CN106649818A - Recognition method and device for application search intentions and application search method and server - Google Patents

Abstract

The invention discloses an application search intent identification method, device, application search method and server. The method includes: obtaining search words in each query session from the query session log of an application search engine; According to the label system of each search term, the application search intent corresponding to the search term is identified. In this solution, the user intent recognition method tag method matching the app application tag system is proposed, which flexibly expresses the user's fine-grained query intent. Based on unsupervised machine learning technology, a user intent labeling system is built, which abandons the traditional user intent classification method, and implements a set of automated user intent mining processes, which can generate a user intent tag list with high accuracy and recall rate, and combine user intent and Apps are mapped to the same tag system, so that when users search for apps, they can quickly and accurately obtain apps that meet their intentions.

Description

Application search intent recognition method, device, application search method, and server

technical field

The present invention relates to the field of data mining, in particular to an application search intent identification method, device, application search method and server.

Background technique

App search engine is a mobile software application search engine service that provides app search and download on mobile phones, such as 360 Mobile Assistant, Tencent App Store, Google Play, Appstore, etc. Application search engines are mobile search services installed on mobile phones, such as 360 mobile assistant app applications. Due to the limitations of objective conditions such as the small display plane of search results, the best user experience can only be obtained by providing accurate search results. One of the important differences in PC-side web search. The number of mobile app applications is huge, with millions of app applications. The application search engine needs to understand the user's query intention in order to accurately display the app application in the user's mind.

The premise of using a search engine to provide accurate search services is to accurately understand the user's query intention. There is a potential search intent hidden behind each query request of the user. If the application search engine can perceive the user's needs and map the search word text to the corresponding app application function or app application category, the app application that is more in line with the user's intention The results are ranked first, which obviously enhances the user's search experience. Therefore, user intent recognition is the core technology of the application search engine, and it is also the key to realize the functional search technology.

In the existing traditional web search engine technology, user search intentions are manually sorted and classified, and user search intentions are divided into three types: navigation, information, and resource. Applicable to app application scenarios. Because each app application has a fixed application field and provides people with a certain specific function, it is appropriate to use tags to mine users' fine-grained functional requirements, and classification-based methods are not applicable due to their wide granularity and breadth. Therefore, there is still no very flexible and effective method to meet the growing demand of users for fast and precise search of apps.

Contents of the invention

In view of the above problems, the present invention is proposed to provide an application search intent identification method, device, application search method and server that overcome the above problems or at least partially solve the above problems.

According to one aspect of the present invention, a method for identifying application search intent is provided, the method comprising:

Obtain the search terms in each query session from the query session log of the application search engine;

According to the search terms and preset strategies in each query session, the label system of each search term is excavated;

According to the tag system of each search term, the application search intent corresponding to the search term is identified.

Optionally, according to the search terms in each query session and the preset strategy, the tag system for mining each search term includes:

Obtain a training corpus set according to the search terms in each query session;

Input the training corpus set into the LDA model for training, and obtain the search word-topic probability distribution result and the topic-keyword probability distribution result output by the LDA model;

According to the search word-topic probability distribution result and the topic-keyword probability distribution result, the label system of each search word is calculated.

Optionally, said obtaining the training corpus set according to the search terms in each query session includes:

According to the search terms in each query session, the original corpus of each search term is obtained;

The original corpus of each search word constitutes an original corpus set; the original corpus set is preprocessed to obtain a training corpus set.

Optionally, said obtaining the original corpus of each search word according to the search words in each query session includes:

Obtain a set of search word sequences corresponding to multiple query sessions according to the search words in each query session; and obtain a set of search words corresponding to multiple query sessions;

The search term sequence set is trained to obtain an N-dimensional search term vector file;

For each search word in the search word set, calculate the degree of association between the search word and other search words according to the N-dimensional search word vector file; the degree of association with the search word meets the preset conditions The other search terms of the search term are used as the original corpus of the search term.

Optionally, the obtaining the set of search word sequences corresponding to multiple query sessions includes:

For each query session, arrange the search terms in the query session in order into a sequence; if a search term in the sequence corresponds to an application download operation, insert the name of the downloaded application into the corresponding search term in the sequence The back adjacent position of word; Obtain the search word sequence corresponding to this query session;

The obtaining the search term sets corresponding to the multiple query sessions includes: using the search term sets in the multiple query sessions as the search term sets corresponding to the multiple query sessions.

Optionally, training the set of search word sequences to obtain an N-dimensional search word vector file includes:

Each search word in the search word sequence set is taken as a word, and the deep learning toolkit word2vec is used to train the search word sequence set to generate an N-dimensional search word vector file.

Optionally, for each search term in the set of search terms, the degree of association between the search term and other search terms is calculated according to the N-dimensional search term vector file; the association with the search term The other search terms that meet the preset conditions are used as the original corpus of the search term, including:

Using the KNN algorithm to perform operations on the search word set and the N-dimensional search word vector file, and calculate the distance between every two search words in the search word set according to the N-dimensional search word vector file;

For each search word in the set of search words, the search words are sorted according to the distance from the search word in descending order, and the first preset threshold search words are selected as the original corpus of the search word.

Optionally, the preprocessing of the original corpus includes:

In the original corpus,

For each original corpus, word segmentation is carried out to the original corpus to obtain a word segmentation result containing a plurality of words; search for a phrase formed by adjacent words in the word segmentation result; retain the phrase, the word segmentation result The lexical items belonging to nouns and the lexical items belonging to verbs are the corresponding reserved keywords of the original corpus.

Optionally, the search for a phrase formed of adjacent terms in the word segmentation result includes:

The cPMId value of every two adjacent terms in the word segmentation result is calculated, and when the cPMId value of the two adjacent terms is greater than a second preset threshold, it is determined that the two adjacent terms form a phrase.

Optionally, the preprocessing of the original corpus further includes:

The original material of each search word corresponds to the reserved keywords as the first stage training corpus of the search word;

The first-stage training corpus of each search term constitutes a first-stage training corpus set; data cleaning is performed on keywords in the first-stage training corpus set.

Optionally, the data cleaning of keywords in the first-stage training corpus includes:

In the first stage training corpus collection,

For the first-stage training corpus of each search word, calculate the TF-IDF value of each keyword in the first-stage training corpus; Make the TF-IDF value higher than the third preset threshold and/or lower than the first The keywords of four preset thresholds are deleted to obtain the training corpus of the search words;

The training corpus of each search term constitutes a training corpus set.

