ResultList is too small, and there is no desired search result. At this time, the empty set can be used as the search result.

Step 618, return the search result and SumCount to the user.

In the above-mentioned multi-term search method, a segmentation search strategy is used. For example, when the 2-segment reading range is set in advance, and the ordered list of the document list is used as the 1st reading range, and the entire document list is used as the 2nd reading range, the document list will be searched first. Sequence list; if the number of search results obtained from the ordered list is greater than or equal to the required number of items, it means that a sufficient number of search results have been retrieved in the first paragraph of the search process, and there is no need to search for the unordered list at this time. search, but can directly give the sorted search results; only when the number of search results in the ordered list is less than the required number of items, that is, when a sufficient number of search results are not retrieved in the first paragraph of the search process, the A second stage of the search is performed, searching from all records in the document list.

Therefore, using the above-mentioned multi-term search method, it is possible to avoid performing retrieval operations on the entire data set (for example, the entire document list) each time as much as possible, thereby effectively reducing the search space and reducing the complexity of the search.

For example, in the prior art, if the level of search quantity is 100 million (for example, web page or news search is generally T level), each search in the current open source search (for example, Lucene) will search the results Sets (such as million levels) are fully sorted, and then the result set searched by users is obtained on this basis. However, if the above-mentioned multi-term search method is used, and the ordered list of the document list is used as the reading range of the first paragraph, then in most cases, the search results that meet the user's needs can be found at the end of the first stage of the search process, without using The second stage of search is performed, so the amount of data required to be searched can be drastically reduced, and there is no need to sort the search results during the retrieval process, thereby greatly improving the search response speed (generally an order of magnitude higher than Lucene's search response speed) , and can also greatly reduce the occupation of memory and CPU resources.

Through the above steps 601-618, multi-term search can be realized.

The following will take a specific example in the field of music search as an example to further introduce the above-mentioned multi-term search method.

Specific example two:

For the sake of simplicity, in this specific example, the basic settings in the first specific example are still used, and the index posting list shown in FIG. 5 is established, and the length n of the ordered list in the index posting list is set to 100. In addition, it is assumed that a 2-segment reading range is set in the document list of the index posting list (that is, Y=2), and the ordered list of the document list is used as the first reading range, and the entire document list is used as the second segment. Segment read range.

When the search string entered by the user is "Andy Lau AND God's Will", and the search result range entered by the user is: the 1st to 10th search results, the following multi-term search method can be used to search:

Generate an etymology queue TermArray{Andy Lau, Providence} and a query logic sequence SetOperators{AND} according to the search string entered by the user;

Analyze the range of search results entered by the user, and learn that the starting position InitialPositon=1, and the number of entries Count=10;

Retrieve the index inverted list shown in Figure 5 according to the generated etymology "Andy Lau" and "God's Will" respectively, and find the corresponding etymology "Andy Lau" and "God's Will" generated above;

According to the index inverted table shown in Figure 5, for the etymology "Andy Lau", its corresponding value is: <10000, DocList>, therefore, the total number of documents in the DocList corresponding to the etymology "Andy Lau" TotalCount=10000. Therefore, the ordered list in the DocList corresponding to the etymology "Andy Lau" will be read from the index inverted list, and all 100 records in the ordered list will be stored in ResultList1;

According to the index posting table shown in Figure 5, it can be seen that for the etymology "God's Will", its corresponding value is: <60, DocList>, therefore, the total number of documents in the DocList corresponding to the etymology "God's Will" TotalCount=60; Owing to 60＜100, so, the document in the DocList corresponding to etymology " God's will " is all stored in the ordered list; "The ordered list in the corresponding DocList, all 60 records in the described ordered list are all stored in ResultList2;

Read the first logical symbol from the logical sequence SetOperators, because the logical symbol is AND, so use AND logic to merge ResultList1 and ResultList2, and the merged result is added to ResultList;

Since 20 of the above 60 documents corresponding to "God's Will" have singers named "Andy Lau", the above ResultList will store 20 records, and the total number of related results is: SumCount=20.

Since all the Term in TermArray have been used, and (InitialPositon+Count)-1=1+10-1<SumCount, that is, the search result range is included in the ordered list, so the end condition has been satisfied, at this moment, Directly start from the first InitialPositon=1 record in the ResultList, and read Count=10 records as the search result;

Finally, the total SumCount=20 of the above search results and related results will be returned to the user.

Embodiment 3: Hierarchical search method.

