CN111737544A - Search intention recognition method and device, electronic equipment and storage medium - Google Patents
Search intention recognition method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN111737544A CN111737544A CN202010401669.7A CN202010401669A CN111737544A CN 111737544 A CN111737544 A CN 111737544A CN 202010401669 A CN202010401669 A CN 202010401669A CN 111737544 A CN111737544 A CN 111737544A
- Authority
- CN
- China
- Prior art keywords
- search
- intention
- search intention
- intersection
- strength
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 70
- 238000003860 storage Methods 0.000 title claims abstract description 16
- 238000013145 classification model Methods 0.000 claims abstract description 68
- 238000010606 normalization Methods 0.000 claims description 33
- 238000012545 processing Methods 0.000 claims description 32
- 230000008569 process Effects 0.000 claims description 12
- 230000004927 fusion Effects 0.000 claims description 10
- 230000008901 benefit Effects 0.000 abstract description 9
- 238000005457 optimization Methods 0.000 abstract description 5
- 230000009286 beneficial effect Effects 0.000 abstract 1
- 235000013305 food Nutrition 0.000 description 13
- 238000010586 diagram Methods 0.000 description 10
- 238000012549 training Methods 0.000 description 7
- 235000013527 bean curd Nutrition 0.000 description 6
- 238000009826 distribution Methods 0.000 description 6
- 238000003058 natural language processing Methods 0.000 description 6
- 238000004590 computer program Methods 0.000 description 5
- 230000003993 interaction Effects 0.000 description 4
- 238000013459 approach Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 239000004278 EU approved seasoning Substances 0.000 description 1
- 241000287828 Gallus gallus Species 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 235000011194 food seasoning agent Nutrition 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000003340 mental effect Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/903—Querying
- G06F16/9032—Query formulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/906—Clustering; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application discloses a search intention identification method, a search intention identification device, an electronic device and a storage medium. The method comprises the following steps: identifying a first set of search intents for the search request based on a binary model; identifying a second search intention set of the search request based on the multi-classification model, wherein the search intents in the second search intention set have intention strength values; and solving an intersection of the first search intention set and the second search intention set, taking the search intention in the intersection as the search intention corresponding to the search request, and determining a search intention display mode according to the intention strength value of the search intention in the intersection. The technical scheme integrates the advantages of single two-classification and unified multi-classification, ensures the comparability of the intentions, decouples the search intention identification and the search intention strength ordering, and is beneficial to the iterative optimization of the respective search intention identification of the services corresponding to the search intentions.
Description
Technical Field
The application relates to the field of search engines, in particular to a search intention identification method and device, electronic equipment and a storage medium.
Background
Search intent recognition is crucial in search scenarios, and better search results can only be recalled to satisfy a user if the user's search intent is accurately recognized. The search intention generally refers to the real needs of the user reflected in the back of the search behavior, for example, searching for "badminton", possibly because the user wants to buy badminton equipment, possibly searching for badminton stadiums, possibly learning badminton rules, and the like. In this example, "buy instruments", "find venues", and "learn rules" are three different types of search intentions related to the search keyword "badminton".
The search intentions of a user at a certain time, a certain place and a certain scene may be unique or multiple, and currently, the identification of the search intentions is usually solved as a classification problem, that is, multiple types of search intentions are preset to determine which types of search intentions the search request corresponds to.
One method commonly used is to perform a plurality of separate binary classifications, i.e., determine whether a search request corresponds to each type of search intention. The disadvantage of this is that there is no comparability between the results of the individual two classifications, i.e. no strength comparison can be made, but for a certain user, the primary search intention should be primary and secondary with search intention at a specific time, place and scene, and the primary search intention should be stronger than the secondary search intention, so the prior art cannot meet the user's needs.
Disclosure of Invention
In view of the above, the present application is made to provide a search intention identifying method, apparatus, electronic device, and storage medium that overcome or at least partially solve the above problems.
According to a first aspect of the present application, there is provided a search intention identification method including: identifying a first set of search intents for the search request based on a binary model; identifying a second set of search intentions of the search request based on a multi-classification model, the search intentions in the second set of search intentions having intent strength values; and solving an intersection of the first search intention set and the second search intention set, taking the search intention in the intersection as the search intention corresponding to the search request, and determining a search intention display mode according to the intention strength value of the search intention in the intersection.
Optionally, in the above method, the identifying a first search intention set of the search request based on the binary model includes: respectively identifying the search requests after normalization processing based on a plurality of two classification models, wherein each two classification model corresponds to a preset search intention; acquiring the recognition result of each binary model; and determining the first search intention set according to the recognition result of each classification model.
Optionally, in the foregoing method, the identifying the search request after the normalization process includes: matching the search request after normalization processing with a first word list; and under the condition that a complete matching item exists in the first word list, taking a preset search intention corresponding to the binary classification model as the identified search intention.
Optionally, in the foregoing method, the identifying the search request after the normalization process includes: matching the search request after normalization processing with a second word list; and under the condition that the partial matching item exists in the second word list, taking the preset search intention corresponding to the binary classification model as the identified search intention.
Optionally, in the foregoing method, the identifying the search request after the normalization process includes: identifying a predicted score of the search request after normalization processing; and under the condition that the prediction score is larger than a positive threshold value and a negative threshold value, taking a preset search intention corresponding to the binary classification model as the identified search intention.
Optionally, in the above method, the identifying a second search intention set of the search request based on the multi-classification model includes: generating a search intent feature of the search request; and identifying the search intention characteristics by utilizing the multi-classification model to obtain the second search intention set.
Optionally, the determining a search intention presentation manner according to the intention strength value of the search intention in the intersection includes: and generating a search result block according to the search intention in the intersection, and determining the display priority of the search block according to the intention strength value of each search intention in the search block.
Optionally, the determining a search intention presentation manner according to the intention strength value of the search intention in the intersection includes: and determining the display sequence of the search intention according to the intention strength value of the search intention in the intersection.
Optionally, the search intents in the first set of search intents have an intention strength adjustment parameter; the method for determining the search intention display mode according to the intention strength value of the search intention in the intersection comprises the following steps: and determining an intention adjusting value according to the intention strength adjusting parameter and the intention strength value of the search intention in the intersection, and determining a search intention display mode according to the intention strength adjusting value of the search intention in the intersection.
