CN108334632B

CN108334632B - Entity recommendation method and device, computer equipment and computer-readable storage medium

Info

Publication number: CN108334632B
Application number: CN201810162116.3A
Authority: CN
Inventors: 李潇; 郑孙聪
Original assignee: Shenzhen Tencent Computer Systems Co Ltd
Current assignee: Shenzhen Tencent Computer Systems Co Ltd
Priority date: 2018-02-26
Filing date: 2018-02-26
Publication date: 2021-03-23
Anticipated expiration: 2038-02-26
Also published as: CN108334632A

Abstract

The application relates to an entity recommendation method, an entity recommendation device, a computer device and a computer readable storage medium, wherein the method comprises the following steps: acquiring a user label corresponding to the user identification which meets the entity recommendation condition; obtaining an associated label associated with the user label based on the label association relation; obtaining an extended user tag corresponding to the user identifier according to the user tag and the associated tag; according to the relation between the user tag and the associated tag, entity recall is carried out to obtain a candidate entity; and screening entities matched with the expanded user tags from the candidate entities to recommend to the corresponding user identification. The user tags are expanded by utilizing the tag association relationship, the potential interest of the user is fully mined, the user is subjected to combined recall through the relationship between the user tags and the expanded tags, and the entity is recommended to the user according to the candidate entity and the expanded user tags, so that the reduction of recall quality caused by tag expansion is avoided, the recommended entity is ensured to meet the potential interest of the user, and the click rate of the recommended entity is effectively improved.

Description

Entity recommendation method and device, computer equipment and computer-readable storage medium

Technical Field

The present application relates to the field of information processing technologies, and in particular, to an entity recommendation method, an entity recommendation apparatus, a computer device, and a computer-readable storage medium.

Background

With the development of internet technology and electronic commerce, personalized entity recommendation based on big data processing is greatly developed and popularized, and personalized information service and decision support are provided for users by mining required entity data from massive data. For example, in news recommendation, news that may be of interest to a user is screened from a large amount of news by using the user's historical interest data.

However, the historical interest data of the user is limited, entity recalling and recommending are performed by the traditional entity recommending method, and entities recalled and recommended are various and difficult to meet the potential interest requirements of the user, so that the recommended entity click rate is low.

Disclosure of Invention

Based on this, it is necessary to provide an entity recommendation method, apparatus, computer device and computer readable storage medium for solving the technical problem that entities recommended by the conventional entity recommendation method are difficult to satisfy the potential interest needs of users.

An entity recommendation method, the method comprising:

acquiring a user label corresponding to the user identification which meets the entity recommendation condition;

obtaining an associated label associated with the user label based on the label association relation;

obtaining an extended user tag corresponding to the user identifier according to the user tag and the associated tag;

according to the relation between the user tag and the associated tag, entity recall is carried out to obtain a candidate entity;

and screening entities matched with the expanded user tags from the candidate entities to recommend to the corresponding user identification.

An entity recommendation apparatus, the apparatus comprising:

the user tag acquisition module is used for acquiring a user tag corresponding to the user identifier which meets the entity recommendation condition;

the associated tag obtaining module is used for obtaining an associated tag associated with the user tag based on the tag association relation;

the user tag expansion module is used for obtaining an expanded user tag corresponding to the user identifier according to the user tag and the associated tag;

the recalling module is used for recalling the entity according to the relation between the user tag and the associated tag to obtain a candidate entity;

and the recommending module is used for screening the entities matched with the extended user tags from the candidate entities and recommending the entities to the corresponding user identification.

A computer device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method of any of the above embodiments.

A computer-readable storage medium, storing a computer program which, when executed by a processor, causes the processor to perform the steps of the method of any of the above embodiments.

According to the entity recommendation method, the entity recommendation device, the computer equipment and the computer readable storage medium, the associated labels associated with the user labels are obtained based on the label association relation, the expanded user labels are obtained according to the user labels and the associated labels, namely, the label association relation is utilized, the number of the labels of the user is expanded, the labels which can be used during entity recommendation are enriched, and the potential interests of the user are fully mined. On the basis, the combined recall is carried out through the relationship between the user tags and the extension tags, and the recall quality reduction caused by tag extension is avoided. Therefore, when the entity recommendation is performed on the user according to the candidate entity and the extended user tag, the recommended entity can be ensured to meet the potential interest of the user, and the click rate of the recommended entity is effectively improved.

Drawings

FIG. 1 is a diagram of an application environment of a method for entity recommendation in one embodiment;

FIG. 2 is a flow diagram that illustrates a method for entity recommendation in one embodiment;

FIG. 3 is a flowchart illustrating a method for entity recommendation in another embodiment;

FIG. 4 is a flowchart illustrating the steps of extracting recommended entities in one embodiment;

FIG. 5 is a model diagram of a news recommendation method in one embodiment;

FIG. 6 is a flowchart illustrating a tag association establishment procedure in one embodiment;

FIG. 7 is a flowchart illustrating the steps of the associate tag recommendation in one embodiment;

FIG. 8 is a flowchart illustrating a method for entity recommendation in another embodiment;

FIG. 9 is a block diagram of an entity recommendation device in one embodiment;

FIG. 10 is a block diagram illustrating a tag association relationship establishing module in the entity recommending apparatus according to an embodiment of the present invention;

FIG. 11 is a block diagram of a computer device in one embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

FIG. 1 is a diagram of an application environment of a method for entity recommendation in one embodiment. Referring to fig. 1, the entity recommendation method is applied to an entity recommendation system. The entity recommendation system includes a user terminal 110 and a computer device 120. The user terminal 110 and the computer device 120 are connected via a network. The user terminal 110 may be a desktop terminal or a mobile terminal, and the mobile terminal may be at least one of a mobile phone, a tablet computer, a notebook computer, and the like. The computer device 120 is embodied as a server, and may be implemented as a stand-alone server or a server cluster composed of a plurality of servers.