Optionally, according to the search word-topic probability distribution result and the topic-keyword probability distribution result, calculating the label system of each search word includes:

According to the search word-topic probability distribution result and the topic-keyword probability distribution result, calculate the search word-keyword probability distribution result;

According to the search word-keyword probability distribution result, for each search word, the keywords are sorted according to the probability of the search word in descending order, and the keywords with the fifth preset threshold number are selected.

Optionally, the calculation of the search word-keyword probability distribution result according to the search word-topic probability distribution result and the topic-keyword probability distribution result includes:

For each search word, obtain the probability of each topic about the search word according to the search word-topic probability distribution result;

For each topic, obtain the probability of each keyword about the topic according to the topic-keyword probability distribution result;

Then for each keyword, the product of the probability of the keyword about a topic and the probability of the topic about a search word is used as the probability of the keyword based on the topic about the search word; the keyword is based on each topic The sum of the probabilities with respect to the search term is used as the probability of the keyword with respect to the search term.

Optionally, according to the search word-topic probability distribution result and the topic-keyword probability distribution result, calculating the label system of each search word further includes:

Use the keywords corresponding to the selected fifth preset threshold number for each search term as the first-stage tag system for the search term;

For the first-stage tag system of each search word, calculate the semantic relationship value between each keyword in the first-stage tag system of the search word and the search word; for each keyword, the corresponding keyword The product of the semantic relationship value of the keyword and the probability of the keyword with respect to the search term is used as the modified probability of the keyword with respect to the search term; each keyword in the first-stage label system of the search term is adjusted according to the modified probability of the search term The probabilities are sorted from large to small, and the first sixth preset threshold keywords are selected to form the label system of the search term.

Optionally, calculating the semantic relationship value between each keyword in the tag system of the first stage of the search term and the search term includes:

According to the search words in each query session, obtain the search word sequence set corresponding to a plurality of query sessions; train the search word sequence set to obtain an N-dimensional keyword vector file;

Calculate the word vector of the keyword according to the keyword vector file of the N dimension, and calculate the word vector of each term in the search term;

Calculate the cosine similarity between the word vector of the keyword and the word vector of each term as the semantic relationship value between the keyword and the corresponding term;

The sum of the semantic relationship values between the keyword and each term is used as the semantic relationship value between the keyword and the search term.

Optionally, said training the set of search word sequences to obtain an N-dimensional keyword vector file includes:

Carry out word segmentation processing on the set of search word sequences, use the deep learning toolkit word2vec to train the set of search word sequences after the word segmentation processing, and generate N-dimensional keyword vector files.

Optionally, according to the search word-topic probability distribution result and the topic-keyword probability distribution result, calculating the label system of each search word further includes:

The first six preset threshold keywords corresponding to each search term are used as the second stage label system of the search term;

For the second-stage tag system of each search word, count the TF-IDF value of each keyword in the second-stage tag system of the search word in the training corpus of the search word; for each keyword, the The product of the probability of the keyword with respect to the search term and the TF-IDF value is used as the secondary revised probability of the keyword with respect to the search term; The secondary correction probability of words is sorted from large to small, and the first K keywords are selected to form the label system of the search word.

Optionally, the tag system for selecting the first K keywords to form the search term includes:

Obtain the number of queries about the search term within a preset time period from the query session log of the application search engine;

According to the number of queries, the first K keywords are selected to form the label system of the search word; where the value of K is a broken line function of the number of queries corresponding to the search word.

According to another aspect of the present invention, an application search method is provided, the method comprising:

Build a search term label database, which includes a label system of multiple search terms;

Receive the current search term uploaded by the client, and obtain the tag system of the current search term according to the search term tag database;

Calculate the degree of association between the label system of the current search term and the label system of each application;

When the degree of association between the label system of the current search word and the label system of an application meets the preset conditions, the relevant information of the application is returned to the client for display;

The search word label database is constructed by the method described in any one of the first aspects of the present invention.

Optionally, said obtaining the label system of the current search term according to the search term label database includes:

Calculate the semantic similarity between the current search term and each search term in the search term label database, sort according to the semantic similarity from large to small, and select the first preset threshold search term;

According to the selected tag system of each search word, the tag system of the current search word is obtained.

Optionally, the calculating the semantic similarity between the current search term and each search term in the search term label database includes: calculating the Euclidean similarity between the current search term and each search term in the search term label database Distance, using the Euclidean distance between each search term and the current search term as the semantic similarity corresponding to the search term;

According to the selected tag system of each search term, obtaining the tag system of the current search term includes: the semantic similarity corresponding to each search term is used as the weight of each tag in the tag system of the search term; For each tag corresponding to the tag system, add the weights of the same tags to get the final weight of each tag; sort the final weights from large to small, and select the first second preset threshold tags to form the tag system of the current search word.

According to another aspect of the present invention, an application search intent recognition device is provided, the device comprising:

An acquisition unit adapted to acquire the search words in each query session from the query session log of the application search engine;

The mining unit is adapted to mine the label system of each search term according to the search term and the preset strategy in each query session;

The identification unit is adapted to identify the application search intent corresponding to the search term according to the tag system of each search term.

Optionally, the mining unit is adapted to obtain a training corpus set according to the search terms in each query session; input the training corpus set into the LDA model for training, and obtain the search term-topic probability distribution results output by the LDA model and Topic-keyword probability distribution results; according to the search word-topic probability distribution results and the topic-keyword probability distribution results, the label system of each search word is calculated.

Optionally, the mining unit is adapted to obtain the original corpus of each search word according to the search words in each query session; the original corpus of each search word constitutes an original corpus set; preprocess the original corpus set to obtain A collection of training corpora.

Optionally, the mining unit is adapted to obtain a set of search word sequences corresponding to multiple query sessions according to the search words in each query session; and obtain a set of search words corresponding to multiple query sessions; The sequence set is trained to obtain an N-dimensional search term vector file; for each search term in the search term collection, calculate the degree of association between the search term and other search terms according to the N-dimensional search term vector file; Other search words whose degree of association with the search word meets the preset conditions are used as the original corpus of the search word.

Optionally, the mining unit is adapted to, for each query session, arrange the search words in the query session into a sequence; if a search word in the sequence corresponds to an application download operation, the downloaded inserting the name of the application into the adjacent position behind the corresponding search term in the sequence; obtaining the search term sequence corresponding to the query session; using a set of search terms in multiple query sessions as the search term corresponding to the multiple query sessions gather.

Optionally, the mining unit is adapted to use each search word in the set of search word sequences as a word, use the deep learning toolkit word2vec to train the set of search word sequences, and generate N-dimensional search words vector file.

Optionally, the mining unit is adapted to use the KNN algorithm to operate on the search word set and the N-dimensional search word vector file, and calculate the search word set in the search word set according to the N-dimensional search word vector file. The distance between every two search words; for each search word in the search word set, sort according to the distance from the search word from large to small, and select the first preset threshold search words as the search word The original corpus of words.