Optionally, the method further comprises: and under the condition that the search intention in the first search intention set hits a cold start rule, updating the intention strength adjustment parameter of the corresponding search intention according to the hit cold start rule.
According to a second aspect of the present application, there is provided a search intention recognition apparatus including: a first identification unit, configured to identify a first search intention set of the search request based on a classification model; a second identification unit, configured to identify a second search intention set of the search request based on a multi-classification model, where search intents in the second search intention set have intention strength values; and the fusion unit is used for solving an intersection of the first search intention set and the second search intention set, taking the search intention in the intersection as the search intention corresponding to the search request, and determining a search intention display mode according to the intention strength value of the search intention in the intersection.
Optionally, in the above apparatus, the first identifying unit is configured to identify the search request after the normalization processing based on a plurality of two classification models, where each two classification models corresponds to a preset search intention; acquiring the recognition result of each binary model; and determining the first search intention set according to the recognition result of each classification model.
Optionally, in the above apparatus, the first identifying unit is configured to match the search request after the normalization processing with a first vocabulary; and under the condition that a complete matching item exists in the first word list, taking a preset search intention corresponding to the binary classification model as the identified search intention.
Optionally, in the above apparatus, the first identifying unit is configured to match the search request after the normalization processing with a second vocabulary; and under the condition that the partial matching item exists in the second word list, taking the preset search intention corresponding to the binary classification model as the identified search intention.
Optionally, in the apparatus, the first identifying unit is configured to identify a predicted score of the search request after the normalization processing; and under the condition that the prediction score is larger than a positive threshold value and a negative threshold value, taking a preset search intention corresponding to the binary classification model as the identified search intention.
Optionally, in the above apparatus, the second identifying unit is configured to generate a search intention characteristic of the search request; and identifying the search intention characteristics by utilizing the multi-classification model to obtain the second search intention set.
Optionally, the fusion unit is configured to generate a search result block according to the search intention in the intersection, and determine a display priority of the search block according to an intention strength value of each search intention in the search block.
Optionally, the fusion unit is configured to determine a display order of the search intentions according to the intention strength values of the search intentions in the intersection.
Optionally, the search intents in the first set of search intents have an intention strength adjustment parameter; and the fusion unit is used for determining an intention adjusting value according to the intention strength adjusting parameter and the intention strength value of the search intention in the intersection and determining a search intention display mode according to the intention strength adjusting value of the search intention in the intersection.
Optionally, the apparatus further comprises: and the cold start adjusting unit is used for updating the intention intensity adjusting parameters of the corresponding search intentions according to the hit cold start rule under the condition that the search intentions in the first search intention set hit the cold start rule.
In accordance with yet another aspect of the present application, there is provided an electronic device including: a processor; and a memory arranged to store computer executable instructions that, when executed, cause the processor to perform a method as any one of the above.
According to a further aspect of the application, there is provided a computer readable storage medium, wherein the computer readable storage medium stores one or more programs which, when executed by a processor, implement a method as in any above.
According to the technical scheme, a first search intention set of the search request is identified based on a binary model, and search intents in the first search intention set have the strong and weak intents; identifying a second search intention set of the search request based on the multi-classification model, wherein the search intents in the second search intention set have intention strength values; and solving an intersection of the first search intention set and the second search intention set, taking the search intention in the intersection as the search intention corresponding to the search request, and determining a search intention display mode according to the intention strength value of the search intention in the intersection. The technical scheme integrates the advantages of single two-classification and unified multi-classification, the recall rate is ensured through the search intents determined by a plurality of single two-classification, the comparability of the strength relationship is ensured through the search intents determined by unified multi-classification, and the display interaction requirements of the service are met; the search intention identification and the search intention strength ordering are decoupled, so that the iterative optimization of the respective search intention identification of the services corresponding to the search intents is facilitated, the intention strength ordering is uniformly performed on the platform side, and the problems of service conflict and exclusive exhibition can be better solved.
The foregoing description is only an overview of the technical solutions of the present application, and the present application can be implemented according to the content of the description in order to make the technical means of the present application more clearly understood, and the following detailed description of the present application is given in order to make the above and other objects, features, and advantages of the present application more clearly understandable.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the application. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:
FIG. 1 shows a schematic flow diagram of a search intent recognition method according to one embodiment of the present application;
FIG. 2 shows a schematic flow diagram of another search intent recognition method according to an embodiment of the present application;
FIG. 3 shows a schematic structural diagram of a search intention recognition apparatus according to an embodiment of the present application;
FIG. 4 shows a schematic structural diagram of an electronic device according to an embodiment of the present application;
FIG. 5 shows a schematic structural diagram of a computer-readable storage medium according to an embodiment of the present application.
Detailed Description
Instead of using a plurality of individual binary classifications, a unified multi-classification approach may be used, i.e. outputting a plurality of search intents as a whole, such as: the intention strength score of the a search intention is 0.5, the intention strength score of the B search intention is 0.3, and the intention strength score of the C search intention is 0.2. Thus, the search intents have the comparability of the intention strength.
However, the problem of this approach is that multiple classifications depend on multiple-intent unified modeling and sample training implementation, and for those search intent classes with insufficient samples and sparse features, the search intent recognition results generated by the trained search intent recognition model tend to be biased toward search intent classes with many samples and dense features, which results in that the search intent recognition results of multiple classifications are not well classified individually. Meanwhile, if one search intention category is added to the recognition target, the sample needs to be reconstructed, and the whole multi-classification model needs to be retrained, so that the cost is very high, and the expansion of the search intention category is not facilitated.
Based on the above, the present application provides a search intention recognition method that combines a plurality of separate two classifications and a unified multiple classification, and can combine the advantages of the two classifications.
The embodiments of the present application can be applied to various scenarios using search engine technology, including but not limited to general search engines such as Baidu, Google (the trade name is merely exemplary), special search engines in the fields of patent, trademark, etc., and search engines within APP.
The user may generate the search request in various manners such as text, image, voice, etc., for example, the text may be a search keyword or a representation of a search sentence.
For example, the search intention may include take-out, food, menu, comment, offer, etc., which reflect the user's needs, and may be specified by the business, domain expert, etc. for name determination and classification of the search intention. In other words, it can be understood as a generalized user requirement.
Specifically, in a business scenario, the search intention may correspond to a category of goods or services, and the category of goods and services may be defined according to business requirements, for example, the takeout and the food service provided above are categories of service providing manners.