In this embodiment, the hierarchical search method refers to dividing the entire search process into multi-layer search processes according to preset rules, and there is a strict distinction between the search processes at each level. When searching between different layers, The fields or search rules are different. Generally speaking, the score of the upper layer is higher than that of the lower layer, and the rules for sorting according to the key attributes of the information are specified inside each layer. In this way, the hierarchical search method can not only maintain the relevance of the search, but also ensure the search of important information. Post-results top-ranked questions.

Fig. 7 is a flow chart of the multi-layer search method in the present invention. As shown in Figure 7, the multi-layer search method may include the following steps:

Step 701, pre-setting multi-layer search processes according to preset search ordering rules, and setting the priority of each layer search process and the hierarchical search score range of each layer search process.

In this step, corresponding search sorting rules can be preset according to business characteristics. For example, in a specific embodiment of the present invention, for the field of music search, the following search sorting rules can be preset: precise search, spelling search And participle search. According to the above search sorting rules, we can know that in the field of music search, precise search can be carried out first, if there is not enough search results, then full spell search can be carried out, and if there is still not enough search results, word segmentation search can be carried out finally , trying to get the search results desired by a sufficient number of users.

Therefore, according to the aforementioned preset search ordering rules, a multi-layer search process can be set. For example, in the field of music search, the following three-layer search process can be set: the precise layer search process, the full spelling layer search process and the word segmentation layer search process, and the priority of each layer search process is in descending order: Exact layer search process, full spell layer search process, word segmentation layer search process.

Furthermore, in a specific embodiment of the present invention, the above-mentioned search process of each layer may be further divided into multiple sub-layer search processes, and the priority of each sub-layer search process may be set. For example, the above precise layer search process can be divided into three sub-layer search processes arranged in order of priority: song sub-layer search process, singer sub-layer search process and album sub-layer search process. That is, when performing the precise layer search process, the song sub-layer search process will be performed first, then the singer sub-layer search process will be performed, and finally the album sub-layer search process will be performed to complete the precise layer search process. Similarly, the search process at the full spelling layer and the search process at the word segmentation layer can also be divided into the above three sub-layer search processes. I won't repeat them here.

In addition, in a specific embodiment of the present invention, the hierarchical search score ranges of each layer search process can be further preset, so as to facilitate the sorting of subsequent retrieved results. For example, when there are 3 layers of search processes, and the priorities of each layer of search processes in descending order are: the first layer search process, the second layer search process, and the third layer search process, the first The hierarchical search score range of the layer search process is set to: [A1, A2], which means that the hierarchical search scores of all retrieval results obtained in the first layer search process will be between A1 and A2; The hierarchical search score range of the second-level search process is set to: [B1, B2], and the hierarchical search score range of the third-level search process is set to: [C1, C2]. Wherein, A1>A2>B1>B2>C1>C2. Therefore, the search score of any retrieval result obtained in the first-level search process will be higher than that of the retrieval results obtained in the second and third-level search processes. Through the above method, while maintaining the relevance of the retrieval results, it can also ensure that the ranking of the retrieval results about important information is relatively high.

Step 702, when the user inputs a search character string, each layer is searched in order of priority.

In this step, when the user enters the search string, a hierarchical search process will be performed, that is, the highest priority search will be performed first, and then the second highest priority search will be performed, and so on. , until a sufficient number of user-desired search results are found.

In the technical solution of the present invention, the above step 702 may have multiple specific implementation manners. The technical solutions of the present invention will be introduced below by taking one of the specific implementation manners as an example.

FIG. 8 is a flowchart of an implementation method of step 702 in the present invention. As shown in FIG. 8, the above step 702 may include the following steps:

Step 801: Analyze the search character string input by the user according to the pre-set search process at each level, and generate etymology queues and query logic sequences corresponding to the search process at each level.

In the technical solution of the present invention, since the multi-layer search process is preset, the search methods used in each layer search process are not necessarily the same, so in this step, the search character string input by the user can be parsed, thereby Generate etymology queues and query logic sequences corresponding to the search process of each layer respectively.

Further, in a specific embodiment of the present invention, if there is only one etymology in the generated etymology queue, the query logic sequence corresponding to the etymology queue is an empty set.