Optionally, the mining unit is adapted to perform word segmentation processing on each original corpus in the original corpus set to obtain a word segmentation result containing a plurality of lexical items; Phrases formed by adjacent terms in ; keep the phrases, the terms belonging to nouns and the terms belonging to verbs in the word segmentation results, as the corresponding reserved keywords of the original corpus.

Optionally, the mining unit is adapted to calculate the cPMId value of each two adjacent term items in the word segmentation result, and when the cPMId value of the two adjacent term items is greater than a second preset threshold, it is determined that the two adjacent term items Adjacent terms form phrases.

Optionally, the mining unit is also adapted to use the keywords retained corresponding to the original material of each search word as the first-stage training corpus of the search word; the first-stage training corpus of each search word constitutes the first-stage training A corpus collection; performing data cleaning on keywords in the first-stage training corpus collection.

Optionally, the mining unit is adapted to calculate the TF of each keyword in the first-stage training corpus for the first-stage training corpus of each search word in the first-stage training corpus set -IDF value; delete keywords whose TF-IDF value is higher than the third preset threshold and/or lower than the fourth preset threshold to obtain the training corpus of the search term; the training corpus of each search term constitutes a training corpus set.

Optionally, the mining unit is adapted to calculate a search word-keyword probability distribution result according to the search word-topic probability distribution result and the topic-keyword probability distribution result; As a result of the word probability distribution, for each search word, the keywords are sorted according to the probability of the search word in descending order, and the keywords with the fifth preset threshold number are selected.

Optionally, the mining unit is adapted to, for each search word, obtain the probability of each topic with respect to the search word according to the search word-topic probability distribution result; for each topic, according to the topic-keyword probability The distribution results get the probability of each keyword on the topic; then for each keyword, the product of the probability of the keyword on a topic and the probability of the topic on a search word is used as the keyword's probability on the topic based on the topic The probability of the search term: the sum of the probabilities of the keyword with respect to the search term based on each topic is used as the probability of the keyword with respect to the search term.

Optionally, the mining unit is also adapted to use the keywords corresponding to the first fifth preset threshold number selected for each search term as the first stage tag system of the search term; for the first stage of each search term Tag system, calculate the semantic relationship value between each keyword in the first stage tag system of the search word and the search word; for each keyword, the semantic relationship value corresponding to the keyword is related to the keyword The product of the probability of the search term is used as the corrected probability of the keyword with respect to the search term; the keywords in the first-stage label system of the search term are sorted from large to small according to the corrected probability of the search term. The sixth preset threshold keywords constitute the tag system of the search term.

Optionally, the mining unit is adapted to obtain a set of search word sequences corresponding to multiple query sessions according to the search words in each query session; train the set of search word sequences to obtain an N-dimensional keyword vector file; Calculate the word vector of the keyword according to the keyword vector file of the N dimension, calculate the word vector of each term in the search term; calculate the difference between the word vector of the keyword and the word vector of each term The cosine similarity of the keyword is used as the semantic relationship value between the keyword and the corresponding term; the sum of the semantic relationship values between the keyword and each term is used as the semantic relationship value between the keyword and the search term.

Optionally, the mining unit is adapted to perform word segmentation processing on the set of search word sequences, use the deep learning toolkit word2vec to train the set of search word sequences after word segmentation, and generate an N-dimensional keyword vector file.

Optionally, the mining unit is also adapted to use the first sixth preset threshold keywords corresponding to each search term as the second-stage label system of the search term; for the second-stage label of each search term System, count the TF-IDF value of each keyword in the second stage label system of the search word in the training corpus of the search word; for each keyword, the probability of the keyword about the search word is compared with the The product of the above TF-IDF values is used as the secondary correction probability of the keyword with respect to the search term; each keyword in the second-stage tag system of the search term is ranked from large to small according to the secondary correction probability of the search term Sorting, select the first K keywords to form the label system of the search term.

Optionally, the mining unit is adapted to obtain the number of queries about the search term within a preset time period from the query session log of the application search engine; select the first K keywords according to the query times to form the search term label system; where the K value is a broken line function of the number of queries corresponding to the search term.

According to still another aspect of the present invention, an application search server is provided, and the server includes:

The database construction unit is suitable for constructing a search word tag database, and the search word tag database includes a tag system of multiple search words;

an interactive unit adapted to receive the current search term uploaded by the client;

The search processing unit is adapted to obtain the label system of the current search term according to the search term label database; calculate the degree of association between the label system of the current search term and the label system of each application;

The interaction unit is further adapted to return relevant information of the application to the client for display when the degree of association between the label system of the current search term and the label system of an application meets a preset condition;

The process of the database construction unit constructing the search term tag database is the same as that described in any one of claims 22-39 by the device for identifying application search intentions.

Optionally, the search processing unit is adapted to calculate the semantic similarity between the current search term and each search term in the search term tag database, sort the semantic similarity from large to small, and select the first pre-selected Set a threshold number of search words; obtain the label system of the current search word according to the label system of each selected search word.

Optionally, the search processing unit is adapted to calculate the Euclidean distance between the current search term and each search term in the search term tag database, and use the Euclidean distance between each search term and the current search term as the The semantic similarity corresponding to the search word; the semantic similarity corresponding to each search word is used as the weight of each tag in the tag system of the search word; for each tag corresponding to the tag system of each search word, the weight of the same tag is Add to get the final weight of each tag; sort according to the final weight from large to small, and select the first second preset threshold tags to form the tag system of the current search term.

According to the solution of the present invention, a user intent recognition method matching the app application label system - the label method is proposed, which is flexible and effective, accurately excavates the label system corresponding to the search word, and establishes a search word label database. Words can be used to accurately describe search terms with a tag system, thus solving the problem of user intent recognition. Furthermore, user intent and app applications can be mapped to the same label system, and more accurate application search results can be obtained by performing search matching. This not only solves the problem of user intention recognition, but also solves the related calculation problems of the application search engine, and lays the foundation for the application of a core technology in the search engine-functional search technology.

The above description is only an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention, it can be implemented according to the contents of the description, and in order to make the above and other purposes, features and advantages of the present invention more obvious and understandable , the specific embodiments of the present invention are enumerated below.

Description of 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 embodiment. The drawings are only for the purpose of illustrating a preferred embodiment and are not to be considered as limiting the invention. Also throughout the drawings, the same reference numerals are used to designate the same components. In the attached picture:

FIG. 1 shows a flowchart of a method for identifying application search intent according to an embodiment of the present invention;

Fig. 2 shows a flow chart of an application search method according to an embodiment of the present invention;

Fig. 3 shows a schematic diagram of an application search intent recognition device according to an embodiment of the present invention; and

Fig. 4 shows a schematic diagram of an application search server according to an embodiment of the present invention.

detailed description

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided for more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

In the following, app represents an application, query represents a search term, tag represents a tag, and Session represents a query session.

