Official document recommendation method and device

Abstract

The present invention discloses a document recommendation method and device, and relates to the field of computer technology. A specific implementation of the method includes: obtaining association relationships between multiple documents; determining the documents to be recommended among the multiple documents according to the association relationships and characteristic data of a target user; and recommending the documents to be recommended to the target user. This implementation can recommend documents to the target user according to the association relationships between the documents, thereby having a high recommendation accuracy.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a document recommendation method and device. Background Art

[0002] Official documents are obviously different from news and information. They are characterized by relatively fixed content, relatively complete elements, and relatively clear structure. In the current official document recommendation technology, the traditional recommendation method is mainly used, that is, recommendation is made based on the similarity between recommendation objects or the similarity between users. However, in fact, different official documents will show obvious logical correlations based on the content. The above recommendation method does not take this correlation into account, so the accuracy is low. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides a document recommendation method and device, which can recommend documents to target users according to the association relationship between documents, thereby having high recommendation accuracy.

[0004] To achieve the above objective, according to one aspect of the present invention, a method for recommending official documents is provided.

[0005] The document recommendation method of the embodiment of the present invention includes: obtaining the association relationship between multiple documents; determining the document to be recommended among the multiple documents according to the association relationship and the characteristic data of the target user; and recommending the document to be recommended to the target user.

[0006] Optionally, the association relationship between the multiple official documents includes: at least one official document that has an association relationship with any one of the multiple official documents and the type of the association relationship; and determining the official document to be recommended among the multiple official documents based on the association relationship and characteristic data of the target user, includes: for any one of the multiple official documents: determining the feature vector of any one of the multiple official documents based on the official document that has an association relationship with the any one official document and the type of the association relationship; determining the first potential attention of the target user to the any one official document based on the feature vector of the any one official document and the characteristic data of the target user; obtaining the comprehensive potential attention of the target user to the any one official document based on the first potential attention; and determining the any one official document as a document to be recommended when the comprehensive potential attention meets preset recommendation conditions.

[0007] Optionally, determining the feature vector of any official document based on the official documents that have an associated relationship with the any official document and the type of the associated relationship includes: for any official document among the multiple official documents: inputting the original features of the any official document, the feature vector of each official document that has an associated relationship with the any official document, and the preset weight value corresponding to the type of the associated relationship into a preset function for calculation to obtain the feature vector of the any official document.

[0008] Optionally, determining the feature vector of any official document based on the official documents that have an association relationship with the any official document and the type of the association relationship includes: inputting the original features of the multiple official documents and the association relationship data between the multiple official documents into a pre-trained feature extraction model to obtain the feature vector of any official document among the multiple official documents.

[0009] Optionally, the characteristic data of the target user includes a current characteristic vector of the target user; and the method further includes: obtaining an initial characteristic vector of the target user; and determining a current characteristic vector of the target user according to the initial characteristic vector of the target user and historical behavior data of the target user with respect to official documents.

[0010] Optionally, the obtaining of the initial feature vector of the target user includes: determining the initial feature vector of the target user according to a preferred document category provided by the target user.

[0011] Optionally, determining the current feature vector of the target user based on the initial feature vector of the target user and the historical behavior data of the target user with respect to official documents includes: obtaining the feature vector of each official document corresponding to the historical behavior data; multiplying the feature vector of any official document by the influence score of the official document to obtain the influence vector of the official document; wherein the influence score is determined based on the type of historical behavior of the target user with respect to the official document; and determining the current feature vector of the target user based on the influence vector of each official document and the initial feature vector of the target user.

[0012] Optionally, determining the first potential attention of the target user to any official document based on the feature vector of any official document and the feature data of the target user includes: for any official document among the multiple official documents, calculating the similarity between the feature vector of any official document and the current feature vector of the target user; and determining the first potential attention of the target user to any official document based on the similarity.

[0013] Optionally, determining the first potential attention of the target user to any official document based on the feature vector of the any official document and the feature data of the target user includes: inputting the feature vector of any official document among the multiple official documents and the current feature vector of the target user into a pre-trained intention recognition model to obtain the probability of the target user's attention to the any official document; and determining the first potential attention of the target user to the any official document based on the attention probability.

[0014] Optionally, obtaining the target user's comprehensive potential attention to any official document based on the first potential attention includes: determining the target user's first potential attention to any official document among the multiple official documents as the target user's comprehensive potential attention to any official document.

[0015] Optionally, the method further includes: for any official document among the multiple official documents, calculating the similarity between the original features of the any official document and the feature data of the target user; determining the second potential attention of the target user to the any official document based on the similarity; and obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention, including: determining the comprehensive potential attention of the target user to the any official document by combining the first potential attention and the second potential attention.

[0016] Optionally, the method further includes: inputting the original features of any official document among the multiple official documents and the comprehensive features of the target user into a pre-trained intention discrimination model to obtain the probability of the target user's attention to the any official document, and determining the second potential attention of the target user to the any official document based on the attention probability; wherein the comprehensive features of the target user are determined by the target user's historical behavior data for the official document and the target user's attribute characteristics; and obtaining the target user's comprehensive potential attention to the any official document based on the first potential attention includes: determining the target user's comprehensive potential attention to the any official document in combination with the first potential attention and the second potential attention.

[0017] Optionally, the method further includes: for any official document among the multiple official documents, determining a first time penalty item based on the time between the release time of any official document and the current time, and using the first time penalty item to determine the third potential attention of the target user to the any official document; and obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention, including: determining the comprehensive potential attention of the target user to the any official document by combining the first potential attention, the second potential attention and the third potential attention.

[0018] Optionally, the method further includes: for any official document among the multiple official documents: when the target user has browsed any official document at a historical moment, determining a second time penalty item based on the length of time between the moment when the target user last browsed any official document and the current moment; and using the first time penalty item to determine the target user's third potential attention to any official document, including: using the first time penalty item and the second time penalty item to determine the target user's third potential attention to any official document.

[0019] Optionally, the method further includes: for any official document among the multiple official documents, determining a fourth potential attention of the any official document according to the number of views of the any official document in at least one historical time interval; and obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention, including: determining the comprehensive potential attention of the target user to the any official document by combining the first potential attention, the second potential attention, the third potential attention and the fourth potential attention.

[0020] Optionally, the method further comprises: when any official document among the plurality of official documents is related to a predetermined hot event, increasing the fourth potential attention level of the any official document according to a preset strategy.

[0021] Optionally, the method further includes: when any official document among the multiple official documents and the target user satisfy at least one preset rule, determining the fifth potential attention of the any official document according to the recommendation weight set in the preset rule; obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention includes: determining the comprehensive potential attention of the target user to the any official document in combination with the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention.

[0022] Optionally, determining the fifth potential attention degree of any official document based on the recommendation weight set in the preset rules includes: when there is one preset rule, determining the recommendation weight set in the preset rule as the fifth potential attention degree of any official document; when there are multiple preset rules, determining the maximum value of the recommendation weights in the multiple preset rules as the fifth potential attention degree of any official document.

[0023] Optionally, obtaining the target user's comprehensive potential attention to any official document based on the first potential attention includes: determining the weighted sum of the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention as the target user's comprehensive potential attention to any official document.