For example, if the pre-set multi-layer search process is: the precise layer search process, the full spelling layer search process and the word segmentation layer search process, and the search character string input by the user is "Andy Lau God Will", then the search character input by the user After parsing the string, there is only one etymology "Andy Lau" in the etymology queue corresponding to the precise layer search process, and the query logic sequence corresponding to the etymology queue is an empty set; There is only one etymology "liudehuatianyi" in the corresponding etymology queue, and the query logic sequence corresponding to this etymology queue is an empty set; there are two etymologies in the resulting etymology queue corresponding to the word segmentation layer search process: " Andy Lau" and "God's will", there is a logic symbol in the query logic sequence corresponding to the etymology queue: OR.

Step 802: Analyze the search result range input by the user to obtain the starting position and the number of items.

In this step, the initial position InitialPositon and the number of items Count can be obtained by analyzing the search result range input by the user.

Step 803, according to the preset priority, determine the currently unexecuted search process with the highest priority as the current search process.

In this step, it is necessary to determine which layer the current search process is. Therefore, according to the preset priority, it can be determined that the current search process is the search process that is not currently executed and has the highest priority. For example, if the search process is executed for the first time, the current search process is the search process with the highest priority; if the search process is performed for the second time, the current search process is the search process with the second highest priority; and so on.

Step 804: Search the index posting list according to the etymology queue and query logic sequence corresponding to the current search process, store the search results in the hierarchical search result set and obtain the total number of current hierarchical search results.

Since the etymology queue and query logic sequence corresponding to each layer of search process have been generated in step 801, in this step, the index can be reversed directly according to the etymology queue and query logic sequence corresponding to the current search process table, and store the search results in the layered search result set (LayerResultList), so as to obtain the layered search result set; at the same time, the total number of search results in the current layered search result set can also be obtained, that is, the current layered search The total number of results (LayerResultCount).

In addition, in the current search process of this step, the above-mentioned multi-term search method can be used to search the index posting list, and the specific search process will not be repeated here.

Step 805: Calculate the hierarchical search score of each retrieval result in the hierarchical retrieval result set according to the KFScore of each retrieval result and the hierarchical search score range of the current search process.

In a specific embodiment of the present invention, various calculation methods may be used to calculate the hierarchical search score of each search result in the hierarchical search result set.

For example, according to the hierarchical search score range of the current search process and the sequence of each search result in the hierarchical search result set according to the KFScore from large to small, a corresponding hierarchical search score can be set for each retrieval result. Other calculation methods will not be repeated here.

Step 806: Add the search results in the hierarchical search result set to the total search result set according to the hierarchical search score, and delete duplicate search results; calculate the total number of current total search results, and clear the score A collection of layer retrieval results.

In this step, all the search results in the hierarchical search result set obtained in step 804 can be inserted into the search results in the total search result set (ResultList) according to the size of the hierarchical search score, and duplicate search results can be deleted. result. Wherein, the initial value of the total retrieval result set is an empty set, that is, no retrieval result is stored in the total retrieval result set under initial conditions. Therefore, if the current search process is the first-level search process, when the search results in the layered search result set are added to the total search result set, there will be no repeated search results, that is, the number of repeated search results is 0 .

In addition, in a specific embodiment of the present invention, it is also necessary to calculate the current total number of search results (SearchResultCount), that is, the total number of search results in the current total search result set. For example, the current total search result set can be directly counted to obtain the total number of search results; or, the total number of search results before this calculation can be added to the total number of current layered search results, and repeated search results can be subtracted to obtain the current total number of search results.

Step 807, judge whether the end condition is satisfied; if yes, execute step 808; otherwise, return to execute step 803.

Wherein, in a specific embodiment of the present invention, the end condition may be:

The current total number of search results is greater than the maximum value of the search result range, or the current search process is the last layer search process.

In the technical solution of the present invention, when the current total number of search results is greater than the search result range (InitialPositon+Count), it means that a sufficient number of search results have been retrieved, thus satisfying the condition for ending the search. And when the current search process is the last layer search process, the search can also be ended at this time, so that the end condition is met.

Step 808, according to the range of search results, read a required number of search results from the total search result set as search results; use the current total number of search results as the total number of relevant results (SumCount).

For example, in a specific embodiment of the present invention, reading a required number of search results from the total search result set according to the search result range as search results includes:

When SumCount≥(InitialPositon+Count)-1, start from the first InitialPositon record in the total retrieval result set, read the Count record as the search result;

When InitialPositon≤SumCount＜(InitialPositon+Count)-1, then show that the number of records in the total retrieval result collection is insufficient, at this moment, from the InitialPositon article record in the described total retrieval result collection, read (SumCount- InitialPositon+1) records as search results;

When SumCount<InitialPositon, it indicates that the number of records in the total search result set is too small, and there is no desired search result. At this time, the empty set can be used as the search result.