The user terminal 110 may intelligently interact with the user and receive the user's operation at the user terminal. When it is detected that the condition recommended by the entity is reached according to the operation of the user, the corresponding user identification is transmitted to the computer device 120. The operation of the user at the user terminal may be an input operation or a click operation of text content. In addition, the user terminal 110 may also extract tags from the relevant entities viewed by the user as the user tags of the current user and/or accept user tags that the end user himself types. The process of extracting the tag from the related entity browsed by the user may be performed in the user terminal 110, or may be performed in the computer device 120, and the computer device 120 extracts the tag according to the browsing request triggered by the user terminal 110 and the data sent by the computer device 120 according to the browsing request. Further, the user tag may be stored locally through the user terminal 110, or may be stored in association with the corresponding user identifier through the computer device 120, for example, the user terminal 110 sends the extracted and/or user-entered user tag to the computer device 120 for storage in association with the corresponding user identifier. The computer device 120 is further configured to process the mass tags to obtain a tag association relationship, expand the tags of the user according to the tag association relationship, recommend the entity according to the expanded user tags, send the recommended entity to the user terminal 110, and display the recommended entity by the user terminal 110.

FIG. 2 illustrates an entity recommendation method in one embodiment. The embodiment is mainly illustrated by applying the method to the computer device 120 in fig. 1. Referring to fig. 2, the entity recommendation method specifically includes the following steps:

s201, obtaining a user label corresponding to the user identification which reaches the entity recommendation condition.

The entity refers to objects or things which exist in the real world in a guest manner and can be distinguished from each other, and may be specific human things or abstract concepts and relations. In the internet field, an entity is often referred to as a generic term of something, for example, an entity may specifically be news, video, games, and the like, and further, in a game application, an entity may also be a game character, a publisher, and the like.

User tags refer to tags that can be used to effectively characterize the attribute characteristics of the relevant entities of interest to the user. For example, the user tag may specifically include tags of news of interest to the user, such as department, jodan, NBA, and the like. The user tag may be obtained from data historically viewed by the user or may be set by the user. A corresponding user portrait can be established through the user tags, and relevant recommendation is carried out based on the user portrait. As described above, the user tag may be stored locally through the user terminal, or may be stored in association with the corresponding user identifier through the server.

It should be noted that, one user identifier is correspondingly set for different users, so that different users and user tags corresponding to different users are distinguished through the user identifiers. Specifically, the user identifier may be a user account registered by the user, or may be a corresponding identifier of the terminal, which is not limited herein.

The entity recommendation condition refers to a relevant condition set by an application program for triggering entity recommendation. It should be understood that different entity recommendation conditions may be set according to application scenarios of different application programs, and the entity recommendation conditions are not limited in this embodiment. Specifically, taking a news application as an example, in the application, the object recommended by the entity is news. The entity recommendation condition may be that entity recommendation is performed according to the user tag when the user opens the application. Taking a video playing application program as an example, in the application program, an object recommended by an entity may be video content or advertisement content. The entity recommendation condition can be when the user opens the application program or when the user opens the video content.

Specifically, when the user tag is stored in the server and the user identifier meeting the entity recommendation condition is obtained, the user tag corresponding to the user identifier can be found and obtained based on the corresponding relationship between the user identifier and the user tag. For example, when the server receives trigger information which is sent by the terminal and carries a corresponding user identifier, it is determined that the user identifier meets the entity recommendation condition, the user identifier is obtained by analyzing the trigger information, and then the user identifier corresponding to the user identifier is obtained based on the corresponding relationship between the user identifier and the user identifier.

Taking news recommendation in a news APP as an example, the news APP is in an online state, when a user opens the news APP through a terminal or logs in a user account of the news APP, trigger information carrying corresponding user identification is generated and sent to a server, the server analyzes the user identification from the received trigger information, and a user tag stored in association with the user identification is obtained according to the user identification.

And when the user label is stored in the server, directly receiving the user label corresponding to the user identification which meets the entity recommendation condition. For example, when receiving trigger information which is sent by a terminal and carries a corresponding user identifier, it is determined that the user identifier reaches an entity recommendation condition, and a user tag which is stored in the terminal and corresponds to the user identifier is obtained according to the user identifier.

Still take news recommendation in the news APP as an example, the news APP is in an online state, when a user opens the news APP through a terminal or logs in a user account of the news APP, trigger information carrying corresponding user identification is generated and sent to a server, the server analyzes the user identification from the received trigger information and sends a user label acquisition signal to the corresponding terminal, and the terminal acquires the user label stored in association with the user identification according to the user identification and feeds the user label back to the server.

S203, obtaining an associated label associated with the user label based on the label association relation.

The tag association relation refers to the correlation between all tags and their corresponding associated tags, which specifically indicates whether the correlation exists between the tags and/or the degree of association. For example, a user tag of one user is NBA, and the tag association relationship corresponding to the tag records an associated tag (e.g., department, jodan, etc.) corresponding to the tag and the degree of correlation between the tag and the associated tag.

Specifically, the tag association relationship may be obtained by learning the correlation of a large amount of tag data in the internet data. The mass label data is obtained, the associated labels of the obtained labels are obtained based on a corresponding learning method, and the label association relation is established according to the labels and the corresponding relation between the associated labels.

In this embodiment, the obtained tag association relationship is used, and the tag association relationship is queried according to the user tags, so that the association tags corresponding to the user tags can be obtained respectively.

And S205, obtaining an extended user tag corresponding to the user identifier according to the user tag and the associated tag.

The expanded user tags are tags which are obtained by expanding the tag association relation to the associated tags through the user tags and are used for representing attribute characteristics of related entities which are interested by the corresponding users. In one particular embodiment, the extended user tags include a user tag and an associated tag.

Specifically, the tag of the user is expanded according to the obtained associated tag, so that an expanded user tag comprising the user tag and the associated tag is obtained, and the expanded user tag corresponds to the user identifier corresponding to the user tag. Because the relevance between the associated tag and the corresponding user tag exists to a certain degree, the tag of the user is expanded by utilizing the associated tag, so that the expanded user tag is ensured to meet the historical interest of the user, and the potential interest of the user can be fully mined.

And S207, recalling the entity according to the relation between the user tag and the associated tag to obtain a candidate entity.

The entity recall refers to retrieving related entities meeting retrieval conditions from a mass of entities.

