CN101820592A - Method and device for mobile search - Google Patents

Abstract

The invention discloses a mobile search method and device. The method includes: receiving a search request, the search request includes one or more query keywords; calculating the score value of each search type field, and the score value is as follows The score value of any item or the comprehensive score value of multiple items: the similarity between the search request and the search type domain, the search rate of the search request corresponding to the search type domain, the personalized user of the search type domain Interest scoring value; the popular search rate: the number of popular searches, or the number of clicks on popular search results; according to the scoring values of each search type field, select one or several search type fields to search for the query keyword. With the present invention, personalized and accurate search results can be provided for users.

Description

Mobile search method and device

technical field

The invention relates to mobile communication technology, in particular to a mobile search method and device.

Background technique

At present, mobile search, which is the combination of search engine and mobile communication, two hot fields in the current information industry, has become a new highlight and growth point of mobile value-added services. The mobile search framework is an open platform based on meta-search, which integrates the capabilities of many professional/vertical search engines to provide users with a comprehensive search capability.

When a user uses a mobile search, he usually searches directly after inputting a search keyword without selecting a search type domain (domain). Therefore, how to correctly understand the user's search intention and provide the user with personalized and accurate search results, there is no good solution in the prior art.

Contents of the invention

Embodiments of the present invention provide a mobile search method and device, which can provide users with personalized and accurate search results.

An embodiment of the present invention provides a mobile search method, including:

Receiving a search request, the search request includes one or more query keywords;

Calculate the score value of each search type field, and the score value is the score value of any one of the following items or the comprehensive score value of multiple items: the similarity between the search request and the search type field, the search request corresponding to the The popular search rate of the search type domain, the personalized user interest score value of the search type domain;

According to the scoring value of each search type field, one or several search type fields are selected to search for the query keyword.

An embodiment of the present invention provides a mobile search device, including:

a receiving unit, configured to receive a search request, wherein the search request includes one or more query keywords;

A calculation unit, configured to calculate the score value of each search type field, and the score value is a score value of any one of the following items or a comprehensive score value of multiple items: the similarity between the search request and the search type field, the The search request corresponds to the popular search rate of the search type domain and the personalized user interest score value of the search type domain;

Selection unit, select one or several search type fields according to the scoring value of each search type field;

A search unit, configured to search for the query keyword using the search type field selected by the selection unit.

The mobile search method and device provided by the embodiments of the present invention determine the user's personalized query category by analyzing the user's public interest and the user's personalized interest, thereby providing the user with personalized and accurate search results.

Description of drawings

Fig. 1 is the flow chart of the mobile search method of the embodiment of the present invention;

Fig. 2 is a kind of implementation flowchart of the mobile search method of the embodiment of the present invention;

Fig. 3 is another implementation flowchart of the mobile search method in the embodiment of the present invention;

Fig. 4 is another implementation flowchart of the mobile search method in the embodiment of the present invention;

Fig. 5 is another implementation flowchart of the mobile search method according to the embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a mobile search device according to an embodiment of the present invention;

FIG. 7 is a schematic structural diagram of a mobile search device according to an embodiment of the present invention;

Fig. 8 is a schematic diagram of another specific structure of a mobile search device according to an embodiment of the present invention;

FIG. 9 is a schematic diagram of another specific structure of a mobile search device according to an embodiment of the present invention;

Fig. 10 is a schematic structural diagram of an interest model extraction subunit in the device shown in Fig. 9;

Fig. 11 is another schematic structural diagram of the interest model extraction subunit in the device shown in Fig. 9;

Fig. 12 is a schematic diagram of another specific structure of a mobile search device according to an embodiment of the present invention.

Detailed ways

In order to enable those skilled in the art to better understand the solutions of the embodiments of the present invention, the embodiments of the present invention will be further described in detail below in conjunction with the drawings and implementations.

The mobile search method and device of the embodiments of the present invention determine the user's personalized query category by analyzing the user's corresponding public interest and user's personalized interest for the user's search request, specifically, calculate the scoring value of each search type field, The score value is the score value of any one of the following items or the comprehensive score value of multiple items: the similarity between the search request and the search type domain, the public search rate of the search request corresponding to the search type domain, search The personalized user interest score value of the type domain; the mass search rate is: the number of mass searches, or the number of clicks on the mass search results; then, according to the scoring value of each search type field, select one or several search type fields to search the Query keywords to provide users with personalized and accurate search results.

As shown in FIG. 1 , it is a flowchart of a mobile search method according to an embodiment of the present invention.

Step 101, receiving a search request, where the search request includes one or more query keywords.

Step 102, calculate the score value of each search type field, the score value is the score value of any one of the following items or the comprehensive score value of multiple items: the similarity between the search request and the search type field, the search request Corresponding to the popular search rate of the search type field and the personalized user interest score value of the search type field; the popular search rate: the number of times of popular searches, or the number of clicks of popular search results.

Step 103, select one or more of the search type fields to search for the query keyword according to the scoring values of each search type field.

In the embodiment of the present invention, when determining the user's personalized query category, there may be multiple implementation methods, for example, it may be based on the similarity between the search request and the search type field, select the one with a high similarity or Several search type domains are searched; it is also possible to select one or several search type domains with high public search rate according to the popular search rate of the search type domain corresponding to the search request; it is also possible to search according to the search type domain Personalized user interest score value, select one or several search type domains with high personalized user interest score value to search. Of course, it is also possible to comprehensively consider the above items, calculate the comprehensive score value of each search type field, and select one or several search type fields with high comprehensive score values for searching. Examples are given below to explain this in detail.

Referring to FIG. 2 , it is a flowchart of an implementation of a mobile search method according to an embodiment of the present invention.

In this embodiment, according to the similarity between the search request and the search type field, a search type field is selected for searching, so as to provide users with personalized and accurate search results.

Step 201, receiving a search request, where the search request includes one or more query keywords.

Step 202, calculate the similarity between the search request and each search type field according to the query keyword.