Step 809, return the search result and the total number of search results to the user.

Through the above steps 701-702, multi-layer search can be realized.

The following will take a specific example in the field of music search as an example to further introduce the above-mentioned multi-layer search method.

Specific example three:

For the sake of simplicity, in this specific example, the basic settings in the first specific example are still used, and the index posting list shown in FIG. 5 is established, and the length n of the ordered list in the index posting list is set to 100.

When the search string entered by the user is "Andy Lau's providence" and the range of search results entered by the user is: the 1st to 50th search results, the following multi-layer search method can be used to search:

In this specific example, three-level search processes are preset according to the business characteristics in the music search field: precise level search process, full-level search process and word-segmentation level search process. Moreover, the priorities of the search process at each level are in descending order: the search process at the precise level, the search process at the full spelling level, and the search process at the word segmentation level. Then, the above-mentioned search processes of each layer are all divided into three sub-layer search processes according to the order of priority from high to low: song sub-layer search process, singer sub-layer search process and album sub-layer search process.

According to the above setting situation, pre-set the hierarchical search score range of each layer search process: set the hierarchical search score range of the precise layer search process to [A1, A2]; The score range is set to [B1, B2], and the hierarchical search score range of the word segmentation layer search process is set to [C1, C2]; wherein, A1>A2>B1>B2>C1>C2.

After the above settings are completed, the search string "Andy Lau God Will" input by the user can be analyzed according to the pre-set search process of each layer, and the etymology queue and query logic sequence corresponding to the search process of each layer are respectively generated, respectively:

There is only one etymology "Andy Lau God Will" in the etymology queue corresponding to the precise layer search process, and the query logic sequence corresponding to the etymology queue is an empty set;

There is only one etymology "liudehuatianyi" in the etymology queue corresponding to the full-level search process, and the query logic sequence corresponding to the etymology queue is an empty set;

There are two etymologies in the etymology queue corresponding to the word segmentation layer search process: "Andy Lau" and "God's will", and there is a logic symbol in the query logic sequence corresponding to this etymology queue: AND.

Analyze the range of search results entered by the user, and learn that the starting position InitialPositon=1, and the number of entries Count=50;

Since the song sub-layer search process has the highest priority in the precise layer search process, the song sub-layer search process can be determined as the current search process.

In the song sub-layer search process, according to the etymology "Andy Lau's providence", the index posting list shown in Figure 5 is retrieved using a single-term search method. Since the corresponding etymology "Andy Lau's providence" cannot be found from the index posting table, an empty search result is returned; therefore, the total search result set is an empty set, and the current total search result is 0.

Since the current total number of search results is less than the range of search results, and the current search process is not the last search process, the end condition is not satisfied, and the subsequent search process will continue. At this time, the currently unexecuted search process with the highest priority is the singer sub-layer search process, so the singer sub-layer search process can be determined as the current search process for subsequent retrieval.

In the same way, in the singer sub-level search process and the album sub-level search process, the corresponding etymology "Andy Lau's will" cannot be found from the index inverted list, so an empty search result is returned, so it is still not available. If the end condition is met, the subsequent full layer search process will continue.

Since the etymology in the full-level search process is "liudehuatianyi", the corresponding etymology cannot be found from the index inverted table when performing the search process of each sub-layer in the full-level search process "liudehuatianyi", therefore returns an empty search result, so the end condition is still not met, and the subsequent word segmentation layer search process will continue.

In the word segmentation layer search process, the song sub-layer search process will be carried out first, then the artist sub-layer search process will be carried out, and the album sub-layer search process will be carried out at last.

Since there are two etymologies in the word segmentation layer search process: "Andy Lau" and "God's Will", and there is a logical symbol in the query logic sequence corresponding to the etymology queue: AND, therefore, in each sublayer of the word segmentation layer search process During the search process, the corresponding etymology can be found from the index inverted list, and corresponding search results can be obtained.

When the number of search results obtained in the word segmentation layer search process is greater than 50, the search process will be terminated, and the search results of the required number of items will be read from the total search result collection according to the search result range as the search results, and calculated The total number of search results, and then return the search results and the total number of search results to the user.

And when the number of search results when the word segmentation layer search process ends is still less than 50, because the word segmentation layer search process has been the last layer of search process, it also satisfies the end condition. At this time, the total number of search results can be calculated, and then to the user Returns the search results and the total number of search results.