[0024] Optionally, when the comprehensive potential attention meets the preset recommendation conditions, any official document is determined as a document to be recommended, including: arranging the multiple official documents in descending order according to the comprehensive potential attention; and determining a preset number of documents arranged in front as documents to be recommended.

[0025] Optionally, the characteristic data of the target user includes: official documents that the target user has browsed, collected or subscribed to in a preset historical time period; and determining the official documents to be recommended from the multiple official documents based on the association relationship and the characteristic data of the target user includes: determining the official documents from the multiple official documents that the target user has browsed, collected or subscribed to in the preset historical time period as target official documents; and determining at least one official document that has an association relationship with the target official document as the official document to be recommended.

[0026] Optionally, the types of the association relationship include: revision relationship, cancellation relationship, based on relationship and mention relationship; the types of the historical behavior include: browsing, collection, subscription, ignoring, uncollection and unsubscription.

[0027] To achieve the above objective, according to another aspect of the present invention, a document recommendation device is provided.

[0028] The document recommendation device of the embodiment of the present invention may include: an information extraction unit, used to obtain the association relationship between multiple documents; a document determination unit, used to determine the document to be recommended among the multiple documents based on the association relationship and the characteristic data of the target user; and a recommendation unit, used to recommend the document to be recommended to the target user.

[0029] To achieve the above objective, according to another aspect of the present invention, an electronic device is provided.

[0030] An electronic device of the present invention includes: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the document recommendation method provided by the present invention.

[0031] To achieve the above objective, according to yet another aspect of the present invention, a computer-readable storage medium is provided.

[0032] A computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the document recommendation method provided by the present invention.

[0033] According to the technical solution of the present invention, the embodiments of the above invention have the following advantages or beneficial effects:

[0034] First, the associations such as revision, abolition, basis, and mention among multiple official documents are extracted, and then the official documents to be recommended are determined based on the associations and the characteristic data of the target user, and finally the official documents to be recommended are recommended to the target user. Based on the above settings, the logical associations between official documents can be fully considered to determine the official documents that match the target user, thereby improving the accuracy of the recommendation.

[0035] Secondly, when determining the official document to be recommended based on the association relationship between official documents, the embodiment of the present invention can first calculate the feature vector of any official document, and then determine the first potential attention of the target user to the official document (i.e., the potential attention based on the association relationship of the official document) according to the feature vector of the official document and the feature data of the target user, and finally determine the official document to be recommended based on the first potential attention. When calculating the feature vector of any official document, the original features of the official document, the feature vectors of the official documents with which the official document has an association relationship, and the weight value corresponding to the type of the association relationship can be directly calculated by inputting the preset function, or the original features of the official document and the association relationship data between the official documents can be input into a pre-trained feature extraction model for calculation. When calculating the first potential attention, it can be based on the similarity between the official document feature vector and the target user feature data, or it can be based on the probability of the target user's attention to the official document obtained by inputting the official document feature vector and the target user feature data into a pre-trained intention recognition model. Through the above settings, the calculation of official document features applicable to a variety of specific scenarios and the matching of official documents and users are realized.

[0036] Third, in an embodiment of the present invention, the characteristic data of the target user may be the current characteristic vector of the target user, and the current characteristic vector may change dynamically based on the initial characteristic vector of the target user as the target user browses, collects, subscribes to, and other behaviors with respect to official documents. Thus, after the target user browses, collects, or subscribes to a certain official document, its current characteristic vector may change toward the characteristic vector of the official document. This change may result in a higher degree of match between the target user and the associated official documents of the official document (i.e., the official documents that have an associated relationship with the official document) (the degree of match may be the characteristic vector similarity, the probability of attention, or the first potential degree of attention), thereby enabling accurate and real-time recommendations of associated official documents.

[0037] Fourthly, in an embodiment of the present invention, in addition to considering the correlation between official documents, collaborative recommendations can also be made in combination with one or more of the following methods: recommendations based on user behavior, recommendations based on time penalty items, recommendations based on the number of official document views, and recommendations based on preset rules. This can further improve the accuracy of recommendations.

[0038] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0040] Figure 1 Schematic diagram of the main steps of the document recommendation method in an embodiment of the present invention;

[0041] Figure 2is a schematic diagram of a knowledge graph constructed by multiple official documents in an embodiment of the present invention;

[0042] Figure 3 Schematic diagram of specific implementation steps of the document recommendation method in an embodiment of the present invention;

[0043] Figure 4 It is a schematic diagram of components of the document recommendation device in an embodiment of the present invention;

[0044] Figure 5 is an exemplary system architecture diagram in which embodiments of the present invention may be applied;

[0045] Figure 6 It is a schematic diagram of the structure of an electronic device used to implement the document recommendation method in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0047] It should be pointed out that the embodiments of the present invention and the technical features therein may be combined with each other without conflict.

[0048] Figure 1 Schematic diagram of main steps of the document recommendation method according to an embodiment of the present invention.

[0049] like Figure 1 As shown, the document recommendation method of the embodiment of the present invention can be specifically performed according to the following steps:

[0050] Step S101: Obtain association relationships between multiple official documents.

[0051] In the embodiment of the present invention, official documents refer to written materials formed in a certain format by various types of institutions or organizations, including various policies, documents and internal documents of state organs and government agencies at all levels, as well as various documents and internal documents of public institutions and enterprise units. Official documents can be displayed in various formats, such as text, pictures, audio, and video. Official documents generally have a specific format and are formed through a certain processing procedure. At the same time, official documents often have content and logical associations. The types of associations can include revision relationships, abolition relationships, based on relationships, and reference relationships. For example, when the subordinate management functional department needs to form official document B based on official document A issued by the superior management functional department, there is a basis relationship between official document B and official document A, that is, official document B is based on official document A; when official document F abolishes official document B, there is an abolition relationship between official document F and official document B; when official document F revises official document E, there is a revision relationship between official document F and official document E; when official document C mentions official document A, there is a reference relationship between official document C and official document A. It should be noted that the relationship between official documents is not limited to the above four types and can be expanded according to actual conditions.

[0052] Figure 2 is a schematic diagram of a knowledge graph constructed by multiple official documents in an embodiment of the present invention. Figure 2 In the example, circular nodes represent official documents, and edges between nodes represent the relationship between official documents. Figure 2 As shown, there is a basis relationship between official document A and official document B (official document B is based on official document A), there is a basis relationship between official document E and official document B (official document E is based on official document B), there is a mention relationship between official document C and official document A (official document C mentions official document A), there is a revision relationship between official document F and official document E (official document F revised official document E), there is a basis relationship between official document F and official document C (official document F is based on official document C), there is a basis relationship between official document D and official document B (official document D is based on official document B), there is a revision relationship between official document D and official document C (official document D revised official document C), and there is an abolition relationship between official document F and official document B (official document F abolished official document B).

[0053] It can be seen that the relationship between any two documents is directional. Figure 2 A directed edge with an arrow is used to indicate this directionality. Generally speaking, for any two official documents with an associated relationship, the issuance time of the official document pointed to by the arrow of the directed edge is earlier than the issuance time of the other official document. It can be understood that the above-mentioned issuance time can correspond to any time precision, such as accurate to the year, month, day, hour, minute or second. In addition, in specific scenarios, different association relationships between official documents often represent different degrees of correlation between official documents. For example, the degree of correlation represented by the abolition relationship and the revision relationship is higher than the according relationship, and the degree of correlation represented by the according relationship is higher than the mention relationship.