In the embodiment, the effect of expanding the user tags is achieved by associating the tags, and the tags which can be used when the entity recommends are enriched. If the entity recalls directly from the associated tag, the retrieved candidate entities are excessive and contain a lot of noise, which will add extra burden to the subsequent recommendation calculation. In this embodiment, the relationship between the user tag and the associated tag is used as a search condition to perform entity recall, the related entities meeting the search condition are recalled from the massive entities, and the recalled related entities are used as candidate entities, so as to perform entity recommendation according to the candidate entities in the following. The combined recall is carried out through the relationship between the user label and the extension label, so that the recall quality reduction caused by label extension is avoided.

S209, screening entities matched with the expanded user tags from the candidate entities and recommending the entities to the corresponding user identification.

And further screening candidate entities according to the extended user tags to obtain entities meeting the matching conditions, and recommending the entities to the user identifications corresponding to the extended user tags. Specifically, the candidate entities and the extended user tags are matched according to a pre-constructed recommendation model, the recommended entities are obtained according to the matching result, and the recommended entities obtained by matching and screening the candidate entities based on the extended user tags can better meet the potential interest or demand of the user, so that the interest of the user in browsing the recommended entities is promoted, and the click rate of the recommended entities is effectively improved.

According to the entity recommendation method, the associated labels associated with the user labels are obtained based on the label association relation, the expanded user labels are obtained according to the user labels and the associated labels, namely, the label association relation is utilized, the number of the labels of the user is expanded, the labels which can be used during entity recommendation are enriched, and the potential interests of the user are fully mined. On the basis, the combined recall is carried out through the relationship between the user tags and the extension tags, and the recall quality reduction caused by tag extension is avoided. Therefore, when the entity recommendation is performed on the user according to the candidate entity and the extended user tag, the recommended entity can be ensured to meet the potential interest of the user, and the click rate of the recommended entity is effectively improved.

In an embodiment, obtaining an extended user tag corresponding to a user identifier according to the user tag and the associated tag includes: and combining the user label and the associated label to obtain an extended user label corresponding to the user identifier.

In this embodiment, based on a preset merge rule, the user tags and the associated tags corresponding to the user tags are merged to obtain extended user tags corresponding to the user identifiers. The specific merging rule may include merging tags with the same name, sorting and merging tags according to the association degree of associated tags, and the like.

In one embodiment, the tag association relationship includes an association coefficient of the associated tag. Wherein the correlation coefficient is used to represent the degree of correlation between two different tags. Specifically, the correlation coefficient may be represented by a number between 0 and 1, and the larger the number, the greater the degree of correlation between the two tags. The specific range of the correlation coefficient can be set according to the requirement, and is not limited herein.

In a specific embodiment, there is a case where there are multiple user tags for the same user, and different user tags have the same associated tag, or the associated tag is the same as one of the user tags, which results in an extended user tag duplication. For example, the user tags corresponding to one user include user tag a and user tag E. Wherein, the associated tag of the user tag A comprises: label B, label C and label D, the correlation coefficient is respectively: 0.6, 0.8 and 0.6. The associated tag of the user tag E includes: label A, label C and label F, the correlation coefficient is respectively: 0.5, 0.7 and 0.8, if directly merged, the extended user tag corresponding to the user includes: user tag a, tag B (0.6), tag C (0.8), tag D (0.6), user tag E, tag a (0.5), tag C (0.7) and tag F (0.8), that is, an associated tag C comprising one user tag a and one same associated tag a, and two different association coefficients. In order to avoid the problem of repeated recommendation calculation caused by repeated expansion tags, in this embodiment, the associated tags are processed in the merging process, so as to reduce the redundancy of the expansion user tags.

Specifically, merging the user tag and the associated tag to obtain an extended user tag corresponding to the user identifier includes: when the associated label is the same as the user label, deleting the associated label; and/or when at least two associated labels with the same name exist, keeping the associated label with the maximum association coefficient in the associated labels with the same name; and obtaining an extended user label corresponding to the user identifier according to the reserved user label and the associated label.

Taking the user tag and the associated tag as an example, based on the method of this embodiment, the associated tag a (0.5) is deleted, the user tag a is retained, the associated tag C (0.7) is deleted, and the associated tag C (0.8) is retained, so as to obtain a merged extended user tag, where the extended user tag includes: user tag a, tag B (0.6), tag C (0.8), tag D (0.6), user tag E, and tag F (0.8).

By combining the labels with the same name, only the same label with the maximum association coefficient is reserved, so that the integrity of the expanded user label is ensured, the maximum association between the associated label and the user label is ensured, and an effective basis is provided for subsequent entity recall and entity recommendation.

In an embodiment, the recalling the entity according to the relationship between the user tag and the associated tag to obtain the candidate entity includes: jointly searching the entity library by using the user tags and the associated tags corresponding to the user tags; and obtaining candidate entities according to the retrieval result.

In this embodiment, a manner of a joint recall of the user tag and the associated tag corresponding to the user tag is adopted. Specifically, an entity library is jointly retrieved by using the user tags and the associated tags corresponding to the user tags, and when a certain entity simultaneously comprises at least one user tag and at least one associated tag corresponding to the user tag, the entity is recalled as a candidate entity. Since the extended user tags include both the user tags and the associated tags, if the user tags or the associated tags are directly used for entity recall, the number of recalled entities is too large, a lot of noise is included, and it is difficult to obtain the related entities in which the user is most interested. For example, if the user tag includes "department", "department" corresponds to the associated tag includes "leisure law foreign", and the corresponding entity library includes an drama recommendation list including "leisure law foreign", "friend remember" and other tags, the drama recommendation list may be recalled by the tag "leisure law foreign", and the recalled drama recommendation list is not of interest to the user. Therefore, the entity recalling is directly carried out by using the user tag or the associated tag, so that the recalled entity is difficult to conform to the interest and hobbies of the user, and the quality of the recalled entity is reduced. And the method of the combined retrieval of the user tags and the corresponding associated tags is adopted, the content of the recalled entities is limited through the user tags, the recalled entities are ensured to accord with the historical interests of the users, meanwhile, the recalled entities are further enabled to be more fit with the potential interests of the users by utilizing the associated tags, and the quality of the recalled entities is improved to a certain extent.