In the technical solution of the present invention, a search device is also proposed. Fig. 9 is a schematic structural diagram of the search device in the present invention. As shown in FIG. 9 , the search device in the present invention includes: a posting list generation module 901 , a storage module 902 , an etymology generation module 903 , and a retrieval module 904 .

The posting table generation module 901 is used to index all documents to be retrieved with the etymology Term as the keyword, and establish the Term as the keyword index, the total TotalCount of the documents containing the Term and the list of documents containing the Term DocList is an index posting list of values; each record in the document list includes a document number and a key attribute score; the document list is composed of an ordered list and an unordered list, and the ordered The list includes n key attribute scores that are the largest and are arranged in descending order according to the key attribute scores; wherein, the n is a predetermined value; the index inverted list is sent to the storage module 902;

The storage module 902 is configured to store the index posting list;

The etymology generating module 903 is configured to generate a corresponding etymology according to the search character string input by the user; send the etymology and the search result range input by the user to the retrieval module 904;

The retrieval module 904 is used for retrieving the index posting list stored in the storage module 902 according to the generated etymology, and preferentially obtains from the ordered list corresponding to the generated etymology according to the range of search results input by the user record to obtain the desired total of search results and related results; and send the search results and the total of related results to the user.

To sum up, in the technical solution of the present invention, since the key attributes and key attribute weights are set for the same type of documents, the key attribute scores of each document can be calculated, and then the index posting list can be established according to the above key attribute scores , so that the document list in the index posting list is composed of an ordered list and an unordered list, and the ordered list includes n records with the largest key attribute scores and arranged in descending order of key attribute scores. Since the key attribute score of the document has been completed when the index posting list is established, and part of the sorting operation has been completed in advance for the records stored in the index posting list, it is possible to sort all the records according to the search string entered by the user. When retrieving the above index inverted list, the range of search results entered by the user is included in the ordered list in most cases, so that the required number of search results can be directly obtained from the ordered list, ensuring Most of the search results can be obtained directly from the ordered list, and the obtained search results have been arranged in the order of KFScore size, so there is no need to sort the search results, which greatly reduces the calculation of the full data score and the full sort operation. The number of operations of the CPU is greatly reduced, the search speed is significantly improved, and the search speed is increased by an order of magnitude compared with the search speed in the prior art, and the search response time of the search engine is also greatly improved.

In addition, in the index posting table of the present invention, what is stored is the key attribute score of each document, and there is no need to store the factors required for calculating the score, so the stored fields are reduced, and the occupied memory resources are greatly saved, thereby Save a lot of hardware and software resources.

In addition, in the index inverted list of the present invention, the basis for sorting the records in the ordered list is the key attribute and key attribute score of each document, and the key attribute and key attribute score of each document can be based on the actual The application situation is set in advance, so the relevance of text and the relative value of information can be fully considered, making the ranking of search results closer to user needs; at the same time, key attributes and key attribute scores can be flexibly set, so that text relevance can be ensured Based on the sorting operation of custom search results.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the present invention. within the scope of protection.

Claims (25)

1. a searching method is characterized in that, this method comprises:

A, for same at least one determinant attribute of class document setup and corresponding determinant attribute weight, and according to the determinant attribute score value KFScore of described determinant attribute and each document of determinant attribute weight calculation;

B, be that key word carries out index with whole documents to be retrieved with etymology Term, to set up with Term be key word index, be the index inverted list of value with the total TotalCount of the document that comprises this Term and the lists of documents DocList that comprises this Term; Include document code and the determinant attribute score value of a document in every record in the described lists of documents; Described lists of documents is made up of ordered list and unordered list, comprises n determinant attribute score value maximum and the record of arranging by determinant attribute score value descending order in the described ordered list; Wherein, described n is predetermined value;

C, the searching character string of importing according to the user generate corresponding etymology, according to the etymology that generates described index inverted list is retrieved, and preferentially from the ordered list of the etymology correspondence that generates, obtain record according to the Search Results scope of user input, to obtain required Search Results and correlated results sum.

2. method according to claim 1 is characterized in that, the formula that calculates the KFScore of document is:

KFScore＝KeyField ₁*W ₁+KeyField ₂*W ₂+......+KeyField _X*W _X；

Wherein, W ₁+ W ₂+ ...+W _X=1; Described KeyField _xThe value of representing x KeyField of described document, described W _xThe value of representing the Weight corresponding with x KeyField of described document.