Corresponding weights may be set for the query keywords in the search request, and the query vector Query(q1, q2,...qn') is generated by the weights of the query keywords; where, q1, q2,...qn ' is the weight corresponding to each query keyword; specifically, the same weight can be set for all keywords, such as weight = 1; different weights can also be set for different keywords, for example, for the top keywords Set the maximum weight, such as weight=1, set the weight of the middle size for the keywords ranked in the middle, such as 0.5<weight<1, set the minimum weight for the keywords ranked last, such as weight=0.5.

The domain vector corresponding to the search type domain is generated from the weights of each word in the search type domain, such as setting a certain weight for all subject words and related words in each search type domain, which is composed of the weights of these subject terms and related words A domain vector Domain(t1, t2, ..., tn) corresponding to the search type domain, where t1, t2, ..., tn is the weight of each word in the search type domain. The similarity between the search request and the search type domain is obtained by calculating the query vector and the domain vector.

The similarity between the vector Domian(t1, t2, ..., tn) and the vector Query(q1, q2, ..., qn') can be calculated according to the following formula:

$\mathrm{<span\; class="notranslate">\; Sim</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Query</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; q</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; q</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; qn</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; Domain</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; tn</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

$<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; q</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; q</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; qn</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; *</span>{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; in</span>}}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\sqrt{{\mathrm{<span\; class="notranslate">\; q</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; q</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; qn</span>}}^{<span\; class="notranslate">\; \&prime;</span>\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; *</span>\sqrt{{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>{{\mathrm{<span\; class="notranslate">\; t</span>}}_{\mathrm{<span\; class="notranslate">\; in</span>}}}^{<span\; class="notranslate">\; \&prime;</span>\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Among them, t _i1 , t _i2 , ..., t _in ' are the words corresponding to the same query keywords corresponding to weights q1, q2, ..., qn' in the vector Domian(t1, t2, ..., tn) the weight of.

Suppose there are m search type domains, and the corresponding domain vectors are Domain1(t1, t2, ..., tn), Domain2(t1, t2, ..., tn), ..., Domainm(t1, t2, ..., tn), then calculate the similarity between the vector Query(q1, q2, ..., qn') and the above domain vectors according to the formula (1).

Step 203, select one or more search type domains with high similarity for searching.

In this embodiment, the subject words, related words, and weights of each word in each search type field can be set in various ways.

1. Manual allocation method

Set the maximum weight for the subject words, set the middle size weight for the strong related words, and set the minimum weight for the weak related words.

For example: subject words (such as "Sichuan Cuisine" in the catering search type field) set the weight to 1, strong related words (such as "spicy" in the catering search type field) set the weight to 0.8, weakly related words (such as catering search type field " "Incense" in ) set the weight to 0.5.

2. By learning the automatic allocation method

The specific process is as follows:

(1) For each search type field, obtain a training text corpus sample corresponding to the search type field;

(2) Carry out word segmentation to described corpus sample, generate the thesaurus of this search type field;

(3) Calculate the weight of each word in the thesaurus, the weight of each word=TF*GIDF, wherein TF is the total word frequency of this word in all corpus samples of this search type domain, GIDF is the global reverse document frequency, GIDF =log(1+N/GDF), where N is the total number of all corpus samples in all search type domains, and GDF is the global corpus sample frequency, which is the number of all corpus samples that include the word in all search type domains;

(4) determine subject words and related words in the search type field according to the weight of each word;

Assume that there are n words in the thesaurus of a certain search type domain, and the corresponding weights are T1, T2, ..., Tn, among them, T1>T2>...>Tn, so that the word corresponding to T1 can be considered as the topic words, and other words are related words.

Further, it is also possible to divide all the words in the thesaurus into sets of different grades according to their weights, set the final score value for the set of each grade, and use the final score value of each grade as each word in the grade the weight of. For example, there are L grades in total, the highest score value is set for the first grade, the intermediate grade value is set for the middle grade, and the minimum score value is set for the L grade. In this way, the domain vector of the corresponding search type domain can be formed from the words in the part of speech and their final scoring values.

Of course, the embodiments of the present invention are not limited to the above-mentioned setting methods, and other methods can be used to set the subject words, related words, and weights of each word in each search type field, which will not be described in detail here.

In the mobile search method of the embodiment of the present invention, for the user's search request, by calculating the similarity between the query vector of the search request and the field vectors of each search type field, one or several search type fields with high similarity are selected for searching, so that Determine personalized query categories for users, and provide users with personalized and accurate search results.

Referring to FIG. 3 , it is a flow chart of another implementation of the mobile search method according to the embodiment of the present invention.

In this embodiment, according to the popular search rate corresponding to the search type field in the search request, the search type field is selected for searching, so as to provide users with personalized and accurate search results.

Step 301: Receive a search request, where the search request includes one or more query keywords.

Step 302, calculating the popular search rate of each search type domain corresponding to the search request according to the query keyword.

Step 303, select one or more search type domains with high public search rate to search.

In the embodiment of the present invention, the popular search rate may specifically be: the number of popular searches, or the number of clicks on popular search results.

The process of calculating the number of popular searches and the number of clicks on popular search results corresponding to each search type field of the search request will be described in detail below.

The process of calculating the number of public searches for a certain search type domain corresponding to the search request is as follows:

(1) Calculate the total number of public searches of a certain search type field corresponding to each keyword in the search request;

According to historical records, the sum of the number of times all users choose to search with a certain search type field for a search request containing a certain keyword in the search request can be used as the number of times the public corresponding to the keyword searches for this search type field The total number of times, that is, the total number of public searches corresponding to the search type domain;

(2) The sum of the total number of popular searches in the search type field corresponding to all the keywords in the search request is taken as the total number of popular searches in the search type field corresponding to the search request.

Similarly, the process of calculating the number of clicks on the public search results of a certain search type domain corresponding to the search request is as follows:

(1) calculating the total number of clicks on popular search results of a certain search type field corresponding to each keyword in the search request;

According to historical records, the sum of the number of times all users click on the search results of a search request containing a certain keyword in the search request and choose a certain search type domain is collected as the search type corresponding to the keyword. The total number of clicks on the search results of the domain, that is, the total number of clicks on the public search results corresponding to the search type domain;

(2) The sum of the total number of clicks on the popular search results of the search type field corresponding to all the keywords in the search request is used as the total number of clicks on the popular search results of the search type field corresponding to the search request.