Assuming that when the user tag is used for entity recall, a certain entity can only be recalled through the user tag a, and the associated tag of the user tag a includes: for example, if the combined recall method in this embodiment is adopted, the entity is recalled only when the entity also includes at least one of the tag B, the tag C, and the tag D. However, if an entity only includes user tag a, or only includes associated tag B, tag C and/or tag D, it cannot be recalled. By the method, the recalled entity does not deviate from the user interest represented by the user label, and the content of the recalled entity is further limited to be more in line with the potential interest of the user. The entity recall is carried out by combining the user tags and the corresponding associated tags, so that the quality of the recalled entity is ensured, and the complexity of subsequent entity recommendation is further reduced.

In one embodiment, as shown in fig. 3, an entity recommendation method includes:

s301, obtaining a user label corresponding to the user identification meeting the entity recommendation condition.

The user tag can be stored locally through the user terminal, and can also be stored in association with the corresponding user identifier through the server.

S303, obtaining an associated label associated with the user label based on the label association relation.

S305, obtaining an extended user label corresponding to the user identification according to the user label and the associated label.

And S307, recalling the entity according to the relation between the user tag and the associated tag to obtain a candidate entity.

In this embodiment, the relationship between the user tag and the associated tag is used as a search condition to perform entity recall, the related entities meeting the search condition are recalled from the massive entities, and the recalled related entities are used as candidate entities, so as to perform entity recommendation according to the candidate entities in the following. The combined recall is carried out through the relationship between the user label and the extension label, so that the recall quality reduction caused by label extension is avoided.

S309, according to the correlation coefficient of the correlation label, the relation between the user label and the correlation label distributes corresponding weight to the expanded user label.

Specifically, the degree of association between the tags may be used for assignment of the weights. For example, the correlation coefficient of each correlation label is used as the corresponding weight, and the maximum value of the preset correlation coefficient range is used as the weight of each user label, for example, if the preset correlation coefficient range is 0-1, the weight 1 is assigned to each user label. And distributing corresponding weight to each label in the expanded user labels so as to perform entity recommendation operation based on the weight.

S311, matching the extended user tag with the candidate entity to obtain a correlation result of the candidate entity, wherein the correlation result is obtained by calculating according to the weight of the extended user tag.

And S313, extracting the recommending entity from the candidate entities according to the preset recommending strategy and the correlation result, and recommending the recommending entity to the corresponding user identification.

Specifically, the step of matching the extended user tag with the candidate entity to obtain a correlation result of the candidate entity, where the correlation result is obtained by calculating according to the weight of the extended user tag, includes: and extracting entity labels of the candidate entities, and respectively carrying out correlation calculation on the entity labels of the candidate entities and the extended user labels to obtain correlation results of the candidate entities, wherein the correlation results are obtained by calculation according to the weight of the extended user labels. The extended user tags include all corresponding user tags and associated tags.

And after the correlation result of each candidate entity is obtained, extracting the recommending entity from the candidate entities and recommending the recommending entity to the corresponding user identification according to a preset recommending strategy and the correlation result. Specifically, the preset recommendation policy may be to recommend the entities corresponding to the preset number of correlation results according to the size of the correlation results, recommend the correlation results when the correlation results reach a preset recommendation threshold, or perform other recommendation policies set according to the user requirements, which is not limited herein.

According to the entity recommendation method, the associated labels associated with the user labels are obtained based on the label association relation, the expanded user labels are obtained according to the user labels and the associated labels, namely, the label association relation is utilized, the number of the labels of the user is expanded, the labels which can be used during entity recommendation are enriched, and the potential interests of the user are fully mined. On the basis, the combined recall is carried out through the relationship between the user tags and the extension tags, and the recall quality reduction caused by tag extension is avoided. Moreover, weights are distributed to all the labels in the expanded user labels based on the association coefficients, when entity recommendation is carried out by utilizing the correlation results obtained based on the weights, the accuracy of the correlation results can be further improved, the fact that the correlation results of entities meeting the potential interest of the user are large is guaranteed, the entities meeting the user requirements best are mined from mass entities for recommendation, and then the click rate of the recommended entities is effectively improved.

In another embodiment, as shown in fig. 4, when the preset recommendation policy is to recommend an entity corresponding to a preset number of correlation results according to the size of the correlation results, extracting a recommendation entity from the candidate entities according to the preset recommendation policy and the correlation results to recommend the recommendation entity to the corresponding user identifier, including:

s401, sorting the candidate entities according to the sequence of the correlation results from big to small.

And S403, taking a preset number of candidate entities as recommended entities according to the sorting result.

Specifically, the top N candidate entities are taken as the recommending entities according to the sorting result. Wherein, N is a preset number and is a positive integer.

S405, recommending the recommending entity to the corresponding user identification.

When the preset recommendation strategy is to recommend when the correlation result reaches a preset recommendation threshold, extracting the recommending entity from the candidate entity according to the preset recommendation strategy and the correlation result, and recommending the recommending entity to the corresponding user identifier, wherein the recommending strategy comprises: comparing the correlation result with a preset recommendation threshold; and when the correlation result is greater than or equal to a preset recommendation threshold value, recommending the entity corresponding to the correlation result to the corresponding user identifier as a recommending entity.

The server sends the recommending entities to the user terminal corresponding to the user identifier for display, and the specific display mode can be arranged and displayed from large to small according to the correlation of the recommending entities.

In the entity recommending process, the matching degree of the user tags and the candidate entities is considered, meanwhile, the matching degree between the associated tags and the candidate entities is also considered, and the recommended entities are finally obtained by considering the influence of the weights of the user tags and the associated tags on the entity tag matching result, so that the candidate entities containing more user tags and associated tags and the associated tags with higher weights are easier to serve as the recommended entities to preferentially recommend and show to the users, and the click rate of the recommended entities is improved.

Further, matching the extended user tag with the candidate entity to obtain a correlation result of the candidate entity, wherein the correlation result is obtained by calculating according to the weight of the extended user tag, and the method comprises the following steps: and matching the extended user tags, the preset summary features and the candidate entities to obtain correlation results of the candidate entities, wherein the correlation results are obtained by calculation according to the weights of the extended user tags.

The preset summary feature refers to a relevant feature for further verifying valid information of the candidate entity. For example, the preset summary features include: the system comprises a junk information detection characteristic, a heat detection characteristic, a timeliness characteristic and the like. And matching the candidate entities by further combining the preset summary features, so that the obtained recommended entities are more in line with the actual requirements of the user. For example, when the candidate entity is detected to be spam, the calculated correlation result is very low; when the popularity of the candidate entity is higher, the relevance result of the candidate entity can be correspondingly improved, so that the recommended entity can further improve the interest of the user in clicking and browsing, and the click rate of the recommended entity is further improved.