The present invention proposes a new user intent recognition method for application search engines, the label method, which flexibly and effectively expresses the user's fine-grained query intent, builds a user intent tag system based on unsupervised machine learning technology, and abandons the traditional user intent The intent classification method implements a set of automated user intent mining processes, which can generate a user intent tag list with high accuracy and recall rate, map user query and app to a shared tag system, and solve the problem of user intent identification and application The correlation calculation problem of the search engine has achieved very good results.

Fig. 1 shows a flow chart of a method for identifying application search intent according to an embodiment of the present invention. As shown in Figure 1, the method includes:

Step S110, obtaining the search words in each query session from the query session log of the application search engine;

Step S120, according to the search words in each query session and the preset strategy, dig out the label system of each search word;

Step S130, according to the tag system of each search term, identify the application search intent corresponding to the search term.

It can be seen that the traditional user intent recognition is a classification method for web design, which is not suitable for app application scenarios. Each application has a fixed application field, providing people with a specific function, and using tags to mine user fine-grained The functional requirements are appropriate, and the classification-based approach is too granular and broad to be applicable. This solution proposes a user intent recognition method that matches the application tag system - the tag method, which is flexible and effective, and maps user intent and applications to the same tag system, which not only solves the problem of user intent recognition, but also solves the problem of application search engines. The problem of correlation calculation is the basis of realizing a core technology of application search engine - function search technology.

Usually, the user's search term is a short text, and the search term constructed by the user according to the demand in his mind has sparse features and cannot fully describe the demand itself. However, if a user is only looking for an app with a single functional scenario in a short period of time, they often rewrite the query search terms around a single requirement, and these query words usually have a strong semantic relationship between them. An important feature of search engines.

In the search engine service, the system will automatically record the information related to the user's search and save it in the query log. For example, when a user opens a Baidu search page, enters the search words such as "game", "game software", "fun game", "game application download" in sequence, and then enters the search page, or after entering a certain search page, continues to enter a certain These search words are searched until the user completes the search event and closes the entire Baidu search page. The whole process is called a query session.

In one embodiment of the present invention, in step S120, according to the search terms in each query session and the preset strategy, digging out the label system of each search term includes: obtaining a training corpus set according to the search terms in each query session; The training corpus is input into the LDA model for training, and the search word-topic probability distribution result and the topic-keyword probability distribution result of the LDA model output are obtained; according to the search word-topic probability distribution result and the topic-keyword probability The distribution results are calculated to obtain the label system of each search term.

In the process of obtaining the training corpus, the technical difficulty is the expansion of the short text of the query. After expanding the long text, a query can be regarded as a document, which is the key to our effective use of the LDA topic model, resulting in high accuracy and high recall. tag. Intent tags are divided into categorical tags and functional tags. The categorical tags reflect the application fields of user needs, and the functional tags reflect the specific needs of users.

Wherein, according to the search terms in each query session, obtaining the training corpus set includes:

According to the search words in each query session, the original corpus of each search word is obtained; the original corpus of each search word forms an original corpus set; the original corpus set is preprocessed to obtain a training corpus set. Specifically, the obtaining the original corpus of each search word according to the search words in each query session includes: obtaining a set of search word sequences corresponding to multiple query sessions according to the search words in each query session; and obtaining multiple query A collection of search terms corresponding to a session.

Keep a query search word sequence inside a query session, and treat the search words as a whole. If a user downloads some apps under a certain query, put the app name next to the query sequence. For example: a user session sequence is query1, query2, query3, the user downloads an app1 after entering query2, and spells app1 after query2 and before query3, namely query1, query2, app1, query3. Each session sequence is a line, which is output to the file session_query-app_list.txt, which is a collection of search word sequences. And output all queries to another file query_all.txt, which is the set of search terms.

The search word sequence set is trained to obtain an N-dimensional search word vector file; for each search word in the search word set, calculate the relationship between the search word and other search words according to the N-dimensional search word vector file. The degree of association among them; the other search words whose association degree with the search term meets the preset conditions are used as the original corpus of the search term.

In one embodiment of the present invention, the obtaining the set of search word sequences corresponding to multiple query sessions includes: for each query session, arranging the search words in the query session into a sequence; A search word corresponds to an application download operation, inserting the name of the downloaded application into the adjacent position behind the corresponding search word in the sequence; obtaining the search word sequence corresponding to the query session; The set of search terms includes: using a set of search terms in multiple query sessions as a set of search terms corresponding to the multiple query sessions.

For example, a user sequentially inputs "search term 1", "search term 2", and "search term 3" in a query session, and the user downloads an app1 after inputting "search term 2". Therefore, the search word sequence corresponding to the query session is: search word 1, search word 2, app1, search word 3. The search word sequence corresponding to each query session is one line, and the search word sequence corresponding to multiple query sessions is set into multiple lines.

The above-mentioned training of the search word sequence set to obtain the N-dimensional search word vector file includes: using each search word in the search word sequence set as a word, and using the deep learning toolkit word2vec to process the search word sequence set Perform training to generate N-dimensional search word vector files. For example, use the deep learning toolkit word2vec training to generate a 300-dimensional query vector, and generate a query vector file query_w2v_300.dict, which is the search word vector file.

In fact, when users search for the desired application, they input various search words, either as a noun (such as: "game"), or as a phrase (such as "fun game"), or as a sentence ( Such as: "I want to download a fun game.").

In one embodiment of the present invention, the search term vector file obtained above is used as the basis for calculating the word vector for each search term in the search term set, and for each search term in the search term set word, according to the N-dimensional search word vector file to calculate the degree of association between the search word and other search words; the degree of association with the search word meets other search words that meet the preset conditions as the search word The original corpus specifically includes:

Using the KNN algorithm to perform operations on the search word set and the N-dimensional search word vector file, and calculate the distance between every two search words in the search word set according to the N-dimensional search word vector file; For each search word in the set of search words, the search words are sorted according to the distance from the search word in descending order, and the first preset threshold search words are selected as the original corpus of the search word.

Table 1 shows the top 10 nearest neighbors whose search term is "Sogou" in one embodiment of the present invention. The nearest neighbors include the search term and app application name, as shown in the first column of Table 1, " Sogou Mobile Phone Input Method", "Sogou Input Method", etc. 10 is the first preset value in this example, and the second column in Table 1 indicates the distance between the corresponding nearest neighbor and the search word "Sogou".