[0054] In particular, the association relationship described above is a direct association relationship between official documents. In addition, official documents may also have an indirect association relationship, for example, Figure 2 The official document F in the document abolished the official document B, and the official document B was based on the official document A. The official document F and the official document A have a secondary indirect association relationship (which can be referred to as a secondary association relationship), in which the first-level association relationship is abolition and the second-level association relationship is based on. This secondary association relationship can be expressed as an abolition-based on relationship. In this way, the first-level associated official documents (that is, the official documents that have a direct association relationship with any official document) and multi-level associated official documents (that is, the official documents that have a multi-level association relationship with any official document) of any official document can be obtained from the knowledge graph. For example, the first-level associated official documents of official document A are official documents B and C, and the second-level associated official documents are official documents D, E, and F. If official document A is taken as the root node, official documents B and C are at the first level, and official documents D, E, and F are at the second level.

[0055] In step S101, a pre-trained information extraction model may be used to obtain the association relationship between multiple documents that are pre-acquired for subsequent recommendation, and the association relationship may include: at least one document that has an association relationship with any document among the multiple documents and the type of the association relationship. In practical applications, the text information of any document among the multiple documents may be input into the information extraction model to obtain the document that has an association relationship with the document and the type of the association relationship.

[0056] Step S102: determining the official documents to be recommended from among the multiple official documents according to the association relationship and the characteristic data of the target user.

[0057] In this step, the target user refers to the user targeted for recommendation, the official document to be recommended refers to the official document that is determined by calculation and will be recommended to the user, and the characteristic data of the target user refers to the data that can reflect the user's characteristics, which may include one or more of the following aspects: the target user's gender, age, occupation, address and other attribute characteristics, the target user's preferred official document category (official document categories can be divided from multiple dimensions, for example, can be divided into commands, decisions, resolutions, instructions, etc.), and the target user's historical behavior data on official documents. In actual applications, users can have multiple types of behaviors on official documents, such as browsing, collecting, subscribing, searching, watching (can be for official documents presented in video form), feedback, ignoring, canceling collection and unsubscribing. The above historical behavior data may include: the target user's browsing, collection, subscription, uncollection and unsubscription of official documents at historical moments, the duration of browsing a certain official document, the progress percentage of watching a certain official document, the search term entered in the search box, and feedback information on a certain official document (such as evaluation information). Exemplarily, the relevant data of an official document may include the title, abstract, content, original features and feature vector of the official document; wherein, the original features of the official document refer to the features of the official document obtained based on the title, abstract or content of the official document using algorithms or models such as TF-IDF (term frequency-inverse document frequency), word2vec (a word vector model), BERT (a natural language processing model), etc., and the feature vector of the official document refers to the features of the official document formed after considering the correlation relationship of the official document on the basis of the original features of the official document.

[0058] In an embodiment of the present invention, step S102 can be specifically performed according to the following steps: for any official document among the multiple official documents obtained in advance, first determine the feature vector of any official document according to the official documents that have an association relationship with the any official document and the type of the association relationship, then determine the first potential attention of the target user to the any official document based on the feature vector of the any official document and the characteristic data of the target user, then obtain the comprehensive potential attention of the target user to the any official document based on the first potential attention, and finally, when the comprehensive potential attention meets the preset recommendation conditions, determine the any official document as a document to be recommended. Among them, the potential attention can represent the probability of the user's interest in the official document, the first potential attention is the potential attention obtained after considering the association relationship of the official document, and the comprehensive potential attention is a potential attention index obtained after evaluating the probability of the user's interest in the official document from at least one aspect.

[0059] Preferably, in some embodiments, two methods can be used to calculate the feature vector of the official document. In the first method, for any official document among the multiple official documents obtained in advance, the original features of the any official document, the feature vector of each official document that has an association relationship with the any official document, and the preset weight value corresponding to the type of the association relationship can be input into the preset function for calculation, thereby obtaining the feature vector of the any official document. Before performing the above calculation, it can be determined according to the actual scenario which levels of related documents need to be selected for calculation, and weight values ​​can be set for different association relationships according to the degree of relevance, such as setting a weight value of 0.5 for the abolition relationship and the revision relationship, a weight value of 0.3 for the according relationship, a weight value of 0.1 for the mention relationship, and a weight value of 0.03 for the according-mention relationship.

[0060] In addition, in an optional implementation, the associated official documents with an earlier issuance time than any official document can be selected for calculation. For example, in the case where it is necessary to select the associated official documents of the first level for calculation, for official document F, official documents B, C, and E with earlier issuance time can be selected for calculation; for official document B, official document A with an earlier issuance time can be selected for calculation. In the case where it is necessary to select the associated official documents of the first and second levels for calculation, for official document F, the first-level associated official documents B, C, and E with earlier issuance time and the second-level associated official documents A and D can be selected for calculation. In another optional implementation, any associated official document of any official document can be selected for calculation, not limited to the official documents with an earlier issuance time than any official document. In this case, the original features of the associated official documents can be used for calculation. The following will take the case of selecting the calculation of the associated official documents of the first level as an example for explanation. It can be understood that when calculating the feature vector of a certain official document, the associated official documents of any level can be selected for calculation as needed, and the above examples do not form any restrictions on this.

[0061] The above preset function for calculating the document feature vector can be set according to actual needs, for example, it can be set as follows:

[0062]

[0063] Among them, M i The feature vector representing document i (i is the document identifier), M i0 represents the original features of document i, M j represents document j that is associated with document i. j is a positive integer starting from 1 and represents the sequence number of the associated document. k j represents the weight value of the association relationship between document j and document i, and p and q are preset parameters. It can be understood that the above formula is only an example and does not form any limitation on the form of the above preset function.

[0064] In the second method for calculating the feature vector of an official document, the original features of a plurality of official documents acquired in advance and the data on the association relationship between the official documents can be input into a pre-trained feature extraction model to obtain the feature vector of any of the plurality of official documents. The data on the association relationship between the official documents can include the identifier of the associated official document of any of the plurality of official documents and the type of the corresponding association relationship, and the feature extraction model can be a neural network model such as CNN (Convolutional Neural Networks) and RNN (Recurrent Neural Network). In one embodiment, the feature extraction model can use the following formula to calculate the feature vector of the official document:

[0065]

[0066] in, represents the feature vector of the document node in the feature extraction model, the subscript v represents the index number of the document node, the superscript k represents the layer of the document node in the feature extraction model, σ represents the activation function (such as the linear rectification function ReLU), W k , B k Both are weight matrices, N(v) represents the set of associated documents of document node v, AGG(·) is an aggregation function, which can collect the features of associated documents by calculating the average value, weighted sum, convolution pooling, etc. AGG(·) can contain the weight value of the corresponding relationship of the associated document. In the above two methods of calculating the document feature vector, the original features of the document are used, which can solve the cold start problem of the system.