A search result may correspond to one or more search intentions, for example, if a restaurant offers both hall food sales and takeaway services, then the search intentions corresponding to the restaurant may include takeaway and hall food; and another restaurant only provides takeaway services, the search intent corresponding to that restaurant only includes takeaway. Conversely, it is apparent that a search intent can also correspond to one or more search results, and typically a plurality of search results, such as many restaurants that provide take-away services. The more the search intent matches the user's real needs, the more easily the search results presented to the user will correspondingly achieve the user's search goals.
Exemplary embodiments of the present application will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
Fig. 1 shows a schematic flow diagram of a search intention recognition method according to an embodiment of the present application. As shown in fig. 1, the method includes:
in step S110, a first search intention set of the search request is identified based on the binary model.
Step S120, a second search intention set of the search request is identified based on the multi-classification model, and the search intentions in the second search intention set have intention strength values.
Here, the intention strength values may be compared with each other for intention strength, such as the intention strength scores exemplified above. That is, the intentions may be compared between the search intentions in the second set of search intentions. In contrast, the search intents in the first set of search intents do not have a basis for comparison with each other.
Step S130, solving an intersection of the first search intention set and the second search intention set, taking the search intention in the intersection as the search intention corresponding to the search request, and determining a search intention display mode according to the intention strength value of the search intention in the intersection.
The search intention can be directly displayed in a search result page as guide information, and the ranking of the search intention can be indirectly reflected through the ranking of the search results.
It can be seen that the method shown in fig. 1 integrates the advantages of the single two-classification and the unified multi-classification, ensures the recall rate through the search intents determined by the single two-classification, ensures the comparability of the strong and weak relationship through the search intents determined by the unified multi-classification, and meets the display interaction requirements of the service; the search intention identification and the search intention strength ordering are decoupled, so that the iterative optimization of the respective search intention identification of the services corresponding to the search intents is facilitated, the intention strength ordering is uniformly performed on the platform side, and the problems of service conflict and exclusive exhibition can be better solved.
In one embodiment of the present application, the method wherein identifying the first set of search intents for the search request based on the classification model comprises: respectively identifying the search requests after the normalization processing based on a plurality of two classification models, wherein each two classification model respectively corresponds to a preset search intention; acquiring the recognition result of each binary model; and determining a first search intention set according to the recognition result of each classification model.
In the embodiment of the present application, the search request may be normalized in advance, and the normalization process is a preprocessing such as word segmentation in Natural Language Processing (NLP).
For example, three categories of searching intentions, such as food, tour, and hotel, can be respectively identified by using a two-category model, and whether the searching intention of a searching request is food, tour, and hotel is accordingly identified. Since the difficulty in determining "yes" or "no" of a problem is less than the difficulty in accurately classifying an object, the recall rate and accuracy of a two-class model are generally better than those of a multi-class model. In this embodiment, a plurality of separate binary models are used to determine whether the two models are true or not. Of course, the recognition result finally output by the binary classification model may not only be simple yes or no, but also contain intention strength information.
In an embodiment of the present application, in the method, identifying the search request after the normalization process includes: matching the search request after the normalization processing with a first word list; and under the condition that a complete matching item exists in the first word list, taking a preset search intention corresponding to the binary classification model as the identified search intention.
The first vocabulary may be referred to herein as a perfect match vocabulary, i.e., a search request may require a perfect match with an entry in the first vocabulary to consider the result of the second classification as "yes".
In an embodiment of the present application, in the method, identifying the search request after the normalization process includes: matching the search request after the normalization processing with a second word list; and under the condition that the partial matching item exists in the second word list, taking the preset search intention corresponding to the binary classification model as the identified search intention.
The second vocabulary can be regarded as a pattern matching vocabulary, pattern (pattern) matching is a way for a user to perform fuzzy matching in search engine technology, and whether a search request hits a certain pattern can be judged by presetting a pattern rule. For example, the search request of the user is "log-house barbeque", and the pattern rule may be ". multidot.barbeque" (i.e., "xx barbeque" or "barbeque xx" both belong to this pattern), so that it can be recognized that the search intention is the "barbeque" classification under "food".
Of course, the specific type of the search intention may be determined according to the needs, and in the last example, the search intention of "barbeque in a cabin" may be considered as "food", or may be considered as "barbeque under" food ", and may be determined according to the needs in business.
In an embodiment of the present application, the determining, according to the intention strength value of the search intention in the intersection, a search intention presentation manner includes: and generating a search result block according to the search intention in the intersection, and determining the display priority of the search block according to the intention strength value of each search intention in the search block.
The search result cluster is an aggregation of search results as the name implies, and specifically, the search results may be clustered according to categories to obtain the search result cluster. There are many clustering methods, such as clustering according to semantic relevance, clustering according to geographical locations corresponding to search results, and so on.
One preferred approach is to cluster according to the user's search intent. In one embodiment of the present application, each search result block corresponds to one or more search intents, and each search intention corresponds to one or more business categories.
For example, each search intention corresponds to a local life service category, which may include take-out, flash purchase, menu, and the like. See the following example of this particular scenario: the user searches for the wife bean curd, and the search result related to the wife bean curd includes a merchant who uses the wife bean curd as a takeaway commodity, a supermarket that sells bean curd and seasonings, a practice of the wife bean curd, and the like. Then at least three search result clusters may be generated: take-out blocks, flash purchase blocks and menu blocks, take-out blocks as examples, and the search results covered by the blocks are the merchants taking Mapo tofu as take-out commodities.
A search result may correspond to one or more search intentions, for example, if a restaurant offers both hall food sales and takeaway services, then the search intentions corresponding to the restaurant may include takeaway and hall food; and another restaurant only provides takeaway services, the search intent corresponding to that restaurant only includes takeaway. Conversely, it is apparent that a search intent can also correspond to one or more search results, and typically a plurality of search results, such as many restaurants that provide take-away services. The more the search intent matches the user's real needs, the more easily the search results presented to the user will correspondingly achieve the user's search goals.
Therefore, the search results are clustered according to the search intention and the search result cluster block is generated, so that the user can pay attention to the required search results more quickly. And determining the display priority of the search result blocks according to the intention strength value of the search intention, so that the search result with stronger search intention can be displayed preferentially.