Table 1

nearest neighbor Statistical indicators (based on Euclidean distance) Sogou mobile phone input method 38 303.827 0.838104 sogou Input 26 323.494 0.845153 Sogou 20332 372.525 0.778589 Take the dog 6986 385.809 0.76965 Sogou Pinyin 14577 410.986 0.753037 Sogou Input Method Xiaomi Version 4042 423.929 0.746941 Sogou Pinyin Input Method 4927 435.273 0.736172 Sohu input method 18233 452.955 0.724872 Sogou input 10274 455.505 0.720034 Mobile phone Sogou input method 3075 476.93 0.721099

Table 2 shows the top 10 nearest neighbors whose search term is "lottery lottery drawing query" in an embodiment of the present invention, and the corresponding representative meanings are similar to those in Table 1, and will not be repeated here.

Table 2

In one embodiment of the present invention, after obtaining the original corpus set corresponding to each search term, the preprocessing of the original corpus set includes:

In the original corpus collection, for each original corpus, the original corpus is subjected to word segmentation processing to obtain a word segmentation result containing a plurality of lexical items; search for phrases formed by adjacent lexical items in the word segmentation results; retain The phrase, the lexical items belonging to nouns and the lexical items belonging to verbs in the word segmentation result are used as keywords correspondingly reserved for the original corpus.

For example, if a user enters a search term of "download game", then the term of the search term that belongs to a noun is "game", and the term that belongs to a verb is "download".

Wherein, the search for phrases formed by adjacent terms in the word segmentation results includes:

The cPMId value of every two adjacent terms in the word segmentation result is calculated, and when the cPMId value of the two adjacent terms is greater than a second preset threshold, it is determined that the two adjacent terms form a phrase.

Equation 1 shows the cPMId calculation method, where d(x, y) represents the co-occurrence frequency of two terms x, y, d(x) represents the frequency of occurrence of term x, and d(y) represents the term y The frequency of occurrence of , D represents the total number of apps, δ=0.7.

Formula 1

For example, according to the reverse order of the cPMId value, select the combination of terms whose cPMId is higher than the threshold value 5 as a phrase, merge with the verbs and nouns just reserved, and generate a new file query_corpus_seg_nouns_verb_phrase.txt, which is the training corpus after the first preprocessing.

Further, in one embodiment of the present invention, the preprocessing of the original corpus further includes: using the keywords retained corresponding to the original material of each search word as the first-stage training corpus of the search word; The first-stage training corpus of each search term constitutes a first-stage training corpus set; data cleaning is performed on keywords in the first-stage training corpus set.

Specifically, the data cleaning of the keywords in the first-stage training corpus includes: in the first-stage training corpus, for each search word in the first-stage training corpus, calculating the first The TF-IDF value of each keyword in the one-stage training corpus; delete the keywords whose TF-IDF value is higher than the third preset threshold and/or lower than the fourth preset threshold to obtain the training corpus of the search term ; The training corpus of each search term constitutes a training corpus set.

This step is to mine the non-tag words in the first-stage training corpus for data cleaning. A high-frequency or low-frequency item is less likely to be a tag. In the first stage of the training corpus, the tf-idf statistical method is used to calculate the tf-idf weight of each term and phrase, which will be higher than a certain threshold. Items or phrases with a value or lower than a certain threshold are regarded as non-tag words. This threshold is related to specific corpus. The specific value is not listed here. Non-tag words generate a blacklist black_tag.list, and filter out the first file Phase non-tag words in the training corpus to generate a new training corpus in the format: search word_id\tterm1term2...termn.

Table 3 shows some non-label words that can be discarded in data cleaning. These words appear too frequently or appear too infrequently, which are meaningless for users to search.

table 3

After obtaining the training corpus, the LDA model uses the GibbsLDA++ version. It is necessary to modify the GibbsLDA++ source code to initialize the topic of the same term in the query corpus as the same. In the original code, each term is randomly initialized to a topic, resulting in the same repeated term being initialized to multiple topics, because in the vertical application field, the possibility of ambiguity of terms is small, so the same term is initialized Forming the same theme is in line with the status quo in the vertical application field, and can also improve the effect of the LDA model. For example, LDA training selects 120 topics, iterates for 300 rounds, and outputs two sets of data, which are topic-term probability distribution and document-topic probability distribution.

Then this solution needs to calculate the label system of each search word according to the search word-topic probability distribution result and the topic-keyword probability distribution result, including:

According to the search word-topic probability distribution result and the topic-keyword probability distribution result, calculate the search word-keyword probability distribution result; according to the search word-keyword probability distribution result, for each search word, The keywords are sorted according to the probability of the search word in descending order, and the keywords with the fifth preset threshold number are selected.

Wherein, the calculation of the search word-keyword probability distribution result according to the search word-topic probability distribution result and the topic-keyword probability distribution result includes: for each search word, according to the search word-topic The probability distribution result obtains the probability of each topic about the search word; for each topic, obtains the probability of each keyword about the topic according to the topic-keyword probability distribution result; then for each keyword, the keyword about the The product of the probability of a topic and the probability of the topic with respect to a search word is used as the probability of the keyword based on the topic with respect to the search word; the sum of the probability of the keyword based on each topic with respect to the search word is used as the keyword The probability of the word with respect to the search term.

This step is the process of initial LDA tag generation, and this step gets the tag generated by LDA. The output of LDA is the topic probability distribution under each query, and the term probability distribution under each topic. In order to get the tag of each query, we sort the topic probability distribution and term probability distribution in reverse order according to the probability from large to small. We select the top 50 topics under each query, and select the top 120 terms under each topic. The probability of the term is weighted and sorted by the probability of the topic. Each tag term has an lda weight, which indicates the importance of the query. According to the reverse order of the tag weight, the tag list generated by LDA is obtained. Less noise, the order of tags is not accurate.

Further, it is necessary to fine-tune the prediction results of the LDA model so that the order of the important tags of each query is higher. In one embodiment of the present invention, the search term-topic probability distribution results and the The topic-keyword probability distribution results, the calculation of the label system of each search term also includes: using the keywords corresponding to the first fifth preset threshold number selected for each search term as the first stage label system of the search term; The first-stage tag system of a search word, calculate the semantic relationship value between each keyword in the first-stage tag system of the search word and the search word; for each keyword, the corresponding semantic value of the keyword The product of the relationship value and the probability of the keyword with respect to the search term is used as the modified probability of the keyword with respect to the search term; each keyword in the first-stage label system of the search term is changed from Sort from large to small, and select the first sixth preset threshold keywords to form the label system of the search term.