In the mobile search method of the embodiment of the present invention, for a user's search request, by calculating the popular search rate of each search type domain corresponding to the search request, one or several search type domains with a high public search rate are selected for searching, thereby providing users with Determine the classification of personalized queries to provide users with personalized and accurate search results.

Referring to FIG. 4 , it is another implementation flowchart of the mobile search method in the embodiment of the present invention.

In this embodiment, according to the personalized user interest score value of the search type field, a search type field with a high score value is selected for searching, so as to provide users with personalized and accurate search results.

Step 401, receiving a search request, where the search request includes one or more query keywords.

Step 402, extract the user's interest model from the user data.

The user's interest model is a vector composed of rating values of the user data for multiple interest dimensions, such as IM(I1, I2, ..., In), where Ii is the rating value of the i-th interest dimension of the user. User interest models can be extracted from user personalized data (such as static files, search and click history data, presented business information, local information, etc.); the corresponding user interest models can also be extracted from user personalized data in advance and saved. When needed, the required user interest models are directly extracted from these saved user interest models.

The user's interest model may be a static interest model or a dynamic interest model, of course, it may also be an interest model generated by combining the static interest model and the dynamic interest model.

The user's static interest model can be extracted from the user's static profile. The specific process can be in the following two ways:

(1) Calculate the sum of the word frequencies of all words belonging to each interest dimension in the user's static file, and use it as the score value corresponding to each interest dimension, and generate the user by using the score value corresponding to each interest dimension as a vector interest model;

(2) Calculating the static file of the user and the similarity scoring value of each dimension of interest, and using it as the scoring value corresponding to each dimension of interest, generating the user interest model as a vector by the scoring value corresponding to each dimension of interest;

Extract the user's dynamic interest model from the user data. The specific process can be in the following two ways:

(1) Calculate the sum of the word frequencies of all words belonging to each dimension of interest in the user's search click history, and use it as the score value corresponding to each dimension of interest, which is generated by the score value corresponding to each dimension of interest as a vector Describe the user's dynamic interest model;

(2) Calculate the similarity score value between the search click history record and each interest dimension, and use it as the score value corresponding to each interest dimension, and generate the dynamic interest of the user by using the score value corresponding to each interest dimension as a vector Model.

The interest model generated by combining static interest model and dynamic interest model can be:

(1) first normalize the static interest model and the dynamic interest model respectively, and then calculate the sum of one or more static interest models and one or more dynamic interest models after normalization, And use this sum as the user's interest model.

(2) First, one or more of the static interest models and one or more of the dynamic interest models are weighted and added, and then the sum of the weighted additions is normalized, and the normalized The final result is used as the user's interest model.

Step 403: The sum of the score values of the search type field corresponding to one or more interest dimensions of the user interest model is used as the personalized user interest score value of the search type field.

Step 404: Select one or more search type domains with high scoring values to search for the query keyword.

For example, the user's interests are represented by n dimensions, such as: news, sports, entertainment, finance, technology, real estate, games, women, forums, weather, commodities, home appliances, music, reading, blogs, mobile phones, military, education , tourism, MMS, CRBT, catering, civil aviation, industry, agriculture, computer, geography, etc. The user interest model is a vector W(r1, r2, r3, .

When extracting the user interest model from the user's personalized data, it can be extracted from the user's static profile, or can be extracted from the historical data of the user's search.

There are several ways to extract the user interest model W1 from the user's static profile:

(1) W1=(p1, p2, p3, . . . , pn), where pi is the sum of the word frequencies of all words whose type belongs to the i-th dimension of interest in the static file.

(2) W1=(p1, p2, p3, . . . , pn), where pi is the similarity score between the static profile and the i-th dimension of interest.

Among them, the process of calculating the similarity pi between a static file and a dimension of interest is as follows:

(a) Extract the feature vocabulary of the classifier, specifically:

(i) Collect corresponding corpus for each interest dimension of the user to generate a corpus;

(ii) performing word segmentation on the corpus to form a series of entries;

(iii) To determine whether the entry after word segmentation is a feature word, specifically, the chi-square statistical algorithm (CHI) can be used:

${\mathrm{<span\; class="notranslate">\; \&chi;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; AD</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; BC</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ;</span>$

Among them, the meaning of each parameter is as follows: t: a certain entry; c: a certain category; N: the total number of training texts; A: the number of training texts that belong to c and contain t; B: the number of texts that do not belong to c but contain t ;C: the number of texts belonging to c but not including t; D: the number of texts not belonging to c nor including t. If both C and D are 0, then χ ² (t, c)=0;

或

低于指定阈值的词条可不考虑作为特征词。The CHI value of term t to the entire training set can be defined as:

or

Entries below the specified threshold may not be considered as feature words.

The calculation process of P(c) is as follows:

Let the categories be C ₁ , C ₂ , ..., C _n ,

其中，N(C_i)是类别C_i所包含的训练文本的数量；but

Among them, N(C _i ) is the number of training texts contained in category C _i ;

其中，M(C_i)是类别C_i的所有训练文本所包含的词条总数，M是所有训练文本所包含的词条总数。or,

Among them, M(C _i ) is the total number of entries contained in all training texts of category C _i , and M is the total number of entries contained in all training texts.

The final feature words are denoted as t1, t2, ..., tn.

Of course, when judging whether the word-sliced entry is a feature word, it is not limited to the above-mentioned CHI algorithm, and other algorithms can also be used, for example, χ ² (t, c)=|AD-BC|.

(b) Generate the feature vector Wi=(wi1,wi2,...,wii,...,win) of the i-th dimension of interest according to the feature words obtained in step (a), where wii is the feature word ti Weights in the i dimensions of interest.