Specifically, a search result set can be obtained according to each search intention, so that the intention strength value of the search intention can be converted into the display weight of the search result, and the display priority of the whole search result block is determined according to the display weight of each search result in each search result block.
In an embodiment of the present application, the determining, according to the intention strength value of the search intention in the intersection, a search intention presentation manner includes: and determining the display sequence of the search intention according to the intention strength value of the search intention in the intersection.
Besides indirectly reflecting the strength of the search intention in a search result block mode, the search intention can also be directly shown, and the strength of the search intention is reflected. For example, the search intention may appear on the search result page in the form of a guide term or the like, that is, the respective guide terms may be presented in the determined order of presentation of the search intention.
For example, the search intention in the second search intention set is A, B, C, D, E, and the intention strength values are 0.3, 0.15, 0.25, 0.2 and 0.1 in turn. The search intention in the first search intention is A, B, E, F, the search intentions in the intersection are A, B and E, and the search intention presentation order may be a → B → E since the intention strength value of a is 0.3, the intention strength value of B is 0.15, and the intention strength value of E is 0.1.
In one embodiment of the present application, in the above method, the search intention in the first search intention set has an intention strength adjustment parameter; determining a search intention display mode according to the intention strength value of the search intention in the intersection comprises the following steps: and determining an intention adjusting value according to the intention strength adjusting parameter and the intention strength value of the search intention in the intersection, and determining a search intention display mode according to the intention strength adjusting value of the search intention in the intersection.
The intention strength adjustment parameter may be a coarser grained division of the intention strength, for example, into three levels (levels): the items "strong", "medium" and "weak" may also be intention scores, but there is no comparability between the search intentions, for example, the user searches "palace chicken dices", the user's demand for take-out is higher than that of cannibalism, but the thick attribute of the intention of the search intention of "take-out" may be "weak", and the thick attribute of the intention of the search intention of "cannibalism" may be "medium".
Although the intention strength adjustment parameters of the search intentions in the first search intention set are not comparable, the intention strength adjustment parameters may be used to adjust the intention strength values, for example, to determine weights according to the intention strength adjustment parameters and to perform weighting processing on the intention strength values, respectively.
In the foregoing embodiment, the intention strength adjustment parameter may be preset for each item in the first vocabulary, and the intention strength adjustment parameter of a completely matching item may be used as the identified intention strength adjustment parameter.
Similarly to the first vocabulary, each item in the second vocabulary may also be preset with an intention strength adjustment parameter, and the intention strength adjustment parameter of a partially matching item may be used as the identified intention strength adjustment parameter.
In addition, even if the intention strength value is not adjusted by the intention strength adjustment parameter, the intention strength value of each search intention in the intersection may be subjected to numerical normalization processing (different from the normalization processing of the natural language processing) to facilitate other processing to be performed subsequently.
For example, the search intention in the first search intention is A, B, E, F, the search intention in the second search intention set is A, B, C, D, E, and the intentions are 0.3, 0.15, 0.25, 0.3 and 0.05 in sequence. Then the search intentions in the intersection are A, B and E, the intentions of A are 0.3, B are 0.15, E are 0.05, after normalization, the adjustment of the intentions of A is 0.6, B is 0.3, and E is 0.1.
In an embodiment of the present application, in the method, identifying the search request after the normalization process includes: identifying a predicted score of the search request after the normalization processing; and under the condition that the prediction score is greater than the positive and negative threshold values, taking the preset search intention corresponding to the two classification models as the identified search intention, and determining the identified intention strength adjustment parameters according to the prediction score and the intention strength grade threshold value.
In addition to using vocabularies for matching, recognition of search intent may also be accomplished using other natural language processing techniques, such as using BERT (Bidirectional Encoder Representation from Transformers, a model proposed by Google that relies on self-attention to compute its input and output representations rather than using a sequence-aligned recurrent neural network or a convolutional transformation model). The BERT model can be obtained by pre-selecting a sample for training.
Of course, in other embodiments, the model for identifying the predicted score may also be other natural language processing models, and when different natural language processing models are used for identifying the predicted score, the positive and negative thresholds and the intention strength level threshold need to be correspondingly predetermined according to the numerical distribution. It is easy to understand that the positive and negative thresholds are the recognition results of yes and no obtained by the user, and the intention strength level threshold is used to determine which level the intention strength coarse attribute belongs to.
For example, if the prediction score is 0.66 and the positive and negative thresholds are 0.6, a recognition result of "yes" is obtained, and it is further determined that it belongs to the "weak" intention level based on the intention level threshold of "medium" level 0.8 and the intention level threshold of "weak" level 0.6.
In one embodiment of the present application, the method wherein identifying the second set of search intents for the search request based on the multi-classification model comprises: generating a search intention characteristic of the search request; and identifying the search intention characteristics by using a multi-classification model to obtain a second search intention set.
Regarding the generated search intention features, the technology of feature engineering is mainly involved, and specifically, two aspects of feature selection and feature processing can be included.
Regarding feature selection, the search intention feature may be a single feature, such as a text-type feature, a spatio-temporal-type feature, a user-type feature, a statistical-type feature, or a composite feature obtained by performing a connection (Concat) operation on features. The search request may be obtained only according to information such as text included in the search request itself, or may be obtained according to search scenario information associated with the search request, which is not limited in the present application.
Regarding the feature processing, different types of features have different feature processing methods, for example, the text features may be embedded to obtain the Embedding vector features of the text, and other types of feature processing are not listed here.
The infrastructure of the multi-classification model may select Neural network models such as CNN (Convolutional Neural Networks), DNN (Deep Neural Networks), and so on. The training of the multi-classification model can be realized by referring to the prior art, model parameters are optimized through iterative training, and only the processing of training samples is briefly introduced here:
in a search scenario, each candidate search result item has its own background category (the background category is a concept of a service operator, and at present, the background category to which a certain item belongs is determined by a service operator in normal service without additional processing), and then the background category is mapped with a search intention classification system, so that the search result item and the search intention are associated. For a search request query, when a user clicks/purchases a search result item, the search intention of the query associated with the corresponding item is considered to be inclined, so that the click distribution/purchase distribution of the search query in a log can be obtained as the search intention distribution, and a training sample is obtained. Of course, in order to prevent the influence of the deviation caused by the mis-click, a corresponding mis-click prevention threshold may be set.