Taking news recommendation as an example, as shown in fig. 5, respectively inputting a news tag, a user tag, an association tag and a preset summary feature of candidate news into a recommendation model, where the recommendation model includes a correlation calculation model, the user tag includes a name and a weight of the user tag, and the association tag includes a name and a weight of the association tag, obtaining a correlation result of the candidate news through the correlation calculation model, and determining whether the candidate news can be recommended as recommended news through a preset recommendation policy model.

In an embodiment, the entity recommendation method further includes a step of establishing a tag association relationship, as shown in fig. 6, the step of establishing the tag association relationship includes:

s601, obtaining the label in the label data source.

Wherein, the tag data source comprises all entities in the mass entity data. Specifically, obtaining the tags in the tag data source may be achieved by obtaining tags that mark all entities.

And S603, recommending the associated labels to the labels respectively to obtain the associated labels and the associated coefficients corresponding to the labels.

And recommending the associated labels according to the correlation among the labels, and respectively obtaining the associated label of each label in the label data source and the associated coefficient corresponding to the associated label. It should be noted that the number of the associated tags of each tag may be limited according to the service requirement, and in general, the preset number of the associated tags of one tag is 20.

And S605, establishing a label association relation according to each label, the corresponding associated label and the association coefficient.

Through the pre-constructed label incidence relation, when label expansion is needed, the incidence label corresponding to the relevant user label and the incidence coefficient thereof can be inquired and obtained based on the label incidence relation, and the expansion of the user label can be quickly realized according to the incidence label and the incidence coefficient thereof.

In an embodiment, as shown in fig. 7, recommending the associated tag for each tag respectively to obtain the associated tag corresponding to each tag includes:

s701, related entities of each label in the label data source are respectively obtained.

In this embodiment, entity recommendation is performed on each tag in the tag data source through a conventional entity recommendation technology, so as to obtain a related entity corresponding to each tag. Specifically, the related entities corresponding to the labels may be obtained by a Learning2rank (machine Learning ranking) method, for example, some training data are manually labeled, and a LambdaMART algorithm is used for training to obtain a ranking model, and the labels and the entity data are input into the ranking model, so that the related entities corresponding to the labels may be obtained, and then the entity labels of the related entities may be extracted.

And S703, obtaining candidate associated labels and associated coefficients corresponding to the labels according to the related entities of the labels.

And acquiring entity labels of related entities corresponding to the labels, and acquiring candidate associated labels and associated coefficients corresponding to the labels based on the acquired entity labels. Specifically, the association coefficients of each tag and the entity tag corresponding to the tag are respectively calculated, and when the association coefficients reach a preset association coefficient threshold, the entity tag corresponding to the association coefficient is used as a candidate association tag. The correlation coefficient threshold may be set according to a requirement, and is not limited herein.

In addition, the correlation coefficient can be obtained by performing weighted calculation according to the relationship of the two labels in the scientific knowledge graph, the aging information, the semantic information and the like. The time efficiency information refers to the co-occurrence of two labels in a recent entity, and the semantic information refers to the similarity of semantics.

S705, obtaining the associated label of each label in the label data source according to the preset associated label recommendation strategy and the associated coefficient of each candidate associated label.

Specifically, for the tags in any tag data source, the corresponding candidate associated tags are sorted according to the size of the association coefficient, and the previous preset number of candidate associated tags with a larger association coefficient are used as the associated tags of the tags. And if the preset number is 20, sorting the candidate associated labels of a certain label according to the size of the association coefficient, and taking the candidate associated label of the top 20 with a larger association coefficient as the associated label of the label. And (3) constructing a complete label association relation by taking an association label from the label in each label data source and based on the association label and the association coefficient.

The following describes an entity recommendation method according to the present application, taking news recommendation as an example. As shown in fig. 8, the method comprises the steps of:

s801, obtaining the label in the label data source.

S802, related entities of each label in the label data source are respectively obtained.

And S803, obtaining candidate associated labels and associated coefficients corresponding to the labels according to the related entities of the labels.

Specifically, some training data can be manually marked, a lamb damart algorithm is used for training, an obtained ranking model is obtained, relevant news corresponding to the label can be obtained by inputting the label and news data into the ranking model, and the news label of the relevant news is extracted to serve as a candidate associated label.

Referring to fig. 8, after obtaining the candidate associated tags corresponding to each tag, the method further includes the following steps:

and S804, obtaining the associated label of each label in the label data source according to the preset associated label recommendation strategy and the associated coefficient of each candidate associated label.

And S805, establishing a label association relation according to each label, the corresponding associated label and the association coefficient.

For example, the association coefficient of each candidate association tag is obtained by performing weighted calculation according to the relation of the news tag and the candidate association tag thereof in the scientific knowledge map, the aging information, the semantic information and the like, the top 20 candidate association tag with a larger association coefficient is taken as the association tag of the news tag, and the tag association relation is established according to the association relation and the association coefficient between the news tags and the association tags, so that the establishment of the tag association relation is completed.

When a certain user identifier reaches an entity recommendation condition, entity recommendation can be performed based on the established tag association relationship, as shown in fig. 8, the method further includes the following steps:

s806, obtaining the user label corresponding to the user identification reaching the entity recommendation condition.

S807, obtaining an associated tag associated with the user tag based on the tag association relation. The label association relationship comprises an association coefficient of the associated label.

For example, when a user opens a news APP through a user terminal, a user identifier corresponding to the user terminal meets an entity recommendation condition, the user terminal generates trigger information carrying the corresponding user identifier and sends the trigger information to a server, the server analyzes the user identifier from the received trigger information, obtains a user tag stored in association with the user identifier according to the user identifier, and searches an associated tag corresponding to the user tag and an associated coefficient thereof based on a pre-established tag association relationship.

Further, when the associated labels and the associated coefficients corresponding to all the user labels are found, the associated labels are compared with the user labels, and the associated labels and the user labels are combined, wherein the combining process includes steps S808 to S809:

s808, deleting the associated label when the associated label is the same as the user label; when at least two associated labels with the same name exist, the associated label with the maximum association coefficient in the associated labels with the same name is reserved.