Wii=TFi*log(1+N/GDFi), TFi is the word frequency of the feature word ti appearing in all corpora belonging to the i-th interest dimension, N is the number of documents in all corpus of the feature word ti in all interest dimensions, GDFi (Global document frequency) is the number of documents containing the feature word ti in all corpora of all dimensions of interest.

(c) Generate the feature vector S=(s1, s2, .

Si = the word frequency of the characteristic word ti appearing in the static file.

(d) Calculate the similarity between the user's static profile vector and the feature vector Wi of the i-th interest dimension, and obtain the score value pi of the similarity,

$\mathrm{<span\; class="notranslate">\; Pi</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; wi</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; wi</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; *</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; S</span>}<span\; class="notranslate">\; |</span>$

$<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wi</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; wi</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; win</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; sn</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span><span\; class="notranslate">\; (</span>\sqrt{{\mathrm{<span\; class="notranslate">\; wi</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; wi</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; win</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; *</span>\sqrt{{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; sn</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>$

There are several ways to extract user interest model W2 from historical data searched by users:

W2=d1+d2+d3+...dm, where di is the interest model vector corresponding to a document clicked by the user;

There are two ways to obtain the interest model vector corresponding to a clicked document:

(1) di=(t1, t2, t3,...,tn), when the user clicks on this document recently, tj is equal to the sum of the word frequencies of all words in the document whose type belongs to the jth dimension of interest.

(2) di=(t1, t2, t3, . . . , tn), where di is the similarity score between the document and the i-th dimension of interest. The process of calculating di is as follows:

(a) Extract the feature vocabulary of the classifier, specifically:

(i) Collect corresponding corpus for each interest dimension of the user to generate a corpus;

(ii) performing word segmentation on the corpus to form a series of entries;

(iii) To determine whether the entry after word segmentation is a feature word, the CHI algorithm can be used specifically:

${\mathrm{<span\; class="notranslate">\; \&chi;</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; \&CenterDot;</span>{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; AD</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; BC</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; A</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; B</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; ;</span>$

Among them, the meaning of each parameter is as follows: t: a certain entry; c: a certain category; N: the total number of training texts; A: the number of texts that belong to c and contain t; B: the number of texts that do not belong to c but contain t; C: the number of texts belonging to c but not including t; D: the number of texts not belonging to c nor including t; if both C and D are 0, then χ ² (t, c)=0.

或

低于指定阈值的词条可不考虑作为特征词。The CHI value of term t to the entire training set can be defined as:

or

Entries below the specified threshold may not be considered as feature words.

Set the categories as C ₁ , C ₂ , ..., C _n , and the calculation process of P(c) is as follows:

其中，N(C_i)是类别C_i所包含的训练文本的数量；

Among them, N(C _i ) is the number of training texts contained in category C _i ;

其中，M(C_i)是类别C_i的所有训练文本所包含的词条总数，M是所有训练文本所包含的词条总数。or,

Among them, M(C _i ) is the total number of entries contained in all training texts of category C _i , and M is the total number of entries contained in all training texts.

The final feature words are denoted as t1, t2, ..., tn.

Of course, when judging whether the word-sliced entry is a feature word, it is not limited to the above-mentioned CHI algorithm, and other algorithms can also be used, for example, χ ² (t, c)=|AD-BC|.

(b) Generate the feature vector Wi=(wi1,wi2,...,wii,...,win) of the i-th dimension of interest according to the feature words obtained in step (a), where wii is the feature word ti Weights in the i dimensions of interest.

Wii=TFi*log(1+N/GDFi), TFi is the word frequency of the feature word ti appearing in all corpora belonging to the i-th interest dimension, N is the number of documents in all corpus of the feature word ti in all interest dimensions, GDFi (Global document frequency) is the number of documents containing the feature word ti in all corpora of all dimensions of interest.

(c) According to the feature words obtained in step (a), generate the feature vector V=(v1, v2, ..., vn) of the document, where vi is the weight of the feature word ti in the document, and vi=the feature word ti in The frequency of the term in the document.

(d) Calculate the similarity between the feature vector v of the document and the feature vector Wi of the i-th dimension of interest, and obtain the score value di of the similarity:

$\mathrm{<span\; class="notranslate">\; di</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; wi</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; /</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; wi</span>}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; *</span><span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; V</span>}<span\; class="notranslate">\; |</span>$

$<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wi</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; v</span>}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; wi</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; v</span>}\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; win</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; vn</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\sqrt{{\mathrm{<span\; class="notranslate">\; wi</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; wi</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; win</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; *</span>\sqrt{{\mathrm{<span\; class="notranslate">\; v</span>}\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; v</span>}\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; vn</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span>$

If the user evaluates a clicked document, if the evaluation is good, the di vector is multiplied by a positive constant c, indicating that the importance of the document increases, that is, di=c*di=(c*ti, c*t2, c*t3,...,c*tn); if the evaluation is not good, the di vector is multiplied by the reciprocal of a positive constant c, indicating that the importance of the document is reduced, that is, di=1/c*di =(1/c*ti, 1/c*t2, 1/c*t3, ..., 1/c*tn);

After a period of time, the value of tj will automatically decrease by a certain percentage, indicating that its importance will decrease as time goes by, until the value of tj decreases to zero after a long period of time, then di can be deleted from the history.

W1 and W2 are normalized respectively to obtain a user interest model W=r1*W1+r2*W2, where r1+r2=1.

In the mobile search method of the embodiment of the present invention, according to the user's search request, by calculating the personalized user interest score value of each search type domain, one or several search type domains with high score value are selected for searching, so as to determine the personalized user interest score value for the user. Query classification to provide users with personalized and accurate search results.

In each of the above embodiments, when selecting a search type field, the similarity between the search request and the search type field, the search rate of the search request corresponding to the search type field, and the search type field The personalized user interest score value is used as the basis for selecting the search type field to determine the user's personalized query classification and provide users with personalized and accurate search results.

In the embodiment of the present invention, any two or more of the above-mentioned items may be considered comprehensively to calculate the comprehensive score value of each search type domain, and select one or several search type domains with high comprehensive score values for searching. Hereinafter, the embodiment of the present invention will be described in detail by taking comprehensive consideration of the above three items as the basis for selecting the search type field as an example.