In an embodiment of the present application, the method further includes: and under the condition that the search intention in the first search intention set hits the cold start rule, updating the intention strength adjusting parameter of the corresponding search intention according to the hit cold start rule.
The cold start generally refers to the first start of the application within a preset time period, for example, the first start after a plurality of days of non-start or the first start after installation. The embodiment of the application can also be expanded to a scene that a business party generates new requirements.
The search intention is a refinement of the user's mental search requirements by the business party, so the refinement can change with the change of business scenes and the development of the user's requirements. For example, in a period of time, the concern of the user on food safety is greatly improved, and then the business party adds a business category "safe store", and the search intention is also "safe store".
It is easy to understand that a search request, if it originally corresponds to the search intention of "food", also has a high probability of corresponding to the search intention of "safe store". However, because the search intention is a newly added search intention, a good recognition effect may not be obtained through the binary model in the absence of a training sample, and intervention may be performed through a cold start rule at this time.
The cold start rule can also be expressed as a cold start word list, and the cold start processing mainly comprises two links of data format design and online validation judgment of the cold start word list.
The data format may include several key fields as follows: search request (Query), city, expiration time, search intent information. The intention information may include a specific search intention (or its search intention identification ID) and a corresponding intention strength level.
The vocabulary format can be designed into a JSON format, such as: { "key _ word: five-way mouth _0", "valid _ time": 2020-04-25"," intent "{" id ": 4", "level": strongg "}. This example representation of an entry in the vocabulary means: when the Query is searched for "five mouths" before 2020-04-25, a strong intention with a search intention ID of 4 is found nationally (key _ word is represented in the format: Query _ cityid, where cityid equal to 0 means nationwide validation). And when the time for requesting the Query is before the expiration time, searching the Query to obtain the corresponding search intention and the strength level thereof. If the search intention does not exist in the first search intention set, the search intention is added to the search intention set, and if the search intention exists in the first search intention set, the original strong and weak levels of the search intention are replaced according to the strong and weak levels of the search intention obtained in the cold-start word list.
By adding the cold start rule, the query operated by a service party can be ensured to generate corresponding search intentions and the intensity levels of the search intentions within a period of time, and the intention intensity final attributes of the search intentions in the intersection are determined according to the intention intensity adjustment parameters and the intention intensity values of the search intentions in the intersection, so that the search results corresponding to the search intentions can be recalled, different degrees of exposure can be obtained according to the intensity levels, and the culture of user cognition is satisfied.
FIG. 2 shows a schematic flow diagram of another search intention identification method according to an embodiment of the present application. As shown in fig. 2, after receiving a search request of a user, normalizing the search request, obtaining a first search intention set through a plurality of separate classification models and a cold start rule, obtaining a second search intention set through a unified multi-classification model, processing an intersection of the first search intention set and the second search intention set according to fusion logic, obtaining a search intention identification result, and outputting the search intention identification result.
In a specific example, there are 5 search intentions, which are hotel, travel, gourmet, income and take-out, respectively, and the corresponding search intentions ID are 0, 1, 2, 3, 4, and query is "five crossing".
(1) A plurality of separate two classifications: and respectively matching the query with the first vocabulary corresponding to each search intention, and if the query is matched with the first vocabulary, outputting the corresponding search intention and the intention strength grade. And if the search intentions are not matched, looking through a second vocabulary whether the search intentions are matched or not, if the search intentions are not matched, requesting a BERT model of the corresponding search intentions, if the prediction scores of the BERT model exceed positive and negative thresholds of the search intentions, outputting the search intentions and determining the strength and weakness level of the search intentions according to the strength and weakness level thresholds of the intentions, and assuming that intention identification results output by a hotel binary classification model, a gourmet binary classification model and a comprehensive binary classification model are positive and are weak intentions, and intention identification results output by a tourism binary classification model and a takeout binary classification model are negative, the final first search intention set is [0, 2, 3], and the strength and weakness coarse attributes of the intentions of each search intention are weak levels.
(2) Cold start: suppose that this period of time needs to run the query of "five mouths", for example, the query needs to correspond to strong intentions of hotels and weak intentions of takeaway. Then the first search intention set is adjusted by combining the recognition results of the plurality of separate second classifications, and then [0, 2, 3, 4] is obtained, and the intention strength and weakness of each search intention are respectively: a manually intervened strong level, a weak level, and a manually intervened weak level.
(3) Unifying the multi-classification models: the query is input to the unified multi-classification model after feature processing, and the unified multi-classification model identifies an intention distribution, for example, the intention distribution output by the unified multi-classification model is [ <0,0.3>, <1,0.1>, <2,0.3>, <3,0.2>, <4,0.1> ], i.e., the second search intention set is [0, 1, 2, 3, 4 ].
(4) Fusion logic: the searching intentions needing to be recalled are obtained from (2) as [ hotel, gourmet, income, take-out ], the corresponding intentions and weaknesses are obtained as [0.3,0.3,0.2,0.1], and because the hotel intentions are operated by strong intentions in the cold starting stage, the intentions and weaknesses need to be corrected as [1,0.3,0.2,0.1] (the adjustment rule is that the intentions and weaknesses of the manual intervention strong level are 1, while the intentions and weaknesses of the manual intervention weak level are not changed, of course, other embodiments can be adjusted in other ways), and finally, the intentions and weaknesses are normalized as [0.625,0.1875,0.125,0.0625], namely, the intentions and weaknesses of each searching intention are obtained.
(5) The search intention recognition result with the final query being "five road junctions" is output as [ hotel (0.625), gourmet (0.1875), income (0.125), take out (0.0625) ]. When the search results are displayed, the search results under the corresponding search intentions can be displayed according to hotel, gourmet, arrival and takeout modes, and adjustment can also be performed.
In a specific embodiment, the elements in the first search intention set may also be intention strength pairs such as < search intention ID, intention strength level >, and the elements in the second search intention set may also be intention strength pairs such as < search intention ID, intention strength score >, and then intersecting may be performed according to the search intention IDs.
Fig. 3 shows a schematic structural diagram of a search intention recognition apparatus according to an embodiment of the present application. As shown, the search intention recognition apparatus 300 includes:
a first identification unit 310, configured to identify a first set of search intents of the search request based on a classification model.
A second identifying unit 320, configured to identify a second search intention set of the search request based on the multi-classification model, where search intents in the second search intention set have intention strength values.