Wherein, calculating the semantic relationship value between each keyword in the tag system of the first stage of the search term and the search term includes: according to the search terms in each query session, obtaining a set of search term sequences corresponding to multiple query sessions ; The search word sequence set is trained to obtain an N-dimensional keyword vector file; according to the N-dimensional keyword vector file, calculate the word vector of the keyword, and calculate the word of each term in the search word Vector; calculate the cosine similarity between the word vector of the keyword and the word vector of each term, as the semantic relationship value of the keyword and the corresponding term; the relationship between the keyword and the semantic relationship value of each term and as the semantic relationship value between the keyword and the search term.

For example, to calculate the semantic relationship between each tag word and query, this uses the trained word vector term_w2v_300.dict, the method is: separately calculate the co-xuan similarity between the tag word vector and the word vectors of each word in the query, and add them together. The larger the value, the more important the tag is, and it is re-sorted in reverse order after being weighted with the lda weight.

Specifically, the training of the set of search word sequences to obtain an N-dimensional keyword vector file includes: performing word segmentation processing on the set of search word sequences, and using the deep learning toolkit word2vec to segment the set of search word sequences after word segmentation. Perform training to generate N-dimensional keyword vector files.

For example, perform Chinese word segmentation on the set of search word sequences above, use the deep learning toolkit word2vec to train, generate a 300-dimensional query vector, and generate another word vector file term_w2v_300.dict, which is a keyword vector file.

Still further, in an embodiment of the present invention, the calculating and obtaining the tag system of each search word based on the search word-topic probability distribution result and the topic-keyword probability distribution result further includes: The first sixth preset threshold keywords selected corresponding to the search term are used as the second-stage label system of the search term; for the second-stage label system of each search term, each The TF-IDF value of a keyword in the training corpus of the search term; for each keyword, the product of the probability of the keyword with respect to the search term and the TF-IDF value is used as the keyword with respect to the search term The secondary correction probability of the search term; the keywords in the second-stage label system of the search term are sorted from large to small according to the secondary correction probability of the search term, and the first K keywords are selected to form the label system of the search term .

For example, the tf-idf weights in the query extended corpus are properly weighted according to the tags, and the weights are normalized to rearrange the tag order.

After the correction of the above two methods, the accuracy of the tag order expressing the query intention is greatly improved

In one embodiment of the present invention, said selecting the first K keywords to form the label system of the search term includes: obtaining the number of queries about the search term within a preset time period from the query session log of the application search engine; According to the number of queries, the first K keywords are selected to form the label system of the search word; where the value of K is a broken line function of the number of queries corresponding to the search word.

This step is to determine the number of tags for each query, and keep the top k tag words. This k value is used as a broken line function of the number of query searches. We keep 2 to 5 tags for each query, and the accuracy rate is 88%. , with a recall rate of 75%. In this step we generate a query intent dictionary query_intent_tag.txt.

Furthermore, in a specific example, this solution marks about 2.6 million queries with tag words that express user intentions, and considers the query as a whole. When the user synonymously reconstructs and rewrites the query, the new query is not in our In the query intent dictionary, at this time, it is necessary to calculate the semantic similarity between the new query and the query in the dictionary, and assign the intent tag of the semantically similar query to the new query. The calculation method is: add up the amount of terms of each word in the new query as a new query vector, calculate the Euclidean distance with the query vector of the query intent dictionary, select the first 3 nearest neighbor queries, and use KdTree to reduce the computational complexity; The K-distance is smoothed with a Gaussian kernel and used as the weighted weight of the tag words. The intent tag words of the three neighboring queries are integrated to generate the intent tag words of the new query. Keeping the first three tags can satisfy the user's search intent, and the accuracy rate reaches 80%.

Fig. 2 shows a flow chart of an application search method according to an embodiment of the present invention, the method includes:

Step 210, constructing a search word tag database, which includes a plurality of search word tag systems.

Step 220, receiving the current search word uploaded by the client, and obtaining the tag system of the current search word according to the search word tag database.

Step 230, calculating the degree of association between the tag system of the current search word and the tag system of each application.

Step 240, when the degree of association between the tag system of the current search word and the tag system of an application meets the preset condition, return the relevant information of the application to the client for display.

Wherein, in step S210, during the process of constructing the search word tag database, the mining of the search word tag system is the same as the mining process of the search word tag system shown in any embodiment of the method shown in FIG. 1 .

In an embodiment of the present invention, said obtaining the tag system of the current search term according to the search term tag database includes: calculating the semantic similarity between the current search term and each search term in the search term tag database, Sorting from large to small according to the semantic similarity, selecting the first search term with a preset threshold; according to the label system of each selected search term, the label system of the current search term is obtained.

In an embodiment of the present invention, the calculating the semantic similarity between the current search term and each search term in the search term label database includes: calculating the current search term and each search term in the search term label database The Euclidean distance between words, the Euclidean distance between each search term and the current search term is used as the semantic similarity corresponding to the search term; the label system of the selected search terms is used to obtain the label of the current search term The system includes: the semantic similarity corresponding to each search word is used as the weight of each tag in the tag system of the search word; for each tag corresponding to the tag system of each search word, the weights of the same tags are added to obtain each tag The final weight of the final weight; according to the final weight sorted from large to small, select the first second preset threshold tags to form the tag system of the current search term.

Table 4 shows the intent tag words of some search words searched by 360 Mobile Assistant.

Table 4

Fig. 3 shows a device for identifying an application search intent according to an embodiment of the present invention, the device 300 for identifying an application search intent includes:

The obtaining unit 310 is adapted to obtain the search words in each query session from the query session log of the application search engine;

The mining unit 320 is adapted to mine the tag system of each search word according to the search words in each query session and the preset strategy;

The identification unit 330 is adapted to identify the application search intent corresponding to the search term according to the tag system of each search term.

In one embodiment of the present invention, the mining unit 320 is adapted to obtain a training corpus set according to the search words in each query session; input the training corpus set into the LDA model for training, and obtain the search words output by the LDA model - Topic probability distribution results and topic-keyword probability distribution results; according to the search word-topic probability distribution results and the topic-keyword probability distribution results, the label system of each search word is calculated.

Wherein, in one embodiment of the present invention, the mining unit 320 is adapted to obtain the original corpus of each search term according to the search term in each query session; the original corpus of each search term constitutes an original corpus set; The original corpus is preprocessed to obtain the training corpus.

Specifically, in one embodiment of the present invention, the mining unit 320 is adapted to obtain search word sequence sets corresponding to multiple query sessions according to the search words in each query session; A set of search words; the search word sequence set is trained to obtain an N-dimensional search word vector file; for each search word in the search word set, the search word and other words are calculated according to the N-dimensional search word vector file The degree of association between each search word; use other search words whose degree of association with the search word meets the preset conditions as the original corpus of the search word.