3. method according to claim 1 is characterized in that,

Described Search Results scope comprises: reference position InitialPositon and bar are counted Count;

The maximal value of described Search Results scope is: (InitialPositon+Count)-1.

4. method according to claim 3 is characterized in that, described Search Results scope according to user input is preferentially obtained record from the ordered list of the etymology correspondence that generates, comprise to obtain required Search Results:

When the maximal value of described Search Results scope was less than or equal to the length n of ordered list of the etymology correspondence that generates, the InitialPositon bar start-of-record from described ordered list read Count bar record as Search Results;

Be arranged in the ordered list of the etymology correspondence that generates when the reference position of described Search Results scope, and the maximal value of described Search Results scope is during greater than the length n of described ordered list, InitialPositon bar start-of-record from described ordered list reads (n-InitialPositon+1) bar record; Record in the unordered list of the etymology correspondence that generates is arranged by the size order of KFScore, and the 1st start-of-record from described unordered list, read [Count-(n-InitialPositon+1)] bar record; The Count bar that reads is recorded as Search Results;

When the reference position of described Search Results scope is arranged in the unordered list of the etymology correspondence that generates, record in the described unordered list being pressed the size order of KFScore arranges, (InitialPositon-n) bar start-of-record from the unordered list after the ordering reads Count bar record as Search Results then.

5. method according to claim 3 is characterized in that, also further comprises among the step C:

When etymology that the searching character string of not storing in the described index inverted list according to user input generates, with empty set as Search Results.

6. method according to claim 1 is characterized in that, described step C comprises:

C1, the searching character string of importing according to the user generate a corresponding etymology;

C2, according to the etymology search index inverted list that generates, judge in the described index inverted list whether store this etymology; If, execution in step c3 then; Otherwise, execution in step c15;

C3, from described index inverted list, read the lists of documents corresponding with described etymology;

C4, the Search Results scope of user input is resolved, obtain reference position InitialPositon and bar and count Count;

Whether c5, judgement (InitialPositon+Count)-1＞n set up, if, execution in step c6 then; Otherwise, execution in step c12; Wherein, described n is the length of ordered list;

C6, judge whether InitialPositon＞n sets up, if, execution in step c10 then; Otherwise, execution in step c7;

C7, the InitialPositon bar start-of-record from ordered list read (n-InitialPositon+1) bar record and join among the search result list ResultList;

C8, the size order that KFScore pressed in the record in the described unordered list are arranged;

C9, from described unordered list, read [Count-(n-InitialPositon+1)] bar record and join among the ResultList execution in step c13;

C10, the size order that KFScore pressed in the record in the described unordered list are arranged;

C11, (InitialPositon-n) bar start-of-record from the unordered list after the ordering read Count bar record and join among the ResultList execution in step c13;

C12, the InitialPositon bar start-of-record from ordered list read Count bar record and join among the ResultList execution in step c13;

C13, with described ResultList as Search Results, and the total number of documents TotalCount of the lists of documents corresponding with described etymology that will read from described index inverted list is as the correlated results sum;

C14, described Search Results and correlated results sum are returned to the user.Process ends;

C15, with empty set as Search Results, establish the correlated results sum and equal 0, described empty set and correlated results sum are returned to the user, process ends.

7. method according to claim 1 is characterized in that, described step C comprises:

C1, Y section read range is set in the lists of documents of index inverted list; Wherein, Y 〉=2;

C2, the searching character string of importing according to the user generate an etymology formation TermArray and query logic sequence SetOperators;

C3, the Search Results scope of user input is resolved, obtain reference position InitialPositon and bar and count Count;

C4, when the length of TermArray greater than 1 the time, read the first two words source among the TermArray, and initial value is set is that 2 variable i and initial value are 1 variable j;

C5, according to two etymologies that read search index inverted lists respectively, from described index inverted list, read the record in the j section read range in the lists of documents corresponding with described two etymologies respectively, the record that reads is stored in respectively among first result for retrieval tabulation ResultList1 and second result for retrieval tabulation ResultList2;

C6, read (i-1) the individual logical symbol among the SetOperators;

The value of the logical symbol that C7, judgement are read; When the value of described logical symbol is AND, execution in step C8; When the value of described logical symbol is OR, execution in step C9; When the value of described logical symbol is SUB, execution in step C10;

C8, use and logic merge ResultList1 and ResultList2, and amalgamation result joins among the search result list ResultList; Execution in step C11;

C9, use or logic merge ResultList1 and ResultList2, and amalgamation result joins among the ResultList; Execution in step C11;