Referring to FIG. 5 , it is another implementation flowchart of the mobile search method according to the embodiment of the present invention.

Step 501, receiving a search request, the search request includes one or more query keywords.

Step 502, respectively calculating the similarity between the search request and each search type field, the public search rate corresponding to each search type field of the search request, and the personalized user interest score value of the search type field.

Step 503 , performing normalization processing on the obtained values corresponding to the search type fields to obtain the comprehensive score value of each search type field.

For example, calculate the similarity between the search request and a certain search type field, and normalize it to obtain the value Score1;

Calculate the public search rate of the search request corresponding to the search type domain, and normalize it to obtain the value Score2;

Calculate the personalized user interest score value of the search type domain and normalize it to obtain the value Score3;

Calculate the comprehensive score value of the search type field=r1*score1+r2*score2+r3*score3, where r1, r2, r3 are the weights of Score1, Score2, and Score3 respectively, and r1+r2+r3+r4=1.

The comprehensive score value can also be calculated in other ways, such as:

Comprehensive score value = score1*score2*score3, or

Comprehensive score value=(score1+score2+score3)/3, etc.

Step 504, select one or more search type domains with high comprehensive score values to search.

It can be seen that in the embodiment of the present invention, a number of factors are comprehensively considered to determine the user's personalized query classification, the comprehensive score value of each search type field is calculated, and one or several search type fields with high comprehensive score values are selected for searching. , so as to provide users with personalized and accurate search results.

Those of ordinary skill in the art can understand that all or part of the steps in the method of the above-mentioned embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage Media, such as: ROM/RAM, disk, CD, etc.

The embodiment of the present invention also provides a mobile search device, as shown in Figure 6, which is a schematic structural diagram of the device:

In this embodiment, the apparatus includes: a receiving unit 601 , a calculating unit 602 , a selecting unit 603 and a searching unit 604 . in:

A receiving unit 601, configured to receive a search request, where the search request includes one or more query keywords;

Calculation unit 602, configured to calculate the score value of each search type field, the score value is the score value of any one of the following items or the comprehensive score value of multiple items: the similarity between the search request and the search type field, the The search request corresponds to the popular search rate of the search type domain and the personalized user interest score value of the search type domain;

The calculation unit 602 calculates the comprehensive score value of each search type field: according to the similarity between the search request and the search type field, the public search rate of the search request corresponding to the search type field, and the personalized user interest score value of the search type field. Product rating, average rating, or weighted rating.

A selection unit 603, selecting one or several search type fields according to the scoring values of each search type field;

The searching unit 604 is configured to use the search type field selected by the selecting unit to search for the query keyword.

In the embodiment of the present invention, when the calculation unit 602 and the selection unit 603 determine the classification of the user's personalized query, there may be multiple implementation methods, for example, it may be based on the similarity between the search request and the search type domain, Select one or several search type domains with high similarity to search; it may also be to select one or several search type domains with high public search rate according to the search request corresponding to the popular search rate of the search type domain; It is also possible to select one or several search type domains with high personalized user interest score values for searching according to the personalized user interest score values of the search type domains. Of course, it is also possible to comprehensively consider the above items, calculate the comprehensive score value of each search type field, and select one or several search type fields with high comprehensive score values for searching. Therefore, the computing unit 602 includes any one or more of the following units:

a similarity calculation unit, configured to calculate the similarity between the search request and each search type field;

a public search rate calculation unit, configured to calculate the public search rate corresponding to each search type field of the search request;

The user interest scoring value calculation unit is used to calculate the personalized user interest scoring value of each search type domain.

Examples are given below to explain this in detail.

As shown in FIG. 7 , it is a schematic structural diagram of a mobile search device according to an embodiment of the present invention.

In this embodiment, the apparatus includes: a receiving unit 701 , a similarity calculating unit 702 , a selecting unit 703 and a searching unit 704 . Wherein, the receiving unit 701 , the selecting unit 703 and the searching unit 704 are consistent with the corresponding units in the embodiment shown in FIG. 6 , and will not be described in detail here.

The similarity calculation unit 702 includes: a weight setting subunit 721 , a query vector generation subunit 722 , a field vector generation unit 723 and a first calculation subunit 724 . Wherein: the weight setting subunit 721 is used to set the weight for the query keyword; the query vector generation subunit 722 is used to generate a query vector by the weight of the query keyword; the field vector generation unit 723 is used to generate the query vector by the weight of the query keyword The weight of each word in the search type domain generates a domain vector corresponding to the search type domain; the first calculation subunit 724 is used to obtain the similarity between the search request and the search type domain by calculating the query vector and domain vector Spend.

In this embodiment, the device may further include: a setting unit (not shown) or a learning unit 705 . Wherein, the setting unit is used to manually determine the subject words and related words in the search type field, as well as the weight of each word; the learning unit 705 is used to determine the search type field through automatic learning. The subject words and related words in , and the weight of each word.

The learning unit 705 includes: a corpus sample acquisition subunit 751 , a thesaurus generation subunit 752 , a weight calculation subunit 753 and a subject term determination subunit 754 . Wherein: the corpus sample acquisition subunit 751 is used for each search type domain, obtains the training text corpus sample corresponding to the search type domain; search thesaurus for type domains;

The weight calculation subunit 753 is used to calculate the weight of each word in the thesaurus; the topic word determination subunit 754 is used to determine the topic word and related words in the search type field according to the weight of each word.

In the embodiment of the present invention, the learning unit 705 may further include: a grade division subunit 755 and a scoring value setting subunit 756 . Wherein, the grade division subunit 755 is used to divide all words in the thesaurus into sets of different grades according to weight; the score value setting subunit 756 is used to set the final score value for the set of each grade, and set The final scoring value of each class is used as the weight of each word in the class.