And S809, obtaining an extended user label corresponding to the user identifier according to the reserved user label and the associated label.

After the extended user tag is obtained, entity recommendation is further performed according to the extended user tag, and accordingly, the entity recommendation method further includes the following steps:

and S810, jointly searching the entity library by using the user tags and the associated tags corresponding to the user tags, and obtaining candidate entities according to a search result.

For example, news data is jointly retrieved by using the user tag and the associated tag corresponding to the user tag, and when a certain news at least comprises one user tag and at least one associated tag corresponding to the user tag, the news is recalled as candidate news.

S811, according to the correlation coefficient of the correlation label, the relation between the user label and the correlation label distributes corresponding weight for the expansion user label.

And S812, matching the extended user tags, the preset summary features and the candidate entities to obtain correlation results of the candidate entities, wherein the correlation results are obtained by calculation according to the weights of the extended user tags.

Still taking news recommendation as an example, the relevance coefficient of the relevance label is used as the weight of the relevance label, meanwhile, the weight 1 is distributed to the user label, and the extended user label, the spam detection feature, the heat detection feature, the timeliness feature and the candidate news are matched by using the weight of each label to obtain the relevance result of the candidate news.

S813, the candidate entities are sorted according to the sequence of the correlation results from big to small.

S814, selecting a preset number of candidate entities as recommended entities according to the sorting result.

And S815, recommending the recommending entity to the corresponding user identification.

Specifically, the candidate news are ranked according to the relevance, 20 candidate news with high relevance are taken as recommended news and sent to the corresponding user terminal of the user identifier, the news APP is used for ranking and displaying according to the relevance, and the user can browse by clicking the recommended news in the news APP.

Based on the application of the entity recommendation method in news recommendation, the click rate of news in various fields is detected, compared with the click rate when news recommendation is carried out by adopting a traditional news recommendation method, and the comparison result is displayed, so that in entertainment-related news, compared with the traditional news recommendation method, the click rate of the news recommendation method is improved by 8.8%; in automobile-related news, compared with the traditional news recommendation method, the click rate of the news recommendation method is improved by 6.62%. Therefore, the entity recommendation method can fully mine the potential interest of the user, so that the recommended entity can attract the user to browse better, and the click rate of the recommended entity is effectively improved.

Fig. 2-8 are flow diagrams illustrating an entity recommendation method in one embodiment. It should be understood that although the various steps in the flow charts of fig. 2-8 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2-8 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternating with other steps or at least some of the sub-steps or stages of other steps.

In one embodiment, as shown in fig. 9, there is provided an entity recommending apparatus including:

a user tag obtaining module 901, configured to obtain a user tag corresponding to the user identifier that meets the entity recommendation condition.

The user tag may be stored locally through the user terminal 110, or may be stored in association with a corresponding user identifier through the server 120.

Specifically, when the user tag is stored in the server, the user tag obtaining module 901 may find the user tag corresponding to the user identifier based on the corresponding relationship between the user identifier and the user tag when obtaining the user identifier that meets the entity recommendation condition. For example, when the server receives trigger information which is sent by the terminal and carries a corresponding user identifier, it is determined that the user identifier meets the entity recommendation condition, the user identifier is obtained by analyzing the trigger information, and then the user identifier corresponding to the user identifier is obtained based on the corresponding relationship between the user identifier and the user identifier.

When the user tag is stored in the server, the user tag obtaining module 901 directly receives the user tag corresponding to the user identifier that meets the entity recommendation condition. For example, when receiving trigger information which is sent by a terminal and carries a corresponding user identifier, it is determined that the user identifier reaches an entity recommendation condition, and a user tag which is stored in the terminal and corresponds to the user identifier is obtained according to the user identifier.

An associated tag obtaining module 903, configured to obtain an associated tag associated with the user tag based on the tag association relationship.

Specifically, the associated label obtaining module 903 uses the obtained label association relationship and queries the label association relationship according to the user label, so as to obtain associated labels corresponding to the user labels respectively.

And the user tag extension module 905 is configured to obtain an extended user tag corresponding to the user identifier according to the user tag and the associated tag.

And the recalling module 907 is configured to perform entity recalling according to the relationship between the user tag and the associated tag to obtain a candidate entity.

In this embodiment, the recall module 907 recalls the relationship between the user tag and the associated tag as a search condition, recalls a related entity meeting the search condition from the mass entities, and takes the recalled related entity as a candidate entity, so as to perform entity recommendation according to the candidate entity in the following. The combined recall is carried out through the relationship between the user label and the extension label, so that the recall quality reduction caused by label extension is avoided.

And a recommending module 909, configured to filter, from the candidate entities, the entities matching with the extended user tag to recommend to the corresponding user identifier.

Further, the recommending module 909 is configured to filter the candidate entity according to the extended user tag, obtain an entity that meets the matching condition, and recommend the entity to the user identifier corresponding to the extended user tag. Specifically, the candidate entities and the extended user tags are matched according to a pre-constructed recommendation model, the recommended entities are obtained according to the matching result, and the recommended entities obtained by matching and screening the candidate entities based on the extended user tags can better meet the potential interest or demand of the user, so that the interest of the user in browsing the recommended entities is promoted, and the click rate of the recommended entities is effectively improved.

According to the entity recommending device, the associated labels associated with the user labels are obtained based on the label association relation, the expanded user labels are obtained according to the user labels and the associated labels, namely, the label association relation is utilized, the label quantity of the users is expanded, the labels which can be used in entity recommending are enriched, and the potential interests of the users are fully mined. On the basis, the combined recall is carried out through the relationship between the user tags and the extension tags, and the recall quality reduction caused by tag extension is avoided. Therefore, when the entity recommendation is performed on the user according to the candidate entity and the extended user tag, the recommended entity can be ensured to meet the potential interest of the user, and the click rate of the recommended entity is effectively improved.

In an embodiment, the user tag extension module 905 is further configured to combine the user tag and the associated tag to obtain an extended user tag corresponding to the user identifier.