Here, the intention strength values may be compared with each other for intention strength, such as the intention strength scores exemplified above. That is, the intentions may be compared between the search intentions in the second set of search intentions. In contrast, the search intents in the first set of search intents do not have a basis for comparison with each other.
And the fusion unit 330 is configured to find an intersection of the first search intention set and the second search intention set, use the search intention in the intersection as a search intention corresponding to the search request, and determine a search intention display mode according to an intention strength value of the search intention in the intersection.
The search intention can be directly displayed in a search result page as guide information, and the ranking of the search intention can be indirectly reflected through the ranking of the search results.
For example, the search intention in the second search intention set is A, B, C, D, E, and the intention strength values are 0.3, 0.15, 0.25, 0.2 and 0.1 in turn. The search intention in the first search intention is A, B, E, F, the search intentions in the intersection are A, B and E, and the search intention presentation order may be a → B → E since the intention strength value of a is 0.3, the intention strength value of B is 0.15, and the intention strength value of E is 0.1.
It can be seen that the apparatus shown in fig. 3 integrates the advantages of the single two-category and the unified multi-category, guarantees the recall rate through the search intents determined by the single two-category, guarantees the comparability of the strong and weak relationships through the search intents determined by the unified multi-category, and meets the display interaction requirements of the service; the search intention identification and the search intention strength ordering are decoupled, so that the iterative optimization of the respective search intention identification of the services corresponding to the search intents is facilitated, the intention strength ordering is uniformly performed on the platform side, and the problems of service conflict and exclusive exhibition can be better solved.
In an embodiment of the present application, in the above apparatus, the first identifying unit 310 is configured to identify the search requests after the normalization processing based on a plurality of two classification models, respectively, where each two classification models corresponds to a preset search intention; acquiring the recognition result of each binary model; and determining a first search intention set according to the recognition result of each classification model.
In an embodiment of the present application, in the above apparatus, the first identifying unit 310 is configured to match the search request after the normalization processing with the first vocabulary; and under the condition that a complete matching item exists in the first word list, taking a preset search intention corresponding to the binary classification model as the identified search intention.
In an embodiment of the present application, in the above apparatus, the first identifying unit 310 is configured to match the search request after the normalization processing with the second vocabulary; and under the condition that the partial matching item exists in the second word list, taking the preset search intention corresponding to the binary classification model as the identified search intention.
In an embodiment of the present application, in the above apparatus, the first identifying unit 310 is configured to identify a predicted score of the search request after the normalization processing; and under the condition that the prediction score is larger than a positive threshold and a negative threshold, taking a preset search intention corresponding to the binary classification model as the identified search intention.
In an embodiment of the present application, in the above apparatus, the second identifying unit 320 is configured to generate a search intention characteristic of the search request; and identifying the search intention characteristics by using a multi-classification model to obtain a second search intention set.
In an embodiment of the present application, in the above apparatus, the fusing unit 330 is configured to generate a search result cluster according to the search intention in the intersection, and determine a presentation priority of the search cluster according to an intention strength value of each search intention in the search cluster.
In an embodiment of the present application, in the above apparatus, the fusion unit 330 is configured to determine a presentation order of the search intentions according to intention strength values of the search intentions in the intersection.
In one embodiment of the application, in the above apparatus, the search intention in the first search intention set has an intention strength adjustment parameter; and the fusion unit 330 is configured to determine an intention adjustment value according to the intention strength adjustment parameter and the intention strength value of the search intention in the intersection, and determine a search intention display mode according to the intention strength adjustment value of the search intention in the intersection.
In one embodiment of the present application, the apparatus further includes: and the cold start adjusting unit is used for updating the intention intensity adjusting parameters of the corresponding search intentions according to the hit cold start rule under the condition that the search intentions in the first search intention set hit the cold start rule.
It should be noted that, for the specific implementation of each apparatus embodiment, reference may be made to the specific implementation of the corresponding method embodiment, which is not described herein again.
In summary, according to the technical scheme of the application, a first search intention set of a search request is identified based on a binary classification model; identifying a second search intention set of the search request based on the multi-classification model, wherein the search intents in the second search intention set have intention strength values; and solving an intersection of the first search intention set and the second search intention set, taking the search intention in the intersection as the search intention corresponding to the search request, and determining a search intention display mode according to the intention strength value of the search intention in the intersection. The technical scheme integrates the advantages of single two-classification and unified multi-classification, the recall rate is ensured through the search intents determined by a plurality of single two-classification, the comparability of the strength relationship is ensured through the search intents determined by unified multi-classification, and the display interaction requirements of the service are met; the search intention identification and the search intention strength ordering are decoupled, so that the iterative optimization of the respective search intention identification of the services corresponding to the search intents is facilitated, the intention strength ordering is uniformly performed on the platform side, and the problems of service conflict and exclusive exhibition can be better solved.
It should be noted that:
the algorithms and displays presented herein are not inherently related to any particular computer, virtual machine, or other apparatus. Various general purpose devices may be used with the teachings herein. The required structure for constructing such a device will be apparent from the description above. In addition, this application is not directed to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the present application as described herein, and any descriptions of specific languages are provided above to disclose the best modes of the present application.
In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the application, various features of the application are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the application and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be interpreted as reflecting an intention that: this application is intended to cover such departures from the present disclosure as come within known or customary practice in the art to which this invention pertains. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this application.
Those skilled in the art will appreciate that the modules in the device in an embodiment may be adaptively changed and disposed in one or more devices different from the embodiment. The modules or units or components of the embodiments may be combined into one module or unit or component, and furthermore they may be divided into a plurality of sub-modules or sub-units or sub-components. All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or elements of any method or apparatus so disclosed, may be combined in any combination, except combinations where at least some of such features and/or processes or elements are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.
Furthermore, those skilled in the art will appreciate that while some embodiments described herein include some features included in other embodiments, rather than other features, combinations of features of different embodiments are meant to be within the scope of the application and form different embodiments. For example, in the following claims, any of the claimed embodiments may be used in any combination.
The various component embodiments of the present application may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It will be appreciated by those skilled in the art that a microprocessor or Digital Signal Processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components of the search intention identification apparatus according to embodiments of the present application. The present application may also be embodied as apparatus or device programs (e.g., computer programs and computer program products) for performing a portion or all of the methods described herein. Such programs implementing the present application may be stored on a computer readable medium or may be in the form of one or more signals. Such a signal may be downloaded from an internet website or provided on a carrier signal or in any other form.