That is to say, the mining unit 320 is adapted to, for each query session, arranging the search words in the query session into a sequence in order; if a search word in the sequence corresponds to an application download operation, the downloaded The name of the application is inserted into the adjacent position behind the corresponding search term in the sequence; the search term sequence corresponding to the query session is obtained; the set of search terms in multiple query sessions is used as the search term corresponding to the multiple query sessions word collection.

For example, the mining unit 320 is adapted to use each search word in the search word sequence set as a word, and use the deep learning toolkit word2vec to train the search word sequence set to generate an N-dimensional search word vector document.

On this basis, in one embodiment of the present invention, the mining unit 320 is adapted to use the KNN algorithm to perform operations on the search word set and the N-dimensional search word vector file, and according to the N-dimensional The search word vector file calculates the distance between every two search words in the search word set; The first preset threshold search words are used as the original corpus of the search words.

In the process of preprocessing, in one embodiment of the present invention, the mining unit 320 is adapted to perform word segmentation processing on the original corpus for each original corpus in the original corpus set to obtain multiple The word segmentation result of the word item; find the phrase formed by the adjacent words in the word segmentation result; keep the phrase, the word item belonging to the noun and the word item belonging to the verb in the word segmentation result, as the corresponding reservation of the original corpus keywords.

Specifically, the mining unit 320 is adapted to calculate the cPMId value of every two adjacent terms in the word segmentation result, and when the cPMId values of the two adjacent terms are greater than a second preset threshold, it is determined that the two adjacent terms Adjacent terms form phrases.

Further, in one embodiment of the present invention, the mining unit 320 is also adapted to use the keywords retained corresponding to the original material of each search word as the first-stage training corpus of the search word; the first-stage training corpus of each search word The first-stage training corpus constitutes a first-stage training corpus set; data cleaning is performed on keywords in the first-stage training corpus set.

Specifically, in one embodiment of the present invention, the mining unit 320 is adapted to calculate the first-stage training corpus for each search term in the first-stage training corpus set. The TF-IDF value of each keyword in the corpus; the TF-IDF value is higher than the third preset threshold value and/or the keyword lower than the fourth preset threshold value is deleted to obtain the training corpus of the search term; each search The training corpus of words constitutes the training corpus set.

In an embodiment of the present invention, the mining unit 320 is adapted to calculate the search word-keyword probability distribution result according to the search word-topic probability distribution result and the topic-keyword probability distribution result; In the search word-keyword probability distribution result, for each search word, the keywords are sorted according to the probability of the search word in descending order, and the keywords with the fifth preset threshold number are selected.

In one embodiment of the present invention, the mining unit 320 is adapted to, for each search word, obtain the probability of each topic related to the search word according to the search word-topic probability distribution result; for each topic, according to the According to the topic-keyword probability distribution results, the probability of each keyword on the topic is obtained; then for each keyword, the product of the probability of the keyword on a topic and the probability of the topic on a search word is used as the keyword based on The probability of the topic with respect to the search word; the sum of the probabilities of the keyword with respect to the search word based on each topic is used as the probability of the keyword with respect to the search word.

Further, in one embodiment of the present invention, the mining unit 320 is also adapted to use the first fifth preset threshold number of keywords corresponding to each search term as the first-stage tag system of the search term; For the first-stage tag system of each search word, calculate the semantic relationship value between each keyword in the first-stage tag system of the search word and the search word; for each keyword, the corresponding keyword The product of the semantic relationship value of the keyword and the probability of the keyword with respect to the search term is used as the modified probability of the keyword with respect to the search term; each keyword in the first-stage label system of the search term is adjusted according to the modified probability of the search term The probabilities are sorted from large to small, and the first sixth preset threshold keywords are selected to form the label system of the search term.

In one embodiment of the present invention, the mining unit 320 is adapted to obtain search word sequence sets corresponding to multiple query sessions according to the search words in each query session; train the search word sequence sets to obtain N-dimensional The keyword vector file; According to the keyword vector file of the N dimension, calculate the word vector of this keyword, calculate the word vector of each term in this search word; Calculate the word vector of this keyword and each word The cosine similarity between the word vectors of the term is used as the semantic relationship value between the keyword and the corresponding term; the sum of the semantic relationship values between the keyword and each term is used as the semantic relationship between the keyword and the search term relationship value.

In one embodiment of the present invention, the mining unit 320 is adapted to perform word segmentation processing on the set of search word sequences, use the deep learning toolkit word2vec to train the set of search word sequences after word segmentation, and generate N-dimensional Keyword vector file.

Further, in one embodiment of the present invention, the mining unit 320 is also adapted to use the first sixth preset threshold keywords corresponding to each search word as the second stage label system of the search word; for The second-stage label system of each search term, count the TF-IDF value of each keyword in the second-stage label system of the search term in the training corpus of the search term; for each keyword, the key The product of the probability of the word about the search word and the TF-IDF value is used as the secondary revised probability of the keyword about the search word; The secondary correction probability of is sorted from large to small, and the first K keywords are selected to form the label system of the search term.

In one embodiment of the present invention, the mining unit 320 is adapted to obtain the number of queries about the search word within a preset time period from the query session log of the application search engine; select the top K according to the number of queries Keywords constitute the label system of the search term; where the K value is a broken line function of the number of queries corresponding to the search term.

Fig. 4 shows an application search server according to an embodiment of the present invention, the application search server 400 includes:

The database construction unit 410 is adapted to construct a search word tag database, which includes a tag system of multiple search words;

The interaction unit 420 is adapted to receive the current search word uploaded by the client;

The search processing unit 430 is adapted to obtain the tag system of the current search term according to the search term tag database; calculate the degree of association between the tag system of the current search word and the tag systems of each application;

The interaction unit 420 is further adapted to return relevant information of the application to the client for display when the degree of association between the label system of the current search word and the label system of an application meets the preset condition;

Wherein, the scheme of mining the tag system of the search word in the process of the database construction unit 410 constructing the search word tag database is the same as that of the search word identification device 300 according to any one of the above-mentioned embodiments of the present invention. The scheme of the labeling system is the same.

In one embodiment of the present invention, the search processing unit 430 is adapted to calculate the semantic similarity between the current search term and each search term in the search term label database, and sort the terms according to the semantic similarity from large to small , select the first preset threshold search words before; obtain the tag system of the current search word according to the tag system of each selected search word.