[0067] In an embodiment of the present invention, the characteristic data of the target user may be the current characteristic vector of the target user, and the current characteristic vector may be dynamically updated according to the historical behavior data of the target user for the official document. It is calculated according to the following steps: First, the initial characteristic vector of the target user is obtained. In one embodiment, the initial characteristic vector of the target user may be determined according to the preferred official document category provided by the target user, for example, the characteristic vectors of the preferred official document category entered by the target user when registering are averaged to obtain the initial characteristic vector of the target user. Thereafter, the current characteristic vector of the target user may be determined according to the initial characteristic vector of the target user and the historical behavior data of the target user for the official document. Specifically, the characteristic vector of each official document corresponding to the historical behavior data of the target user may be first obtained, and then the characteristic vector of any official document may be multiplied by the influence score of the official document to obtain the influence vector of the official document, wherein the above-mentioned influence score may be determined according to the type of the historical behavior of the target user for the official document, for example, the influence score of browsing may be set to 0.1, the influence score of collecting may be set to 0.2, the influence score of subscribing may be set to 0.3, the influence score of ignoring may be set to -0.1, the influence score of canceling the collection may be set to -0.2, and the influence score of canceling the subscription may be set to -0.3. Finally, the current feature vector of the target user may be determined according to the influence vector of each document and the initial feature vector of the target user. For example, the influence vector of each document may be added to the initial feature vector of the target user to obtain the current feature vector of the target user.

[0068] According to the above settings, the user's current feature vector can be updated in real time as the user's behavior changes. In actual applications, when the target user browses, collects or subscribes to a document, his current feature vector may change towards the feature vector of the document. This change will result in a higher degree of match between the target user and the related documents of the document, thereby achieving accurate related document recommendations and real-time recommendations.

[0069] After obtaining the feature vector of the official document and the current feature vector of the target user, the first potential attention of the target user to any of the multiple official documents obtained in advance can be calculated. Two calculation methods are provided below. In one calculation method, for any official document among the multiple official documents, the similarity between the feature vector of the any official document and the current feature vector of the target user can be calculated, and the first potential attention of the target user to the any official document can be determined based on the similarity, for example, the similarity is multiplied by a preset coefficient to obtain the first potential attention, or the similarity is directly used as the first potential attention. Among them, the above similarity can be cosine similarity, Jaccard similarity, Pearson correlation coefficient or adjusted cosine similarity,

[0070] In the second calculation method, the feature vector of any document among the multiple documents and the current feature vector of the target user can be input into the pre-trained intention recognition model to obtain the probability of the target user's attention to the document, and finally determine the first potential attention of the target user to the document based on the attention probability, for example, the first potential attention is obtained by multiplying the attention probability with a preset coefficient, or directly using the attention probability as the first potential attention. Among them, the intention recognition model can be a classification model such as naive Bayes and logistic regression. In particular, the above-mentioned intention recognition model can be connected with the aforementioned feature extraction model and trained together. For example, the output data of the feature extraction model and the current feature vector of the target user can be input into the intention recognition model to realize the connection between the feature extraction model and the intention recognition model, and the positive and negative samples are determined according to the feedback effect of the user on the recommended document at the historical moment (such as whether to browse, browsing time, evaluation information), and the training set is constructed based on the positive and negative samples to train the above-mentioned feature extraction model and the intention recognition model at the same time.

[0071] In some embodiments, after obtaining the first potential attention of the target user to any official document, the first potential attention can be determined as the comprehensive potential attention of the target user to the any official document for subsequent recommendation.

[0072] Preferably, in an embodiment of the present invention, the first potential attention can be combined with the potential attention of other aspects to calculate the comprehensive potential attention, so that more accurate document recommendation can be achieved. In a specific application, for any document among the multiple documents obtained in advance, the similarity between the original features of the document and the feature data of the target user can be calculated, and the second potential attention of the target user to the document can be determined based on the similarity, for example, the similarity is multiplied by a preset coefficient to obtain the second potential attention, or the similarity is directly used as the second potential attention. Among them, the feature data of the target user can be calculated according to the following steps: first, the initial feature vector of the target user is determined according to the preferred document category provided by the target user, and then the original features of each document corresponding to the historical behavior data of the target user are obtained, and the original features of any document are multiplied by the influence score of the document to obtain the specific vector of the document. Finally, the specific vector of each document can be added to the initial feature vector of the target user to obtain the feature data of the target user.

[0073] In practical applications, the original features of any official document and the comprehensive features of the target user can also be input into a pre-trained intention discrimination model to obtain the probability of the target user's attention to any official document, and the second potential attention of the target user to any official document can be determined based on the attention probability, for example, the second potential attention can be obtained by multiplying the attention probability with a preset coefficient, or directly using the attention probability as the second potential attention. Among them, the comprehensive features of the target user can be determined by the historical behavior data of the target user for the official document and the attribute features of the target user. The above comprehensive features can be spliced ​​from the following parts: data formed by calculating the word vector and averaging the pooling of the search word sequence of the target user's behavior data, data calculated by the word vector and averaging the pooling of the title, abstract, etc. of each official document browsed in the target user's behavior data, and attribute feature data such as the name, gender, and occupation of the target user. In specific scenarios, the above intention discrimination model can be various deep learning models. When training the above intention discrimination model, the positive and negative samples can be first determined based on the feedback effect of the user on the recommended official document at the historical moment (such as whether to browse, browsing time, and evaluation information), and then the training set is constructed based on the positive and negative samples to train the above intention discrimination model at the same time.

[0074] After obtaining the second potential attention of the target user to any official document, the comprehensive potential attention of the target user to the any official document can be determined by combining the first potential attention and the second potential attention. For example, the sum or weighted sum of the first potential attention and the second potential attention can be calculated as the comprehensive potential attention according to a preset method. The present invention does not impose any restrictions on the method of calculating the comprehensive potential attention.

[0075] In actual application scenarios, relatively speaking, people pay more attention to newly released official documents. The longer a certain official document is released, the less interested people are in it. On the other hand, for official documents that people have just read, it is obviously not necessary to recommend them repeatedly. Based on the above considerations, a time penalty item based on the time when the official document is released can be established to determine the corresponding potential attention. Specifically, for any official document, the first time penalty item can be determined based on the length of time between the time when the official document is released and the current time, and the first time penalty item can be used to determine the target user's third potential attention to the any official document (for example, directly use the first time penalty item as the third potential attention). For example, the first time penalty item can be expressed as follows:

[0076] -log(t publish (p i ))*punish publish

[0077] Among them, p i Indicates any official document, t publish (p i) indicates the time between the release time of the document and the current time, punish publish is an adjustable parameter.

[0078] In an optional implementation, for any official document, when the target user has browsed the official document at a historical moment, the second time penalty item can be determined based on the time between the last time the target user browsed the official document and the current moment, and the third potential attention of the target user to the official document can be calculated based on the first time penalty item and the second time penalty item. The second time penalty item can be shown as follows:

[0079] lo g (Tt watch (p i ,u j )+1)*punish watch

[0080] Where T is the preset time constant, u j Indicates the target user, punish watch is an adjustable parameter.

[0081] Exemplarily, the third potential attention of the target user to the document based on the first time penalty item and the second time penalty item can be calculated by adding the first time penalty item and the second time penalty item to obtain the third potential attention. After obtaining the third potential attention, the first potential attention, the second potential attention and the third potential attention can be combined to determine the comprehensive potential attention of the target user to the document, for example, the sum or weighted sum of the first potential attention, the second potential attention and the third potential attention is calculated according to a preset method as the comprehensive potential attention.