For example, fig. 4 shows a schematic structural diagram of an electronic device according to an embodiment of the present application. The electronic device 400 comprises a processor 410 and a memory 420 arranged to store computer executable instructions (computer readable program code). The memory 420 may be an electronic memory such as a flash memory, an EEPROM (electrically erasable programmable read only memory), an EPROM, a hard disk, or a ROM. The memory 420 has a storage space 430 storing computer readable program code 431 for performing any of the method steps described above. For example, the storage space 430 for storing the computer readable program code may include respective computer readable program codes 431 for respectively implementing various steps in the above method. The computer readable program code 431 can be read from or written to one or more computer program products. These computer program products comprise a program code carrier such as a hard disk, a Compact Disc (CD), a memory card or a floppy disk. Such a computer program product is typically a computer readable storage medium such as described in fig. 5. FIG. 5 shows a schematic diagram of a computer-readable storage medium according to an embodiment of the present application. The computer readable storage medium 500 stores computer readable program code 431 for performing the steps of the method according to the present application, which is readable by the processor 410 of the electronic device 400, which computer readable program code 431, when executed by the electronic device 400, causes the electronic device 400 to perform the steps of the method described above, in particular the computer readable program code 431 stored by the computer readable storage medium may perform the method shown in any of the embodiments described above. The computer readable program code 431 may be compressed in a suitable form.
It should be noted that the above-mentioned embodiments illustrate rather than limit the application, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The application may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The usage of the words first, second and third, etcetera do not indicate any ordering. These words may be interpreted as names.
Claims (13)
1. A search intention recognition method, characterized by comprising:
identifying a first set of search intents for the search request based on a binary model;
identifying a second set of search intentions of the search request based on a multi-classification model, the search intentions in the second set of search intentions having intent strength values;
and solving an intersection of the first search intention set and the second search intention set, taking the search intention in the intersection as the search intention corresponding to the search request, and determining a search intention display mode according to the intention strength value of the search intention in the intersection.
2. The method of claim 1, wherein the identifying a first set of search intents for a search request based on a classification model comprises:
respectively identifying the search requests after normalization processing based on a plurality of two classification models, wherein each two classification model corresponds to a preset search intention;
acquiring the recognition result of each binary model;
and determining the first search intention set according to the recognition result of each classification model.
3. The method of claim 2, wherein the identifying the search request after the normalization process comprises:
matching the search request after normalization processing with a first word list;
and under the condition that a complete matching item exists in the first word list, taking a preset search intention corresponding to the binary classification model as the identified search intention.
4. The method of claim 2, wherein the identifying the search request after the normalization process comprises:
matching the search request after normalization processing with a second word list;
and under the condition that the partial matching item exists in the second word list, taking the preset search intention corresponding to the binary classification model as the identified search intention.
5. The method of claim 2, wherein the identifying the search request after the normalization process comprises:
identifying a predicted score of the search request after normalization processing;
and under the condition that the prediction score is larger than a positive threshold value and a negative threshold value, taking a preset search intention corresponding to the binary classification model as the identified search intention.
6. The method of claim 1, wherein the identifying a second set of search intents for a search request based on a multi-classification model comprises:
generating a search intent feature of the search request;
and identifying the search intention characteristics by utilizing the multi-classification model to obtain the second search intention set.
7. The method of claim 1, wherein the determining the search intention presentation manner according to the intention strength value of the search intention in the intersection comprises:
and generating a search result block according to the search intention in the intersection, and determining the display priority of the search block according to the intention strength value of each search intention in the search block.
8. The method of claim 1, wherein the determining the search intention presentation manner according to the intention strength value of the search intention in the intersection comprises:
and determining the display sequence of the search intention according to the intention strength value of the search intention in the intersection.
9. The method of any one of claims 1-8, wherein search intents in the first set of search intents have an intent strength adjustment parameter;
the method for determining the search intention display mode according to the intention strength value of the search intention in the intersection comprises the following steps: and determining an intention adjusting value according to the intention strength adjusting parameter and the intention strength value of the search intention in the intersection, and determining a search intention display mode according to the intention strength adjusting value of the search intention in the intersection.
10. The method of claim 9, wherein the method further comprises:
and under the condition that the search intention in the first search intention set hits a cold start rule, updating the intention strength adjustment parameter of the corresponding search intention according to the hit cold start rule.
11. A search intention recognition apparatus, characterized by comprising:
a first identification unit, configured to identify a first search intention set of the search request based on a classification model;
a second identification unit, configured to identify a second search intention set of the search request based on a multi-classification model, where search intents in the second search intention set have intention strength values;
and the fusion unit is used for solving an intersection of the first search intention set and the second search intention set, taking the search intention in the intersection as the search intention corresponding to the search request, and determining a search intention display mode according to the intention strength value of the search intention in the intersection.
12. An electronic device, comprising: a processor; and a memory arranged to store computer-executable instructions that, when executed, cause the processor to perform the method of any one of claims 1-10.