In one aspect of the present invention, the search processing unit 430 is adapted to calculate the Euclidean distance between the current search term and each search term in the search term label database, and calculate the distance between each search term and the current search term The Euclidean distance is used as the semantic similarity corresponding to the search term; the semantic similarity corresponding to each search term is used as the weight of each label in the label system of the search term; for each label corresponding to the label system of each search term, the same Add the weights of the tags to get the final weight of each tag; sort the final weights from large to small, and select the first second preset threshold tags to form the tag system of the current search term.

It should be noted that the embodiments of the apparatus shown in FIGS. 3-4 are correspondingly the same as the embodiments of the method shown in FIGS. 1-2 , which have been described in detail above and will not be repeated here.

To sum up, the application search intent recognition method, device, application search method, and server in the present invention propose a user intent recognition method that matches the app application tag system—the labeling method, which flexibly expresses the user's fine-grained query intent . Based on unsupervised machine learning technology, the user intent labeling system is built, and the traditional user intent classification method is abandoned, and a set of automatic user intent mining process is realized, which can generate a user intent tag list with high accuracy and recall rate. User intentions and app applications can be mapped to the same label system, thus solving the problem of user intention identification and the calculation of application search engines, which is a core technology in application search engines - functional search technology laid the foundation.

It should be noted:

The algorithms and displays presented herein are not inherently related to any particular computer, virtual appliance, or other device. Various general purpose devices can also be used with the teachings based on this. The structure required to construct such an apparatus will be apparent from the foregoing description. Furthermore, the present invention is not specific to any particular programming language. It should be understood that various programming languages can be used to implement the content of the present invention described herein, and the above description of specific languages is for disclosing the best mode of the present invention.

In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the invention 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 the understanding of this description.

Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, in order to streamline this disclosure and to facilitate an understanding of one or more of the various inventive aspects, various features of the invention are sometimes grouped together in a single embodiment, figure, or its description. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. 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 invention.

Those skilled in the art can understand that the modules in the device in the embodiment can be adaptively changed and arranged in one or more devices different from the embodiment. Modules or units or components in the embodiments may be combined into one module or unit or component, and furthermore may be divided into a plurality of sub-modules or sub-units or sub-assemblies. All features disclosed in this specification (including accompanying claims, abstract and drawings) and any method or method so disclosed may be used in any combination, except that at least some of such features and/or processes or units are mutually exclusive. All processes or units of equipment are combined. Each feature disclosed in this specification (including 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 understand that although some embodiments described herein include some features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention. 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 invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) can be used in practice to implement some or all of the application search intent identification device and application search server according to the embodiment of the present invention. Or full functionality. The present invention can also be implemented as an apparatus or an apparatus program (for example, a computer program and a computer program product) for performing a part or all of the methods described herein. Such a program for realizing the present invention 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 site, or provided on a carrier signal, or provided in any other form.

It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, 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 invention can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means can be embodied by one and the same item of hardware. The use of the words first, second, and third, etc. does not indicate any order. These words can be interpreted as names.

Claims (10)

1. A method for identifying application search intent, comprising: Obtain the search terms in each query session from the query session log of the application search engine; According to the search terms and preset strategies in each query session, the label system of each search term is excavated; According to the tag system of each search term, the application search intent corresponding to the search term is identified. 2. The method as claimed in claim 1, wherein, according to the search term and the preset strategy in each query session, digging out the label system of each search term comprises: Obtain a training corpus set according to the search terms in each query session; Input the training corpus set into the LDA model for training, and obtain the search word-topic probability distribution result and the topic-keyword probability distribution result output by the LDA model; According to the search word-topic probability distribution result and the topic-keyword probability distribution result, the label system of each search word is calculated. 3. The method according to claim 1 or 2, wherein, according to the search term in each query session, obtaining the training corpus collection includes: According to the search terms in each query session, the original corpus of each search term is obtained; The original corpus of each search word constitutes an original corpus set; the original corpus set is preprocessed to obtain a training corpus set. 4. The method according to any one of claims 1-3, wherein, according to the search term in each query session, obtaining the original corpus of each search term comprises: Obtain a set of search word sequences corresponding to multiple query sessions according to the search words in each query session; and obtain a set of search words corresponding to multiple query sessions; The search term sequence set is trained to obtain an N-dimensional search term vector file; For each search term in the search term set, calculate the degree of association between the search term and other search terms according to the N-dimensional search term vector file; Other search terms are used as the original corpus of the search term. 5. An application search method, comprising: Build a search term label database, which includes a label system of multiple search terms; Receive the current search term uploaded by the client, and obtain the tag system of the current search term according to the search term tag database; Calculate the degree of association between the label system of the current search term and the label system of each application; When the degree of association between the label system of the current search word and the label system of an application meets the preset conditions, the relevant information of the application is returned to the client for display; The search word tag database is constructed by the method according to any one of claims 1-4. 6. A recognition device for applying a search intent, comprising: An acquisition unit adapted to acquire the search words in each query session from the query session log of the application search engine; The mining unit is adapted to mine the label system of each search term according to the search term and the preset strategy in each query session; The identification unit is adapted to identify the application search intent corresponding to the search term according to the tag system of each search term. 7. The method of claim 6, wherein, The mining unit is adapted to obtain a training corpus set according to the search words in each query session; input the training corpus set into the LDA model for training, and obtain the search word-topic probability distribution results and the topic-keywords output by the LDA model Probability distribution results: According to the search word-topic probability distribution results and the topic-keyword probability distribution results, the label system of each search word is calculated. 8. The method of claim 6 or 7, wherein, The mining unit is adapted to obtain the original corpus of each search word according to the search words in each query session; the original corpus of each search word forms an original corpus set; preprocess the original corpus set to obtain a training corpus set. 9. The method according to any one of claims 6-8, wherein the mining unit is adapted to obtain a set of search word sequences corresponding to a plurality of query sessions according to the search words in each query session; and, obtain A set of search words corresponding to multiple query sessions; training the set of search word sequences to obtain an N-dimensional search word vector file; for each search word in the search word set, according to the N-dimensional search word vector file Calculate the degree of association between the search word and other search words; use other search words whose association degree with the search word meets the preset conditions as the original corpus of the search word. 10. An application search server, comprising: The database construction unit is suitable for constructing a search word tag database, and the search word tag database includes a tag system of multiple search words; an interactive unit adapted to receive the current search term uploaded by the client; The search processing unit is adapted to obtain the label system of the current search term according to the search term label database; calculate the degree of association between the label system of the current search term and the label system of each application; The interaction unit is further adapted to return relevant information of the application to the client for display when the degree of association between the label system of the current search term and the label system of an application meets a preset condition; The database construction unit is the same as the process of constructing the search word tag database by the device for identifying application search intentions according to any one of claims 6-9.