The mobile search device in the embodiment of the present invention, according to the user's search request, calculates the similarity between the search request and each search type field, and selects one or several search type fields with high similarity for searching, so as to determine the personalized query for the user Classification, to provide users with personalized and accurate search results. For the specific process, reference may be made to the description in the embodiment shown in FIG. 2 above, and details are not repeated here.

As shown in FIG. 8 , it is a schematic diagram of another specific structure of a mobile search device according to an embodiment of the present invention.

In this embodiment, the device includes: a receiving unit 801 , a popular search rate calculating unit 802 , a selecting unit 803 and a searching unit 804 . Wherein, the receiving unit 801 , the selecting unit 803 and the searching unit 804 are consistent with the corresponding units in the embodiment shown in FIG. 6 , and will not be described in detail here.

The public search rate calculation unit 802 includes a second calculation subunit 821 and an addition subunit 822, wherein the second calculation subunit 821 is used to calculate the search type fields corresponding to each query keyword in the search request the popular search rate; the adding subunit 822 is used to use the sum of the popular search rates of the same search type field corresponding to all query keywords in the search request as the popular search rate of the search request corresponding to the search type field .

In this embodiment of the present invention, the public search rate may specifically be the number of public searches. When the second calculation subunit 821 calculates the total number of public searches for a certain search type field corresponding to each keyword in the search request, it can collect all user information about a certain keyword in the search request according to historical records. The sum of the number of times that a certain search type field is used for the search request of a word is selected as the total number of times that the public searches for this search type field corresponding to this keyword, that is, the total number of times that the public searches for this search type field; and then The adding subunit 822 takes the sum of the total number of popular searches in the search type field corresponding to all the keywords in the search request as the total number of popular searches in the search type field corresponding to the search request.

In the embodiment of the present invention, the public search rate may specifically be the number of clicks on public search results. When the second calculation subunit 821 calculates the total number of clicks on the public search results of a certain search type field corresponding to each keyword in the search request, it can collect all users' information about a certain search type field included in the search request according to historical records. The sum of the number of clicks on the search results of a certain search type domain for a search request of a keyword is taken as the total number of clicks on the search results of the search type domain by the public corresponding to the keyword, that is, the number of clicks corresponding to the search type domain The total number of clicks on popular search results; then the summing subunit 822 uses the sum of the total number of clicks on popular search results in the search type field corresponding to all keywords in the search request as the search type corresponding to the search request The total number of popular search result clicks for the domain.

According to the search request of the user, the mobile search device in the embodiment of the present invention calculates the public search rate of each search type domain corresponding to the search request, and selects one or several search type domains with a high public search rate to search, so as to provide users with Determine the classification of personalized queries to provide users with personalized and accurate search results. For the specific process, reference may be made to the description in the embodiment shown in FIG. 3 above, and details are not repeated here.

As shown in FIG. 9 , it is a schematic diagram of another specific structure of a mobile search device according to an embodiment of the present invention.

In this embodiment, the device includes: a receiving unit 901 , a calculation unit 902 for a user interest score value, a selection unit 903 and a search unit 904 . Wherein, the receiving unit 901 , the selecting unit 903 and the searching unit 904 are consistent with the corresponding units in the embodiment shown in FIG. 6 , and will not be described in detail here.

The user interest score calculation unit 902 includes an interest model extraction subunit 921 and a third calculation subunit 922, wherein the interest model extraction subunit 921 is used to extract the user interest model from the user data, and the user interest The model is a vector composed of score values of multiple interest dimensions of the user data; the third calculation subunit 922 is configured to map the search type domain to one of the score values of one or more interest dimensions of the user interest model and the personalized user interest score value as the search type field.

In this embodiment, the interest model of the user is: a static interest model or a dynamic interest model, or an interest model generated by combining the static interest model or the dynamic interest model. For this reason, the interest model extracting subunit 921 may have various structural modes.

The interest model extraction subunit 921 may only include a first extraction subunit (not shown in the figure), which is used to calculate the sum of the word frequencies of all words belonging to each interest dimension in the user's static file, and use it as the sum of the word frequencies corresponding to each interest dimension. The scoring value of each interest dimension, the user interest model is generated as a vector by the scoring value corresponding to each interest dimension;

The interest model extracting subunit 921 may also only include a second extracting subunit (not shown in the figure), which is used to calculate the word frequency of all words belonging to each interest dimension in the clicked document in the history record history record of the user search and take it as a score value corresponding to each interest dimension, and use the score value corresponding to each interest dimension as a vector to generate a dynamic interest model of the user.

As shown in FIG. 10 , the interest model extraction subunit 921 may further include the first extraction subunit 1001 and the second extraction subunit 1002 , as well as a first processing subunit 1003 and a first weighting subunit 1004 . Wherein, the first processing subunit 1003 is used to normalize the static interest model and the dynamic interest model respectively; the first weighting subunit 1004 is used to calculate the normalized static interest model and The sum of dynamic interest models, and use this sum as the user's interest model.

As shown in FIG. 11 , the interest model extraction subunit 921 may further include the first extraction subunit 1101 and the second extraction subunit 1102 , as well as a second weighting subunit 1103 and a second processing subunit 1104 . Wherein, the second weighting subunit 1103 is configured to perform weighted addition of the static interest model and the dynamic interest model; the second processing subunit 1104 is configured to normalize the results output by the second weighting subunit normalized, and use the normalized result as the user's interest model.

According to the user's search request, the mobile search device in the embodiment of the present invention calculates the personalized user interest score value of each search type domain, and selects one or several search type domains with high score values to search, so as to determine the personalization for the user. Query classification to provide users with personalized and accurate search results. For the specific process, reference may be made to the description in the mobile search method in the embodiment of the present invention.

In the mobile search device in each of the above embodiments, when selecting a search type domain, the similarity between the search request and the search type domain, the public search rate of the search request corresponding to the search type domain, And the personalized user interest score value of the search type field is used as the basis for selecting the search type field to determine the user's personalized query category and provide the user with personalized and accurate search results.

In the embodiment of the present invention, any two or more of the above-mentioned items may be considered comprehensively to calculate the comprehensive score value of each search type domain, and select one or several search type domains with high comprehensive score values for searching. Hereinafter, the embodiment of the present invention will be described in detail by taking comprehensive consideration of the above three items as the basis for selecting the search type field as an example.