13. A computer readable storage medium, characterized in that the computer readable storage medium stores one or more programs which, when executed by a processor, implement the method of any of claims 1-10.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010401669.7A CN111737544A (en) | 2020-05-13 | 2020-05-13 | Search intention recognition method and device, electronic equipment and storage medium |
| PCT/CN2021/090257 WO2021227869A1 (en) | 2020-05-13 | 2021-04-27 | Search intent identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010401669.7A CN111737544A (en) | 2020-05-13 | 2020-05-13 | Search intention recognition method and device, electronic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111737544A true CN111737544A (en) | 2020-10-02 |
Family
ID=72647149
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010401669.7A Pending CN111737544A (en) | 2020-05-13 | 2020-05-13 | Search intention recognition method and device, electronic equipment and storage medium |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111737544A (en) |
| WO (1) | WO2021227869A1 (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112541079A (en) * | 2020-12-10 | 2021-03-23 | 杭州远传新业科技有限公司 | Multi-intention recognition method, device, equipment and medium |
| CN112784156A (en) * | 2021-01-13 | 2021-05-11 | 携程旅游信息技术（上海）有限公司 | Search feedback method, system, device and storage medium based on intention recognition |
| WO2021227869A1 (en) * | 2020-05-13 | 2021-11-18 | 北京三快在线科技有限公司 | Search intent identification |
| CN114647756A (en) * | 2022-03-11 | 2022-06-21 | 北京百度网讯科技有限公司 | Image-based searching method and device, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105070288A (en) * | 2015-07-02 | 2015-11-18 | 百度在线网络技术（北京）有限公司 | Vehicle-mounted voice instruction recognition method and device |
| US20180089197A1 (en) * | 2016-09-29 | 2018-03-29 | International Business Machines Corporation | Internet search result intention |
| CN109241255A (en) * | 2018-08-20 | 2019-01-18 | 华中师范大学 | A kind of intension recognizing method based on deep learning |
| CN110555717A (en) * | 2019-07-29 | 2019-12-10 | 华南理工大学 | method for mining potential purchased goods and categories of users based on user behavior characteristics |
| CN110781675A (en) * | 2019-09-25 | 2020-02-11 | 苏宁云计算有限公司 | Text classification method and device |
| CN110807485A (en) * | 2019-10-31 | 2020-02-18 | 电子科技大学 | A method for fusion of two-class semantic segmentation maps into multi-class semantic maps based on high-scoring remote sensing images |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9754037B2 (en) * | 2014-08-27 | 2017-09-05 | Facebook, Inc. | Blending by query classification on online social networks |
| CN109522556B (en) * | 2018-11-16 | 2024-03-12 | 北京九狐时代智能科技有限公司 | Intention recognition method and device |
| CN110162535B (en) * | 2019-03-26 | 2023-11-07 | 腾讯科技（深圳）有限公司 | Search method, apparatus, device and storage medium for performing personalization |
| CN110110795B (en) * | 2019-05-10 | 2021-04-20 | 厦门美图之家科技有限公司 | Image classification method and device |
| CN110162633B (en) * | 2019-05-21 | 2022-02-11 | 深圳市珍爱云信息技术有限公司 | Voice data intention determining method and device, computer equipment and storage medium |
| CN111737544A (en) * | 2020-05-13 | 2020-10-02 | 北京三快在线科技有限公司 | Search intention recognition method and device, electronic equipment and storage medium |
-
2020
- 2020-05-13 CN CN202010401669.7A patent/CN111737544A/en active Pending
-
2021
- 2021-04-27 WO PCT/CN2021/090257 patent/WO2021227869A1/en active Application Filing
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105070288A (en) * | 2015-07-02 | 2015-11-18 | 百度在线网络技术（北京）有限公司 | Vehicle-mounted voice instruction recognition method and device |
| US20180089197A1 (en) * | 2016-09-29 | 2018-03-29 | International Business Machines Corporation | Internet search result intention |
| CN109241255A (en) * | 2018-08-20 | 2019-01-18 | 华中师范大学 | A kind of intension recognizing method based on deep learning |
| CN110555717A (en) * | 2019-07-29 | 2019-12-10 | 华南理工大学 | method for mining potential purchased goods and categories of users based on user behavior characteristics |
| CN110781675A (en) * | 2019-09-25 | 2020-02-11 | 苏宁云计算有限公司 | Text classification method and device |
| CN110807485A (en) * | 2019-10-31 | 2020-02-18 | 电子科技大学 | A method for fusion of two-class semantic segmentation maps into multi-class semantic maps based on high-scoring remote sensing images |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021227869A1 (en) * | 2020-05-13 | 2021-11-18 | 北京三快在线科技有限公司 | Search intent identification |
| CN112541079A (en) * | 2020-12-10 | 2021-03-23 | 杭州远传新业科技有限公司 | Multi-intention recognition method, device, equipment and medium |
| CN112784156A (en) * | 2021-01-13 | 2021-05-11 | 携程旅游信息技术（上海）有限公司 | Search feedback method, system, device and storage medium based on intention recognition |
| CN114647756A (en) * | 2022-03-11 | 2022-06-21 | 北京百度网讯科技有限公司 | Image-based searching method and device, electronic equipment and storage medium |
| CN114647756B (en) * | 2022-03-11 | 2024-08-13 | 北京百度网讯科技有限公司 | Image-based searching method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021227869A1 (en) | 2021-11-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111737544A (en) | Search intention recognition method and device, electronic equipment and storage medium | |
| CN107851097B (en) | Data analysis system, data analysis method, data analysis program and storage medium | |
| US20160260033A1 (en) | Systems and Methods for Similarity and Context Measures for Trademark and Service Mark Analysis and Repository Searchess | |
| US11392645B2 (en) | Computerized domain expert | |
| US11263277B1 (en) | Modifying computerized searches through the generation and use of semantic graph data models | |
| CN112256845B (en) | Intention recognition method, apparatus, electronic device, and computer-readable storage medium | |
| CN107257970A (en) | The problem of being carried out from structuring and unstructured data sources answer | |
| CN111832290A (en) | Model training method and device for determining text relevancy, electronic equipment and readable storage medium | |
| CN112131876B (en) | A method and system for determining standard questions based on similarity | |
| CN109582788A (en) | Comment spam training, recognition methods, device, equipment and readable storage medium storing program for executing | |
| CN112528031A (en) | Work order intelligent distribution method and system | |
| Liu et al. | Open intent discovery through unsupervised semantic clustering and dependency parsing | |
| CN111782676A (en) | Probability prediction model training method, probability prediction method and probability prediction device | |
| CN116975267A (en) | Information processing method and device, computer equipment, medium and product | |
| KR102663632B1 (en) | Device and method for artwork trend data prediction using artificial intelligence | |
| CN113704623A (en) | Data recommendation method, device, equipment and storage medium | |
| WO2018230616A1 (en) | Legal information processing system, method, and program | |
| Bhanuse et al. | Optimal e-learning course recommendation with sentiment analysis using hybrid similarity framework | |
| CN114661951B (en) | Video processing method, device, computer equipment and storage medium | |
| CN118468061B (en) | Automatic algorithm matching and parameter optimizing method and system | |
| CN112487154B (en) | Intelligent search method based on natural language | |
| CN113821612A (en) | Information search method and device | |
| CN118520854A (en) | Text generation method, apparatus, computer device, storage medium, and program product | |
| WO2015117657A1 (en) | A query expansion system and method using language and language variants | |
| CN117009621A (en) | Information searching method, device, electronic equipment, storage medium and program product |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20201002 |