C10, use difference logic merge ResultList1 and ResultList2, and amalgamation result joins among the ResultList; Execution in step C11;

C11, judge among the TermArray whether to also have etymology not use; If, execution in step C12 then; Otherwise, execution in step C13;

C12, establish i=i+1, read i etymology among the TermArray; Empty ResultList1 and ResultList2, and the record among the ResultList is copied among the ResultList2; Retrieve described index inverted list according to the etymology that reads, from described index inverted list, read the record in the j section read range in the lists of documents corresponding with the etymology that reads, the record that reads is stored among the ResultList1; Return execution in step C6;

C13, judge whether to satisfy termination condition; If, execution in step C16 then; Otherwise, execution in step C14;

C14, empty ResultList1 and ResultList2; Read the first two words source among the TermArray, and i=2 and j=j+1 are set; According to above-mentioned two etymologies difference search index inverted lists that read, from described index inverted list, read the record in the j section read range in the lists of documents corresponding with described two etymologies respectively, be stored in respectively among ResultList1 and the ResultList2;

C15, the size order of the record among described ResultList1 and the ResultList2 being pressed KFScore are respectively arranged; Return execution in step C6;

C16, with the number of the record stored in the set of total result for retrieval as the correlated results sum;

C17, return Search Results and correlated results sum to the user.

8. method according to claim 7 is characterized in that,

When Y=2, with the ordered list of lists of documents as the 1st section read range, with the entire document tabulation as the 2nd section read range.

9. method according to claim 7 is characterized in that,

When Y=3, with the ordered list of lists of documents as the 1st section read range, with the unordered list of lists of documents as the 2nd section read range, with the entire document tabulation as the 3rd section read range.

10. method according to claim 7 is characterized in that, also further comprises among the step C5:

If do not store any one etymology in described two etymologies in the index inverted list, the corresponding result for retrieval tabulation of the etymology of then not storing is set to empty set.

11. method according to claim 7 is characterized in that, described use and logic merge ResultList1 and ResultList2, and amalgamation result joined among the ResultList comprise:

Compare each the bar record among ResultList1 and the ResultList2 one by one, KFScore two records identical with Did in two result for retrieval tabulations are joined among the ResultList as a Search Results.

12. method according to claim 7 is characterized in that, described use or logic merge ResultList1 and ResultList2, and amalgamation result is joined among the ResultList and can comprise:

Each bar record among described ResultList1 and the ResultList2 is joined among the ResultList, if there is the KFScore of two records all identical with Did, then in ResultList, delete a record wherein.

13. method according to claim 7 is characterized in that, described use difference logic merges ResultList1 and ResultList2, and amalgamation result is joined among the ResultList and can comprise:

Remove each bar record of storing among the ResultList2 from described ResultList1, the record that will remove among the ResultList1 after the operation joins among the ResultList.

14. method according to claim 7 is characterized in that, also further comprises among the described step C12:

If do not store above-mentioned i the etymology that reads in the index inverted list, then described ResultList1 is set to empty set, returns execution in step C6 again.

15. method according to claim 7 is characterized in that, described termination condition is:

The maximal value of the number great-than search range of results of the record of storing among the ResultList; Perhaps, j equals Y.

16. method according to claim 7 is characterized in that, describedly reads required record comprise as Search Results from described ResultList:

When SumCount 〉=(InitialPositon+Count)-1, the InitialPositon bar start-of-record from described ResultList reads Count bar record as Search Results;

When InitialPositon≤SumCount＜(InitialPositon+Count)-1, the InitialPositon bar start-of-record from described ResultList reads (SumCount-InitialPositon+1) bar record as Search Results;

When SumCount＜InitialPositon, with empty set as Search Results.

17. method according to claim 1 is characterized in that, described step C comprises:

Set in advance the multilayer search procedure according to predefined searching order rule, and the priority of each layer search procedure and the hierarchical search score value scope of each layer search procedure are set;

When the user imports searching character string, carry out each layer search in proper order according to the height of priority.

18. method according to claim 17 is characterized in that, describedly sets in advance the multilayer search procedure according to predefined searching order rule, and the priority that each layer search procedure be set comprises:

In the music searching field, three layers of following search procedure are set: accurate layer search procedure, spelling layer search procedure and participle layer search procedure; And the priority of each layer search procedure by order from high to low is: accurate layer search procedure, spelling layer search procedure, participle layer search procedure;

19. method according to claim 18 is characterized in that, this method also further comprises:

Every layer of search procedure is divided into a plurality of sublayers search procedure, and sets the priority of each sublayer search procedure.