Referring to FIG. 12 , it is another structural diagram of a mobile search device according to an embodiment of the present invention.

In this embodiment, the apparatus includes: a receiving unit 1201 , a calculating unit 1202 , a selecting unit 1203 and a searching unit 1204 . Wherein, the receiving unit 1201 is used to receive a search request, and the search request contains one or more query keywords; the calculation unit 1202 is used to calculate the score value of each search type field, and the score value is any one of the following Score value or comprehensive score value of multiple items: the similarity between the search request and the search type domain, the public search rate corresponding to the search request type domain, and the personalized user interest score value of the search type domain The selection unit 1203 is used to select one or more of the search type fields according to the scoring value of each search type field; the search unit 1204 is configured to use the search type field selected by the selection unit to search for the query keyword.

In this embodiment, the calculation unit 1202 includes: a similarity calculation unit 1221 , a public search rate calculation unit 1222 , a user interest score value calculation unit 1223 , a normalization processing unit 1224 and a comprehensive processing unit 1225 . Among them, the similarity calculation unit 1221 is used to calculate the similarity between the search request and each search type domain; the public search rate calculation unit 1222 is used to calculate the public search rate of each search type domain corresponding to the search request; user interest Score value calculation unit 1223, used to calculate the personalized user interest score value of each search type field; normalization processing unit 1224, used to respectively calculate the similarity calculation unit, the popular search rate calculation unit and the user The value calculated by the interest score value calculation unit is normalized; the comprehensive processing unit 1225 is used for comprehensive calculation of any two or more normalized values obtained by the normalization processing unit 1224, such as: product , average or weighted sum, etc., to obtain the score value of each search type field.

It can be seen that the mobile search device in the embodiment of the present invention comprehensively considers multiple factors to determine the user's personalized query classification, calculates the comprehensive score value of each search type field, and selects one or several search type fields with high comprehensive score values. Search, which can provide users with personalized and accurate search results.

The embodiments of the present invention have been described in detail above, and the present invention has been described using specific implementation methods herein. The descriptions of the above embodiments are only used to help understand the method and equipment of the present invention; meanwhile, for those of ordinary skill in the art, According to the idea of the present invention, there will be changes in the specific implementation and scope of application. To sum up, the contents of this specification should not be construed as limiting the present invention.

Claims (27)