Specifically, the user tag expansion module 905 includes a tag merging module and a tag processing module. The tag merging module is used for deleting the associated tag when the associated tag is the same as the user tag; and/or when at least two associated labels with the same name exist, keeping the associated label with the maximum association coefficient in the associated labels with the same name. And the label processing module is used for obtaining an extended user label corresponding to the user identifier according to the reserved user label and the associated label.

In one embodiment, the recall module includes a retrieval module and a candidate entity acquisition module. The retrieval module is used for jointly retrieving the entity library by using the user tags and the associated tags corresponding to the user tags; and the candidate entity acquisition module is used for acquiring a candidate entity according to the retrieval result.

In this embodiment, a manner of a joint recall of the user tag and the associated tag corresponding to the user tag is adopted. Specifically, an entity library is jointly retrieved by using the user tags and the associated tags corresponding to the user tags, and when a certain entity simultaneously comprises at least one user tag and at least one associated tag corresponding to the user tag, the entity is recalled as a candidate entity. The recalled entity is not deviated from the user interest represented by the user label, and the content of the recalled entity is further limited, so that the recalled entity is more in line with the potential interest of the user. The entity recall is carried out by combining the user tags and the corresponding associated tags, so that the quality of the recalled entity is ensured, and the complexity of subsequent entity recommendation is further reduced.

In an embodiment, the entity recommending apparatus further includes a weight assigning module, configured to assign a corresponding weight to the expanded user tag according to the association coefficient of the association tag and the relationship between the user tag and the association tag.

Further, the recommendation module 909 further includes a correlation calculation module and a recommendation sub-module, wherein,

and the correlation calculation module is used for matching the expanded user tags with the candidate entities to obtain correlation results of the candidate entities, and the correlation results are obtained by calculation according to the weights of the expanded user tags.

And the recommending submodule is used for extracting the recommending entity from the candidate entities according to the preset recommending strategy and the correlation result and recommending the recommending entity to the corresponding user identification.

In an embodiment, the correlation calculation module is further configured to extract entity tags of the candidate entities, perform correlation calculation on the entity tags of the candidate entities and the extended user tags respectively, and obtain correlation results of the candidate entities, where the correlation results are obtained by calculation according to weights of the extended user tags. The extended user tags include all corresponding user tags and associated tags.

In an embodiment, the recommending sub-module is further configured to sort the candidate entities in an order from large to small of the correlation result; taking a preset number of candidate entities as recommended entities according to the sorting result; and recommending the recommending entity to the corresponding user identification.

In an embodiment, as shown in fig. 10, the entity recommending apparatus further includes a tag association relationship establishing module 10, where the tag association relationship establishing module 10 includes: a tag source obtaining module 1001, an associated tag recommending module 1003 and a tag association relationship generating module 1005.

A tag source obtaining module 1001, configured to obtain a tag in a tag data source.

The associated tag recommending module 1003 is configured to recommend an associated tag to each tag, respectively, to obtain an associated tag and an associated coefficient corresponding to each tag.

A tag association relationship generating module 1005, configured to establish a tag association relationship according to each tag, a corresponding association tag, and an association coefficient.

And recommending the associated labels according to the correlation among the labels, and respectively obtaining the associated label of each label in the label data source and the associated coefficient corresponding to the associated label. It should be noted that the number of the associated tags of each tag may be limited according to the service requirement, and in general, the preset number of the associated tags of one tag is 20. Through the pre-constructed label incidence relation, when label expansion is needed, the incidence label corresponding to the relevant user label and the incidence coefficient thereof can be inquired and obtained based on the label incidence relation, and the expansion of the user label can be quickly realized according to the incidence label and the incidence coefficient thereof.

In one embodiment, the associated tag recommendation module 1003 includes: the system comprises a related entity acquisition module, a candidate associated tag acquisition module and an associated tag recommendation sub-module.

The related entity obtaining module is used for obtaining related entities of each label in the label data source respectively.

The candidate associated tag obtaining module is used for obtaining a candidate associated tag and an associated coefficient corresponding to each tag according to the related entity of each tag.

And the associated tag recommending submodule is used for obtaining the associated tag of each tag in the tag data source according to a preset associated tag recommending strategy and the association coefficient of each candidate associated tag.

According to the entity recommending device, the associated labels associated with the user labels are obtained based on the label association relation, the expanded user labels are obtained according to the user labels and the associated labels, namely, the label association relation is utilized, the label quantity of the users is expanded, the labels which can be used in entity recommending are enriched, and the potential interests of the users are fully mined. On the basis, the combined recall is carried out through the relationship between the user tags and the extension tags, and the recall quality reduction caused by tag extension is avoided. Moreover, weights are distributed to all the labels in the expanded user labels based on the association coefficients, when entity recommendation is carried out by utilizing the correlation results obtained based on the weights, the accuracy of the correlation results can be further improved, the fact that the correlation results of entities meeting the potential interest of the user are large is guaranteed, the entities meeting the user requirements best are mined from mass entities for recommendation, and then the click rate of the recommended entities is effectively improved.

FIG. 11 is a diagram illustrating an internal structure of a computer device in one embodiment. The computer device may specifically be the server 120 in fig. 1. As shown in fig. 11, the computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program that, when executed by the processor, causes the processor to implement the entity recommendation method. The internal memory may also have stored therein a computer program that, when executed by the processor, causes the processor to perform the entity recommendation method.

Those skilled in the art will appreciate that the architecture shown in fig. 11 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, the entity recommending apparatus provided in the present application may be implemented in the form of a computer program, and the computer program may be run on a computer device as shown in fig. 11. The memory of the computer device may store various program modules constituting the entity recommending apparatus, such as a user tag acquiring module 901, an associated tag acquiring module 903, a user tag expanding module 905, a recall module 907 and a recommending module 909 shown in fig. 9. The computer program constituted by the respective program modules causes the processor to execute the steps in the entity recommending method of the respective embodiments of the present application described in the present specification.

For example, the computer device shown in fig. 11 may execute step S201 through a 901 module in the entity recommending apparatus shown in fig. 9. The computer device may perform step S203 through the 903 module. The computer device may perform step S205 through the 905 module. The computer device may perform step S207 through module 907. The computer apparatus may perform step S209 through the 909 block.