20. method according to claim 17 is characterized in that, the described hierarchical search score value scope that each layer search procedure is set comprises:

When the setting search procedure that haves three layers, and the priority of each layer search procedure by from high to low order is: when ground floor search procedure, second layer search procedure, the 3rd layer of search procedure,

The hierarchical search score value scope of ground floor search procedure is made as: [A1, A2]; The hierarchical search score value scope of second layer search procedure is made as: [B1, B2]; The hierarchical search score value scope of the 3rd layer of search procedure is made as: [C1, C2]; Wherein, A1＞A2＞B1＞B2＞C1＞C2.

21. method according to claim 17 is characterized in that, and is described when the user imports searching character string, carries out each layer search in proper order according to the height of priority and comprises:

Each layer search procedure that Z1, basis set in advance resolved the searching character string of user's input, generates etymology formation and the query logic sequence corresponding with each layer search procedure respectively

Z2, the Search Results scope of user input is resolved, obtain reference position and bar number;

Z3, according to priority preset, current search procedure unenforced and that priority is the highest is defined as the current search process;

Z4, the index inverted list is retrieved according to etymology formation and the query logic sequence corresponding with the current search process, be stored in result for retrieval in the set of layering result for retrieval and obtain current layering result for retrieval sum

Z5, according to the KFScore of each result for retrieval and the hierarchical search score value scope of current search process, calculate the hierarchical search score value of each result for retrieval in the set of layering result for retrieval.

Z6, the size of the result for retrieval in the layering result for retrieval set according to the hierarchical search score value joined in total result for retrieval set, and the result for retrieval that repeats of deletion; Calculate current total result for retrieval sum, and empty described layering result for retrieval set;

Z7, judge whether to satisfy termination condition; If, execution in step Z8 then; Otherwise, return execution in step Z3;

Z8, from described total result for retrieval set, read the result for retrieval of required number as Search Results according to the Search Results scope; With current total result for retrieval sum as correlated results sum SumCount;

Z9, return Search Results and Search Results sum to the user.

22. method according to claim 21 is characterized in that, and is described according to the KFScore of each result for retrieval and the hierarchical search score value scope of current search process, the hierarchical search score value that calculates each result for retrieval in the set of layering result for retrieval comprises:

According to each result for retrieval in the hierarchical search score value scope of current search process and the set of layering result for retrieval according to KFScore putting in order from big to small, for each result for retrieval arranges corresponding hierarchical search score value.

23. method according to claim 21 is characterized in that, described termination condition can for:

The maximal value of current total result for retrieval sum great-than search range of results; Perhaps, the current search process is last one deck search procedure.

24. method according to claim 21 is characterized in that, the described result for retrieval that reads required number according to the Search Results scope from described total result for retrieval set comprises as Search Results:

When SumCount 〉=(InitialPositon+Count)-1, the InitialPositon bar start-of-record from described total result for retrieval set reads Count bar record as Search Results;

When InitialPositon≤SumCount＜(InitialPositon+Count)-1, the InitialPositon bar start-of-record from described total result for retrieval set reads (SumCount-InitialPositon+1) bar record as Search Results;

When SumCount＜InitialPositon, with empty set as Search Results.

25. a searcher is characterized in that, this searcher comprises: inverted list generation module, memory module, etymology generation module and retrieval module;

Described inverted list generation module, being used for whole documents to be retrieved is that key word carries out index with etymology Term, and foundation is key word index with Term, be the index inverted list of value with the total TotalCount of the document that comprises this Term and the lists of documents DocList that comprises this Term; Include document code and the determinant attribute score value of a document in every record in the described lists of documents; Described lists of documents is made up of ordered list and unordered list, comprises n determinant attribute score value maximum and the record of arranging by determinant attribute score value descending order in the described ordered list; Wherein, described n is predetermined value; Described index inverted list is sent to memory module;

Described memory module is used for the described index inverted list of storage;

Described etymology generation module is used for generating corresponding etymology according to the searching character string of user's input; The Search Results scope of described etymology and user's input is sent to retrieval module;

Described retrieval module, be used for retrieving according to the index inverted list that the etymology that generates is stored memory module, and preferentially from the ordered list of the etymology correspondence that generates, obtain record according to the Search Results scope of user input, to obtain required Search Results and correlated results sum; Described Search Results and correlated results sum are sent to the user.

Images