[0082] In some embodiments, for any document among the multiple documents, the fourth potential attention level of the document may be determined according to the number of views of the document in at least one historical time interval, for example, by the following formula:

[0083] w4(p i ,u j )=log(v 1h (p i ))*a 1h +log(v 1d (p i ))*a 1d +log(v 7d (p i ))*a 7d

[0084] Among them, p i Indicates official documents, u jrepresents the target user, w4(p i ,u j ) represents the fourth potential attention, v 1h Indicates the number of views of the document in the last hour, v 1d Indicates the number of views of the document in the last day, v 7d Indicates the number of views of the document in the past week. 1h 、a 1d 、a 7d are preset parameters.

[0085] In practical applications, when any of the multiple documents is related to a predetermined hot event, the fourth potential attention of the document can be increased according to a preset strategy. For example, the a in the above formula can be increased. 1h 、a 1d 、a 7d After obtaining the fourth potential attention, the first potential attention, the second potential attention, the third potential attention and the fourth potential attention can be combined to determine the comprehensive potential attention of the target user to the document. For example, the sum or weighted sum of the first potential attention, the second potential attention, the third potential attention and the fourth potential attention can be calculated according to a preset method as the comprehensive potential attention.

[0086] In specific applications, in order to achieve specific recommendation requirements, it can support recommending designated documents or documents of designated categories to users of designated ranges or designated categories in a confirmed task manner. Therefore, in an embodiment of the present invention, when any document among multiple documents and the target user meet at least one preset rule, the fifth potential attention degree of any document is determined according to the recommendation weight set in the preset rule. Specifically, when one preset rule is met, the recommendation weight set in the preset rule can be determined as the fifth potential attention degree of any document; when multiple preset rules are met, the maximum value of the recommendation weights in multiple preset rules can be determined as the fifth potential attention degree of any document.

[0087] After obtaining the fifth potential attention, the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention can be combined to determine the target user's comprehensive potential attention to any official document. For example, the sum or weighted sum of the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention can be calculated according to a preset method as the comprehensive potential attention.

[0088] After obtaining the target user's comprehensive potential attention to any document, the document that meets the preset recommendation conditions can be determined as the document to be recommended. For example, multiple documents can be arranged in descending order according to the comprehensive potential attention, and the documents with a preset number of documents arranged in the front can be determined as the document to be recommended.

[0089] In particular, in the embodiment of the present invention, recommendation can be achieved without relying on the feature vector of the document. Specifically, when the target user has browsed, collected or subscribed to the documents in the above documents in a preset historical time period (for example, the last hour), these documents are determined as target documents, and at least one document associated with the target document is determined as a document to be recommended. For example, Figure 2 In the example, when the target user has browsed document A in the last hour, the first-level related documents B and C of document A can be determined as documents to be recommended. Of course, the first-level related documents B and C of document A and the second-level related documents D, E, and F can also be determined as documents to be recommended. Through the above settings, simple and fast recommendations can be achieved based on the document association relationship.

[0090] Step S103: recommending the official document to be recommended to the target user. In this step, the official document to be recommended can be pushed to the terminal corresponding to the user, thereby completing the official document recommendation.

[0091] Figure 3 FIG. 1 is a schematic diagram of specific implementation steps of the document recommendation method in an embodiment of the present invention. Figure 3 As shown, in an embodiment of the present invention, the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention can be calculated through five paths, and the comprehensive potential attention can be calculated based on the above potential attentions, and finally the official document recommendation is performed based on the comprehensive potential attention. When calculating the first potential attention, the official document association formula is first obtained, and the feature vector of the official document is calculated using a preset function or a feature extraction model, and then the first potential attention can be obtained by calculating the similarity or using an intention recognition model. When calculating the second potential attention, the method of calculating the similarity or using the intention discrimination model can be used. When calculating the third potential attention, the time penalty items of the two aspects can be combined. In addition, the fourth potential attention can be calculated using the number of official document views, and the fifth potential attention can be calculated using preset rules.

[0092] In the technical solution of the embodiment of the present invention, the association relationships such as revision, abolition, basis, and mention among multiple official documents are first extracted, and then the official documents to be recommended are determined based on such association relationships and the characteristic data of the target user, and finally the official documents to be recommended are recommended to the target user. Based on the above settings, the logical association relationships between official documents can be fully considered to determine the official documents that match the target user, thereby improving the accuracy of recommendation. When determining the official documents to be recommended based on the association relationship between official documents, the embodiment of the present invention can first calculate the feature vector of any official document, and then determine the first potential attention of the target user to the official document based on the feature vector of the official document and the characteristic data of the target user, and finally determine the official document to be recommended based on the first potential attention. When calculating the feature vector of any official document, the original features of the official document, the feature vectors of the official documents that have an association relationship with the official document, and the weight values ​​corresponding to the type of the association relationship can be input into a preset function for direct calculation, or the original features of the official document and the association relationship data between the official documents can be input into a pre-trained feature extraction model for calculation. When calculating the first potential attention, it can be based on the similarity between the document feature vector and the target user feature data, or it can be based on the probability of the target user's attention to the document obtained by inputting the document feature vector and the target user feature data into a pre-trained intention recognition model. Through the above settings, the document feature calculation applicable to a variety of specific scenarios and the matching of documents and users are realized. In an embodiment of the present invention, the feature data of the target user can be the current feature vector of the target user, and the current feature vector can dynamically change with the target user's browsing, collection, subscription and other behaviors for the document on the basis of the target user's initial feature vector. In this way, after the target user browses, collects or subscribes to a certain document, its current feature vector can change in a direction that tends to the feature vector of the document. This change makes the target user and the associated documents of the document form a higher degree of matching, so that accurate recommendation of associated documents and real-time recommendation can be achieved. In an embodiment of the present invention, in addition to considering the association relationship between documents, one or more of the following methods can also be combined for collaborative recommendation, which methods include: recommendation based on user behavior, recommendation based on time penalty items, recommendation based on the number of document browsing, and recommendation based on preset rules, which can further improve the accuracy of recommendation.

[0093] It should be noted that, for the convenience of description, the aforementioned method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the present invention is not limited by the described action sequence, and some steps can actually be performed in other sequences or simultaneously. In addition, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily necessary to implement the present invention.

[0094] In order to better implement the above-mentioned solution of the embodiment of the present invention, relevant devices for implementing the above-mentioned solution are also provided below.

[0095] See also Figure 4 As shown, the official document recommendation device 400 provided by the embodiment of the present invention may include: an information extraction unit 401, an official document determination unit 402 and a recommendation unit 403.

[0096] Among them, the information extraction unit 401 can be used to obtain the association relationship between multiple documents; the document determination unit 402 can be used to determine the documents to be recommended among the multiple documents based on the association relationship and the characteristic data of the target user; the recommendation unit 403 is used to recommend the documents to be recommended to the target user.