1. A mobile search method, characterized in that, comprising: Receiving a search request, the search request includes one or more query keywords; Calculate the score value of each search type field, and the score value is the score value of any one of the following items or the comprehensive score value of multiple items: the similarity between the search request and the search type field, the search request corresponding to the The popular search rate of the search type domain, the personalized user interest score value of the search type domain; According to the scoring value of each search type field, one or several search type fields are selected to search for the query keyword. 2. The method according to claim 1, wherein the calculation of the comprehensive score value of each search type field is based on the similarity between the search request and the search type field, the search request corresponding to the search The product score value, average score value or weighted score value are calculated from the mass search rate of the type domain and the personalized user interest score value of the search type domain. 3. The method according to claim 1, wherein the calculating the similarity between the search request and the search type field comprises: Setting weights for the query keywords; generating a query vector by the weight of the query keyword; Generate a domain vector corresponding to the search type domain by the weight of each word in the search type domain; The similarity between the search request and the search type domain is obtained by calculating the query vector and the domain vector. 4. method according to claim 3, is characterized in that, described method also comprises: Manually determine the subject terms and related terms in the search type field, and the weight of each term; or The subject words and related words in the search type domain, and the weight of each word are determined through automatic learning. 5. method according to claim 4, is characterized in that, described subject word and related word in described search type domain are determined by automatic learning mode, and the weight of each word comprises: For each search type domain, obtain a training text corpus sample corresponding to the search type domain; Carry out word segmentation to described corpus sample, generate the thesaurus of this search type field; Calculate the weight of each word in the thesaurus; Subject words and related words in the search type domain are determined according to the weights of each word. 6. method according to claim 5, is characterized in that, described subject word and related word in described search type field are determined by automatic learning mode, and the weight of each word also comprises: Divide all the words in the thesaurus into sets of different grades according to weight; Set the final scoring value for the collection of each grade, and use the final scoring value of each grade as the weight of each word in the grade. 7. The method according to claim 3, wherein said setting weight for said query keyword comprises: Set the same weight for all query keywords; or Set a maximum weight for the top keywords, a medium-sized weight for the middle keywords, and a minimum weight for the bottom keywords. 8. The method according to claim 1, wherein said calculating the public search rate of said search request corresponding to said search type field comprises: Calculate the public search rate of each search type field corresponding to each query keyword in the search request; The sum of the popular search rates of the same search type field corresponding to all query keywords in the search request is taken as the popular search rate of the search request corresponding to the search type field. 9 . The method according to claim 8 , wherein the public search rate is: the number of public searches, or the number of clicks on public search results. 10. The method according to claim 1, wherein said calculating the personalized user interest score value of said search type domain comprises: Extracting the user's interest model from the user data, the user's interest model is a vector composed of the score values of the user data for multiple interest dimensions; The sum of the score values of the search type field corresponding to one or more interest dimensions of the user interest model is used as the personalized user interest score value of the search type field. 11. The method according to claim 10, wherein the user's interest model is: a static interest model or a dynamic interest model; The static interest model of users extracted from user data includes: Calculate the sum of the word frequencies of all words belonging to each interest dimension in the user's static profile, and use it as the score value corresponding to each interest dimension; or, calculate the similarity score value between the user's static profile and each interest dimension, And take it as the scoring value corresponding to each dimension of interest; generating the user interest model by using the scoring value corresponding to each interest dimension as a vector; The user's dynamic interest model extracted from user data includes: Calculate the sum of the word frequencies of all words belonging to each interest dimension in the user's search click history, and use it as the score value corresponding to each interest dimension; or, calculate the similarity score between the search click history and each interest dimension value, and use it as the scoring value corresponding to each dimension of interest; The user's dynamic interest model is generated by using the scoring value corresponding to each interest dimension as a vector. 12. The method according to claim 11, wherein said extracting the user's interest model from the user data further comprises: respectively performing normalization processing on the static interest model and the dynamic interest model; The sum of the one or more static interest models and the one or more dynamic interest models after normalization processing is calculated, and the sum is used as the interest model of the user. 13. The method according to claim 11, wherein said extracting the user's interest model from the user data further comprises: performing weighted addition of one or more of the static interest models and one or more of the dynamic interest models; The sum of the weighted additions is normalized, and the normalized result is used as the interest model of the user. 14. The method according to claim 1, wherein said calculating the weighted scoring value of each search type field comprises: calculating the similarity between the search request and the search type field, and normalizing it; calculating the popular search rate of the search request corresponding to the search type domain, and normalizing it; Calculating the personalized user interest score value of the search type domain, and normalizing it; Weighted addition of any two or more normalized values above to obtain the weighted score value of the search type field. 15. A mobile search device, characterized in that it comprises: a receiving unit, configured to receive a search request, wherein the search request includes one or more query keywords; A calculation unit, configured to calculate the score value of each search type field, and the score value is a score value of any one of the following items or a comprehensive score value of multiple items: the similarity between the search request and the search type field, the The search request corresponds to the popular search rate of the search type domain and the personalized user interest score value of the search type domain; Selection unit, select one or several search type fields according to the scoring value of each search type field; A search unit, configured to search for the query keyword using the search type field selected by the selection unit. 16. The device according to claim 15, wherein the calculation unit calculates the comprehensive score value of each search type field according to the similarity between the search request and the search type field, and the corresponding search request. Calculate the product score value, average score value or weighted score value of multiple items in the popular search rate of the above search type domain and the personalized user interest score value of the search type domain. 17. The device according to claim 15, wherein the calculation unit comprises any one or more of the following units: a similarity calculation unit, configured to calculate the similarity between the search request and each search type field; a public search rate calculation unit, configured to calculate the public search rate corresponding to each search type field of the search request; The user interest scoring value calculation unit is used to calculate the personalized user interest scoring value of each search type domain. 18. The device according to claim 17, wherein the similarity calculation unit comprises: a weight setting subunit, configured to set weights for the query keywords; a query vector generating subunit, configured to generate a query vector by the weight of the query keyword; A domain vector generating unit, configured to generate a domain vector corresponding to the search type domain from the weight of each word in the search type domain; The first calculation subunit is configured to obtain the similarity between the search request and the search type domain by calculating the query vector and the domain vector. 19. The device according to claim 18, further comprising: A setting unit is used to manually determine the subject words and related words in the search type field, as well as the weight of each word; or The learning unit is used to determine the subject words and related words in the search type field and the weight of each word through automatic learning. 20. The device according to claim 19, wherein the learning unit comprises: The corpus sample acquisition subunit is used to obtain, for each search type domain, a training text corpus sample corresponding to the search type domain; Thesaurus generation subunit is used to carry out word segmentation to described corpus sample, generates the thesaurus of this search type field; a weight calculation subunit for calculating the weight of each word in the thesaurus; The subject word determination subunit is used to determine subject words and related words in the search type field according to the weight of each word. 21. The device according to claim 20, wherein the learning unit further comprises: A grade division subunit is used to divide all the words in the thesaurus into sets of different grades according to their weights; The scoring value setting subunit is used to set the final scoring value for the set of each grade, and use the final scoring value of each grade as the weight of each word in the grade. 22. The device according to claim 17, wherein the public search rate calculation unit comprises: The second calculation subunit is used to calculate the public search rate of each search type field corresponding to each query keyword in the search request; The adding subunit is configured to use the sum of the popular search rates of the same search type field corresponding to all the query keywords in the search request as the popular search rate of the search request corresponding to the search type field. 23. The device according to claim 17, wherein the user interest score calculation unit comprises: The interest model extraction subunit is used to extract the user's interest model from the user data, and the user's interest model is a vector composed of the score values of the user data for multiple interest dimensions; The third calculation subunit is configured to use the sum of the score values of the search type field corresponding to one or more interest dimensions of the user interest model as the personalized user interest score value of the search type field. 24. The device according to claim 23, wherein the user's interest model is: a static interest model or a dynamic interest model; The interest model extraction subunit includes: The first extraction subunit is used to calculate the sum of the word frequencies of all words belonging to each interest dimension in the user's static profile, and use it as the score value corresponding to each interest dimension, or calculate the user's static profile and each interest The similarity score value of the dimension, and use it as the score value corresponding to each interest dimension, and generate the user interest model by using the score value corresponding to each interest dimension as a vector; or The second extraction subunit is used to calculate the sum of the word frequencies of all words belonging to each interest dimension in the user's search click history record, and use it as a score value corresponding to each interest dimension, or calculate the search click history record and each The similarity score value of each interest dimension is used as the score value corresponding to each interest dimension, and the dynamic interest model of the user is generated by using the score value corresponding to each interest dimension as a vector. 25. The device according to claim 23, wherein the interest model extraction subunit further comprises: A first processing subunit, configured to perform normalization processing on the static interest model and the dynamic interest model respectively; The first weighting subunit is configured to calculate the sum of one or more static interest models and one or more dynamic interest models after normalization processing, and use the sum as the interest model of the user. 26. The device according to claim 23, wherein the interest model extraction subunit further comprises: A second weighting subunit, configured to weight and add one or more of the static interest models and one or more of the dynamic interest models; The second processing subunit is configured to normalize the results output by the second weighting subunit, and use the normalized results as the interest model of the user. 27. The device according to claim 23, wherein the computing unit further comprises: A normalization processing unit is used to perform normalization processing on the values calculated by the similarity calculation unit, the popular search rate calculation unit and the user interest score value calculation unit respectively; A weighting processing unit, configured to perform weighted addition of any two or more normalized values obtained by the normalization processing unit to obtain the scoring value of each search type field.

Images