In one embodiment, there is provided a computer device comprising a memory and a processor, the memory having stored therein a computer program which, when executed by the processor, causes the processor to carry out the steps of:

In one embodiment, the processor, when executing the computer program, further performs the steps of:

and combining the user label and the associated label to obtain an extended user label corresponding to the user identifier.

when the associated label is the same as the user label, deleting the associated label; and/or when at least two associated labels with the same name exist, keeping the associated label with the maximum association coefficient in the associated labels with the same name;

and obtaining an extended user label corresponding to the user identifier according to the reserved user label and the associated label.

jointly searching the entity library by using the user tags and the associated tags corresponding to the user tags;

and obtaining candidate entities according to the retrieval result.

according to the correlation coefficient of the correlation label, the relation between the user label and the correlation label distributes corresponding weight to the expanded user label;

matching the extended user tag with the candidate entity to obtain a correlation result of the candidate entity; the correlation result is obtained by calculation according to the weight of the expanded user label;

and extracting a recommending entity from the candidate entities according to a preset recommending strategy and a correlation result, and recommending the recommending entity to the corresponding user identification.

sorting the candidate entities according to the sequence of the correlation results from large to small;

taking a preset number of candidate entities as recommended entities according to the sorting result;

and recommending the recommending entity to the corresponding user identification.

acquiring a label in a label data source;

recommending the associated labels for the labels respectively to obtain the associated labels and the associated coefficients corresponding to the labels;

and establishing a label association relation according to each label, the corresponding associated label and the association coefficient.

respectively acquiring related entities of each label in a label data source;

obtaining candidate associated labels and associated coefficients corresponding to the labels according to the related entities of the labels;

and obtaining the associated label of each label in the label data source according to a preset associated label recommendation strategy and the associated coefficient of each candidate associated label.

In one embodiment, a computer-readable storage medium is provided, storing a computer program that, when executed by a processor, performs the steps of:

In one embodiment, the computer program when executed by the processor implements the steps of:

and obtaining candidate entities according to the retrieval result.

acquiring a label in a label data source;

respectively acquiring related entities of each label in a label data source;

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. An entity recommendation method, the method comprising:

based on the obtained label incidence relation, inquiring the label incidence relation according to the user label to obtain an incidence label associated with each user label; the label incidence relation indicates whether the correlation relation and/or the incidence degree exist among the labels;

according to the relation between the user tag and the associated tag, entity recall is carried out to obtain a candidate entity; when the entity simultaneously comprises at least one user tag and at least one associated tag corresponding to the user tag, the entity is recalled as a candidate entity;

2. The method according to claim 1, wherein obtaining an extended user tag corresponding to the user identifier according to the user tag and the association tag comprises:

3. The method of claim 2, wherein the tag association relationship comprises an association coefficient of an associated tag; the merging the user tag and the associated tag to obtain an extended user tag corresponding to the user identifier includes:

deleting the associated tag when the associated tag is the same as the user tag; and/or when at least two associated labels with the same name exist, the associated label with the maximum association coefficient in the associated labels with the same name is reserved;

4. The method of claim 1, wherein the recalling entities according to the relationship between the user tag and the associated tag to obtain candidate entities comprises:

jointly searching an entity library by using the user tag and the associated tag corresponding to the user tag;

and obtaining candidate entities according to the retrieval result.

5. The method of claim 3, further comprising: according to the correlation coefficient of the correlation label, the relation between the user label and the correlation label distributes corresponding weight to the expansion user label;

the screening of entities matching the expanded user tag from the candidate entities to recommend to the corresponding user identifier includes:

matching the extended user tag with the candidate entity to obtain a correlation result of the candidate entity; the correlation result is obtained by calculation according to the weight of the extended user tag;

and extracting a recommending entity from the candidate entities according to a preset recommending strategy and the correlation result, and recommending the recommending entity to the corresponding user identification.

6. The method according to claim 5, wherein the extracting, according to a preset recommendation policy and the correlation result, a recommendation entity from the candidate entities to recommend to the corresponding user identifier comprises:

sorting the candidate entities according to the sequence of the correlation results from big to small;

7. The method according to any one of claims 1 to 6, further comprising:

acquiring a label in a label data source;

recommending associated labels for the labels respectively to obtain associated labels and associated coefficients corresponding to the labels;

8. The method of claim 7, wherein the recommending the associated tag for each tag respectively to obtain the associated tag and the associated coefficient corresponding to each tag comprises:

respectively acquiring related entities of each label in a label data source;

9. An entity recommendation apparatus, the apparatus comprising:

the associated tag obtaining module is used for inquiring the tag association relation according to the user tag based on the obtained tag association relation to obtain an associated tag associated with the user tag; the label incidence relation indicates whether the correlation relation and/or the incidence degree exist among the labels;

the recalling module is used for recalling the entity according to the relation between the user tag and the associated tag to obtain a candidate entity; when the entity simultaneously comprises at least one user tag and at least one associated tag corresponding to the user tag, the entity is recalled as a candidate entity;

10. The apparatus of claim 9, wherein the user tag expansion module is further configured to combine the user tag and the associated tag to obtain an expanded user tag corresponding to the user identifier.

11. The apparatus of claim 10, wherein the tag association relationship comprises an association coefficient for associating tags; the user tag expansion module includes:

the tag merging module is used for deleting the associated tag when the associated tag is the same as the user tag; and/or when at least two associated labels with the same name exist, the associated label with the maximum association coefficient in the associated labels with the same name is reserved;

and the label processing module is used for obtaining an extended user label corresponding to the user identifier according to the reserved user label and the associated label.

12. The apparatus of claim 9, wherein the recall module comprises:

the retrieval module is used for jointly retrieving the entity library by utilizing the user tags and the associated tags corresponding to the user tags;

and the candidate entity acquisition module is used for acquiring a candidate entity according to the retrieval result.

13. The apparatus of any one of claims 9 to 12, further comprising:

the tag source acquisition module is used for acquiring tags in the tag data source;

the associated tag recommendation module is used for recommending associated tags to the tags respectively to obtain the associated tags and associated coefficients corresponding to the tags;

and the tag association relation establishing module is used for establishing tag association relations according to the tags, the corresponding associated tags and the association coefficients.

14. A computer device comprising a memory and a processor, the memory storing a computer program, wherein the computer program, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 8.

15. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, causes the processor to carry out the steps of the method according to any one of claims 1 to 8.