[0097] In an embodiment of the present invention, the association relationship between the multiple official documents includes: at least one official document that has an association relationship with any of the multiple official documents and the type of the association relationship; the official document determination unit 402 can be further used for: for any of the multiple official documents: determining the feature vector of any official document based on the official document that has an association relationship with the any official document and the type of the association relationship; determining the first potential attention of the target user to the any official document based on the feature vector of the any official document and the characteristic data of the target user; obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention; and determining the any official document as a document to be recommended when the comprehensive potential attention meets the preset recommendation conditions.

[0098] In some embodiments, the official document determination unit 402 can be further used to: for any official document among the multiple official documents: input the original features of the any official document, the feature vector of each official document that has an association relationship with the any official document, and the preset weight value corresponding to the type of the association relationship into a preset function for calculation to obtain the feature vector of the any official document.

[0099] Preferably, the official document determination unit 402 can be further used to: input the original features of the multiple official documents and the association relationship data between the multiple official documents into a pre-trained feature extraction model to obtain a feature vector of any official document among the multiple official documents.

[0100] In a specific application, the characteristic data of the target user includes the current characteristic vector of the target user. The device 400 may further include a target user characteristic calculation unit, which is used to: obtain the initial characteristic vector of the target user; determine the current characteristic vector of the target user based on the initial characteristic vector of the target user and the historical behavior data of the target user with respect to the official document.

[0101] In practical applications, the target user feature calculation unit can be further used to determine the initial feature vector of the target user according to the preferred document category provided by the target user.

[0102] As a preferred solution, the target user feature calculation unit can be further used to: obtain the feature vector of each official document corresponding to the historical behavior data; multiply the feature vector of any official document by the influence score of the official document to obtain the influence vector of the official document; wherein the influence score is determined based on the type of historical behavior of the target user with respect to the official document; determine the current feature vector of the target user based on the influence vector of each official document and the initial feature vector of the target user.

[0103] In a specific application scenario, the official document determination unit 402 can be further used to: for any official document among the multiple official documents, calculate the similarity between the feature vector of the any official document and the current feature vector of the target user; and determine the first potential attention of the target user to the any official document based on the similarity.

[0104] In some embodiments, the official document determination unit 402 can be further used to: input the feature vector of any official document among the multiple official documents and the current feature vector of the target user into a pre-trained intention recognition model to obtain the probability of the target user's attention to the any official document; and determine the first potential attention level of the target user to the any official document based on the attention probability.

[0105] In the embodiment of the present invention, the official document determining unit 402 may be further configured to: determine the first potential degree of interest of the target user to any official document among the multiple official documents as the comprehensive potential degree of interest of the target user to the any official document.

[0106] In an optional implementation, the device 400 may further include a second attention calculation unit, which is used to: for any document among the multiple documents, calculate the similarity between the original features of the document and the feature data of the target user; determine the second potential attention of the target user to the document based on the similarity; the document determination unit 402 may be further used to: determine the comprehensive potential attention of the target user to the document in combination with the first potential attention and the second potential attention.

[0107] In some embodiments, the second attention calculation unit can be further used to: input the original features of any official document among the multiple official documents and the comprehensive features of the target user into a pre-trained intention discrimination model to obtain the probability of the target user's attention to the any official document, and determine the second potential attention of the target user to the any official document based on the attention probability; wherein the comprehensive features of the target user are determined by the target user's historical behavior data for the official document and the target user's attribute features; the official document determination unit 402 can be further used to: determine the target user's comprehensive potential attention to the any official document in combination with the first potential attention and the second potential attention.

[0108] In an optional implementation, the device 400 may further include a third attention calculation unit, which is used to: for any official document among the multiple official documents, determine a first time penalty item according to the time between the release time of any official document and the current time, and use the first time penalty item to determine the third potential attention of the target user to the any official document. The official document determination unit 402 may be further used to: determine the comprehensive potential attention of the target user to the any official document by combining the first potential attention, the second potential attention and the third potential attention.

[0109] Preferably, the third attention calculation unit can be further used for: for any official document among the multiple official documents: when the target user has browsed any official document at a historical moment, determining the second time penalty item based on the length of time between the time when the target user last browsed any official document and the current moment; and using the first time penalty item and the second time penalty item to determine the target user's third potential attention to the any official document.

[0110] In an embodiment of the present invention, the device 400 may further include a fourth attention calculation unit, which is used to: for any document among the multiple documents, determine the fourth potential attention of the any document according to the number of views of the any document in at least one historical time interval; the document determination unit 402 may be further used to: determine the comprehensive potential attention of the target user to the any document in combination with the first potential attention, the second potential attention, the third potential attention and the fourth potential attention.

[0111] Preferably, the fourth attention degree calculation unit may be further used for: when any official document among the plurality of official documents is related to a predetermined hot event, increasing the fourth potential attention degree of the any official document according to a preset strategy.

[0112] As a preferred solution, the device 400 may further include a fifth attention calculation unit, which is used to: when any official document among the multiple official documents and the target user meet at least one preset rule, determine the fifth potential attention of any official document according to the recommendation weight set in the preset rule; the official document determination unit 402 may be further used to: determine the comprehensive potential attention of the target user to the any official document in combination with the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention.

[0113] Preferably, the fifth attention degree calculation unit can be further used for: when there is one preset rule, determining the recommendation weight set in the preset rule as the fifth potential attention degree of any official document; when there are multiple preset rules, determining the maximum value of the recommendation weights in the multiple preset rules as the fifth potential attention degree of any official document.

[0114] In an embodiment of the present invention, the official document determination unit 402 can be further used to: determine the weighted sum of the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention as the target user's comprehensive potential attention to any official document.

[0115] In one embodiment, the official document determination unit 402 may be further configured to: arrange the plurality of official documents in descending order of comprehensive potential attention; and determine a preset number of official documents arranged in front as official documents to be recommended.

[0116] In actual applications, the characteristic data of the target user includes: official documents that the target user has browsed, collected or subscribed to in a preset historical time period; the official document determination unit 402 can be further used to: determine the official documents among the multiple official documents that the target user has browsed, collected or subscribed to in the preset historical time period as target official documents; and determine at least one official document that has an association with the target official document as a document to be recommended.

[0117] In addition, in an embodiment of the present invention, the types of the association relationships include: revision relationship, cancellation relationship, based on relationship and mention relationship; the types of the historical behaviors include: browsing, collection, subscription, ignoring, uncollection and unsubscription.

[0118] In the technical solution of the embodiment of the present invention, the association relationships such as revision, abolition, basis, and mention among multiple official documents are first extracted, and then the official documents to be recommended are determined based on such association relationships and the characteristic data of the target user, and finally the official documents to be recommended are recommended to the target user. Based on the above settings, the logical association relationships between official documents can be fully considered to determine the official documents that match the target user, thereby improving the accuracy of recommendation. When determining the official documents to be recommended based on the association relationship between official documents, the embodiment of the present invention can first calculate the feature vector of any official document, and then determine the first potential attention of the target user to the official document based on the feature vector of the official document and the characteristic data of the target user, and finally determine the official document to be recommended based on the first potential attention. When calculating the feature vector of any official document, the original features of the official document, the feature vectors of the official documents that have an association relationship with the official document, and the weight values ​​corresponding to the type of the association relationship can be input into a preset function for direct calculation, or the original features of the official document and the association relationship data between the official documents can be input into a pre-trained feature extraction model for calculation. When calculating the first potential attention, it can be based on the similarity between the document feature vector and the target user feature data, or it can be based on the probability of the target user's attention to the document obtained by inputting the document feature vector and the target user feature data into a pre-trained intention recognition model. Through the above settings, the document feature calculation applicable to a variety of specific scenarios and the matching of documents and users are realized. In an embodiment of the present invention, the feature data of the target user can be the current feature vector of the target user, and the current feature vector can dynamically change with the target user's browsing, collection, subscription and other behaviors for the document on the basis of the target user's initial feature vector. In this way, after the target user browses, collects or subscribes to a certain document, its current feature vector can change in a direction that tends to the feature vector of the document. This change makes the target user and the associated documents of the document form a higher degree of matching, so that accurate recommendation of associated documents and real-time recommendation can be achieved. In an embodiment of the present invention, in addition to considering the association relationship between documents, one or more of the following methods can also be combined for collaborative recommendation, which methods include: recommendation based on user behavior, recommendation based on time penalty items, recommendation based on the number of document browsing, and recommendation based on preset rules, which can further improve the accuracy of recommendation.

[0119] Figure 5 An exemplary system architecture 500 is shown to which the official document recommendation method or official document recommendation device according to the embodiment of the present invention can be applied.

[0120] like Figure 5As shown, the system architecture 500 may include terminal devices 501, 502, 503, a network 504 and a server 505 (this architecture is only an example, and the components included in the specific architecture may be adjusted according to the specific application). The network 504 is used to provide a medium for communication links between the terminal devices 501, 502, 503 and the server 505. The network 504 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0121] Users can use terminal devices 501, 502, 503 to interact with server 505 through network 504 to receive or send messages, etc. Various client applications can be installed on terminal devices 501, 502, 503, such as official document reading applications, etc. (only as an example).

[0122] The terminal devices 501 , 502 , and 503 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0123] The server 505 may be a server that provides various services, such as a background server (only as an example) that provides support for the official document reading application operated by the user using the terminal devices 501, 502, 503. The background server may process the received official document recommendation request and feed back the processing result (such as the official document to be recommended - only as an example) to the terminal devices 501, 502, 503.

[0124] It should be noted that the official document recommendation method provided in the embodiment of the present invention is generally executed by the server 505 , and accordingly, the official document recommendation device is generally disposed in the server 505 .

[0125] It should be understood that Figure 5 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0126] The present invention also provides an electronic device. The electronic device of an embodiment of the present invention includes: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the document recommendation method provided by the present invention.

[0127] Reference below Figure 6 , which shows a schematic diagram of the structure of a computer system 600 of an electronic device suitable for implementing an embodiment of the present invention. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0128] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the computer system 600 are also stored. The CPU 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0129] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed so that a computer program read therefrom is installed into the storage section 608 as needed.

[0130] In particular, according to the embodiments disclosed in the present invention, the process described in the main step diagram above can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the main step diagram. In the above embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the central processing unit 601, the above functions defined in the system of the present invention are executed.

[0131] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0132] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0133] The units involved in the embodiments of the present invention may be implemented by software or hardware. The units described may also be provided in a processor, for example, may be described as: a processor including an information extraction unit, an official document determination unit, and a recommendation unit. The names of these units do not, in certain cases, constitute limitations on the units themselves, for example, the information extraction unit may also be described as a "unit that provides the official document determination unit with the relationship between official documents".

[0134] As another aspect, the present invention further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently and not be assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by the device, the steps executed by the device include: obtaining the association relationship between multiple documents; determining the documents to be recommended among the multiple documents according to the association relationship and the characteristic data of the target user; and recommending the documents to be recommended to the target user.

[0135] In the technical solution of the embodiment of the present invention, the association relationships such as revision, abolition, basis, and mention among multiple official documents are first extracted, and then the official documents to be recommended are determined based on such association relationships and the characteristic data of the target user, and finally the official documents to be recommended are recommended to the target user. Based on the above settings, the logical association relationships between official documents can be fully considered to determine the official documents that match the target user, thereby improving the accuracy of recommendation. When determining the official documents to be recommended based on the association relationship between official documents, the embodiment of the present invention can first calculate the feature vector of any official document, and then determine the first potential attention of the target user to the official document based on the feature vector of the official document and the characteristic data of the target user, and finally determine the official document to be recommended based on the first potential attention. When calculating the feature vector of any official document, the original features of the official document, the feature vectors of the official documents that have an association relationship with the official document, and the weight values ​​corresponding to the type of the association relationship can be input into a preset function for direct calculation, or the original features of the official document and the association relationship data between the official documents can be input into a pre-trained feature extraction model for calculation. When calculating the first potential attention, it can be based on the similarity between the document feature vector and the target user feature data, or it can be based on the probability of the target user's attention to the document obtained by inputting the document feature vector and the target user feature data into a pre-trained intention recognition model. Through the above settings, the document feature calculation applicable to a variety of specific scenarios and the matching of documents and users are realized. In an embodiment of the present invention, the feature data of the target user can be the current feature vector of the target user, and the current feature vector can dynamically change with the target user's browsing, collection, subscription and other behaviors for the document on the basis of the target user's initial feature vector. In this way, after the target user browses, collects or subscribes to a certain document, its current feature vector can change in a direction that tends to the feature vector of the document. This change makes the target user and the associated documents of the document form a higher degree of matching, so that accurate recommendation of associated documents and real-time recommendation can be achieved. In an embodiment of the present invention, in addition to considering the association relationship between documents, one or more of the following methods can also be combined for collaborative recommendation, which methods include: recommendation based on user behavior, recommendation based on time penalty items, recommendation based on the number of document browsing, and recommendation based on preset rules, which can further improve the accuracy of recommendation.

[0136] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for recommending official documents, characterized in that: include: Get the relationship between multiple documents; The association relationship between the multiple official documents includes: at least one official document having an association relationship with any official document among the multiple official documents and the type of the association relationship; Determining the official document to be recommended among the multiple official documents according to the association relationship and the characteristic data of the target user, including: for any official document among the multiple official documents: determining the characteristic vector of the any official document according to the official documents that have an association relationship with the any official document and the type of the association relationship; determining the first potential attention of the target user to the any official document according to the characteristic vector of the any official document and the characteristic data of the target user; obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention; when the comprehensive potential attention meets the preset recommendation condition, determining the any official document as the official document to be recommended; the characteristic data of the target user includes the current characteristic vector of the target user; The determining of the feature vector of any official document according to the official documents associated with the any official document and the type of the associated relationship comprises: for any official document among the plurality of official documents: inputting the original feature of the any official document, the feature vector of each official document associated with the any official document, and the preset weight value corresponding to the type of the associated relationship into a preset function for calculation, so as to obtain the feature vector of the any official document; Recommend the documents to be recommended to the target users; The method further comprises: determining an initial feature vector of the target user according to the preferred document category provided by the target user; determining a current feature vector of the target user according to the initial feature vector of the target user and the historical behavior data of the target user with respect to the document; The method of determining the current feature vector of the target user based on the initial feature vector of the target user and the historical behavior data of the target user with respect to the official document includes: obtaining the feature vector of each official document corresponding to the historical behavior data; multiplying the feature vector of any official document by the influence score of the official document to obtain the influence vector of the official document; wherein the influence score is determined according to the type of the historical behavior of the target user with respect to the official document; and determining the current feature vector of the target user based on the influence vector of each official document and the initial feature vector of the target user.

2. The method according to claim 1, characterized in that: The determining the first potential degree of attention of the target user to the any official document according to the feature vector of the any official document and the feature data of the target user comprises: For any official document among the multiple official documents, calculating the similarity between a feature vector of the official document and a current feature vector of the target user; The first potential degree of attention of the target user to any official document is determined based on the similarity.

3. The method according to claim 1, characterized in that The determining the first potential degree of attention of the target user to the any official document according to the feature vector of the any official document and the feature data of the target user comprises: Inputting a feature vector of any official document among the multiple official documents and a current feature vector of the target user into a pre-trained intention recognition model to obtain a probability of the target user paying attention to the any official document; The first potential attention degree of the target user to any official document is determined according to the attention probability.

4. The method according to claim 1, characterized in that: The step of obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention includes: The first potential attention degree of the target user to any official document among the multiple official documents is determined as the comprehensive potential attention degree of the target user to the any official document.

5. The method according to claim 1, characterized in that The method further comprises: for any official document among the plurality of official documents, calculating the similarity between the original feature of the official document and the feature data of the target user; determining the second potential attention of the target user to the official document according to the similarity; and The obtaining of the target user's comprehensive potential attention to the any official document based on the first potential attention includes: determining the target user's comprehensive potential attention to the any official document in combination with the first potential attention and the second potential attention.

6. The method according to claim 1, characterized in that The method further comprises: inputting the original features of any official document among the multiple official documents and the comprehensive features of the target user into a pre-trained intention discrimination model to obtain the probability of the target user paying attention to the official document, and determining the second potential attention degree of the target user to the official document according to the attention probability; wherein the comprehensive features of the target user are determined by the historical behavior data of the target user with respect to the official document and the attribute features of the target user; and, The obtaining of the target user's comprehensive potential attention to the any official document based on the first potential attention includes: determining the target user's comprehensive potential attention to the any official document in combination with the first potential attention and the second potential attention.

7. The method according to claim 5 or 6, characterized in that: The method further comprises: For any official document among the multiple official documents, determining a first time penalty item according to the time between the release time of the official document and the current time, and using the first time penalty item to determine a third potential attention degree of the target user to the official document; and The step of obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention includes: The comprehensive potential attention degree of the target user to the any official document is determined by combining the first potential attention degree, the second potential attention degree and the third potential attention degree.

8. The method according to claim 7, characterized in that The method further comprises: for any official document among the plurality of official documents: when the target user has browsed the official document at a historical moment, determining a second time penalty item according to the time between the last time the target user browsed the official document and the current moment; and The method of using the first time penalty item to determine the third potential attention degree of the target user to the any official document includes: using the first time penalty item and the second time penalty item to determine the third potential attention degree of the target user to the any official document.

9. The method according to claim 7, characterized in that: The method further comprises: for any official document among the plurality of official documents, determining a fourth potential attention degree of the official document according to the number of views of the official document in at least one historical time interval; and The obtaining of the target user's comprehensive potential attention to any official document based on the first potential attention includes: determining the target user's comprehensive potential attention to any official document by combining the first potential attention, the second potential attention, the third potential attention and the fourth potential attention.

10. The method according to claim 9, characterized in that The method further comprises: When any official document among the multiple official documents is related to a predetermined hot event, the fourth potential attention level of the official document is increased according to a preset strategy.

11. The method according to claim 9, characterized in that The method further comprises: when any official document among the plurality of official documents and the target user satisfy at least one preset rule, determining the fifth potential attention degree of the any official document according to the recommendation weight set in the preset rule; The method of obtaining the target user's comprehensive potential attention to any official document based on the first potential attention includes: determining the target user's comprehensive potential attention to any official document by combining the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention.

12. The method according to claim 11, characterized in that The determining the fifth potential attention degree of any official document according to the recommendation weight set in the preset rule includes: When there is only one preset rule, the recommendation weight set in the preset rule is determined as the fifth potential attention degree of any official document; When there are multiple preset rules, the maximum value of the recommendation weights among the multiple preset rules is determined as the fifth potential attention degree of any official document.

13. The method according to claim 11, characterized in that The step of obtaining the comprehensive potential attention of the target user to the any official document based on the first potential attention includes: The weighted sum of the first potential attention, the second potential attention, the third potential attention, the fourth potential attention and the fifth potential attention is determined as the comprehensive potential attention of the target user to the any official document.

14. The method according to claim 1, characterized in that When the comprehensive potential attention meets the preset recommendation conditions, the document is determined as a document to be recommended, including: Arrange the multiple official documents in descending order according to their comprehensive potential attention; A preset number of official documents arranged in front are determined as official documents to be recommended.

15. The method according to claim 1, characterized in that The characteristic data of the target user includes: the official documents that the target user has browsed, collected or subscribed to in a preset historical time period; and determining the official documents to be recommended among the multiple official documents according to the association relationship and the characteristic data of the target user includes: Determine the official documents among the plurality of official documents that the target user has browsed, collected or subscribed to in a preset historical time period as target official documents; At least one official document associated with the target official document is determined as the official document to be recommended.

16. The method according to claim 1, characterized in that The types of association relationships include: revision relationship, abolition relationship, based on relationship and reference relationship; The types of historical behaviors include: browsing, collecting, subscribing, ignoring, uncollecting, and unsubscribing.

17. A document recommendation device, characterized in that: include: An information extraction unit, used to obtain the association relationship between multiple official documents; The association relationship between the multiple official documents includes: at least one official document having an association relationship with any official document among the multiple official documents and the type of the association relationship; An official document determination unit, configured to determine an official document to be recommended among the plurality of official documents according to the association relationship and characteristic data of the target user; The official document determination unit is further used for: for any official document among the multiple official documents: determining a feature vector of the official document according to the official documents associated with the official document and the type of the associated relationship; determining a first potential attention of the target user to the official document according to the feature vector of the official document and the feature data of the target user; obtaining a comprehensive potential attention of the target user to the official document based on the first potential attention; and determining the official document as a document to be recommended when the comprehensive potential attention meets a preset recommendation condition; the feature data of the target user includes a current feature vector of the target user; The document determination unit is further used to: for any document among the multiple documents: input the original features of the document, the feature vector of each document associated with the document, and the preset weight value corresponding to the type of the association into a preset function for calculation to obtain the feature vector of the document; A recommendation unit, used to recommend the official document to be recommended to the target user; The target user feature calculation unit is used to determine the target user's initial feature vector according to the preferred document category provided by the target user; determine the target user's current feature vector according to the target user's initial feature vector and the target user's historical behavior data on the document; The target user feature calculation unit is also used to: obtain the feature vector of each official document corresponding to the historical behavior data; multiply the feature vector of any official document by the influence score of the official document to obtain the influence vector of the official document; wherein the influence score is determined according to the type of historical behavior of the target user with respect to the official document; determine the current feature vector of the target user according to the influence vector of each official document and the initial feature vector of the target user.

18. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 16.

19. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 16 is implemented.

Images