The method and device, the electronic device, and the storage medium recommended in the paper

Abstract

The application provides a paper recommendation method and device, an electronic device and a storage medium. The method comprises the following steps: obtaining a pre-constructed paper relationship graph and reading behavior data of a target user in a current preset time period; the paper relationship graph comprises relationships between multiple papers; inputting the paper relationship graph and the reading behavior data in the current preset time period into a pre-trained recommendation model to obtain a recommendation value of each paper in the paper relationship graph. The recommendation model is pre-trained by using historical reading behavior data of multiple users in a historical preset time period and the paper relationship graph. Therefore, the target user is recommended with papers based on the recommendation value of each paper. Thus, the user is no longer recommended with papers by calculating the similarity between each two papers, but is accurately recommended with each paper according to the reading behavior data of the user and the paper relationship graph data.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and device for paper recommendation, electronic equipment, and storage medium. Background Art

[0002] With the development of the information age, a large number of bachelor's theses in different fields have emerged. When users want to quickly find a bachelor's theses in a field they want to know about from a large number of bachelor's theses, the system will recommend a large number of bachelor's theses in that field to users.

[0003] The existing paper recommendation method recommends papers to users by calculating the similarity between each two papers. When a user views a paper, the system recommends target papers that are highly similar to the paper.

[0004] However, since the recommendations to users are based on the similarity of papers without considering the needs of users, the recommended papers to users are not the papers that users themselves want to know about, resulting in the inability to accurately recommend papers to users. Summary of the Invention

[0005] Based on the above-mentioned deficiencies in the prior art, the present application provides a method and device for recommending papers, an electronic device, and a storage medium to solve the problem of inaccurate paper recommendations in the prior art.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] The first aspect of this application provides a method for recommending a paper, comprising:

[0008] Obtaining a pre-built paper relationship map and target user's reading behavior data within a current preset time period; wherein the paper relationship map includes the relationship between multiple papers;

[0009] Inputting the paper relationship map and the reading behavior data within the current preset time period into a pre-trained recommendation model to obtain a recommendation value for each paper in the paper relationship map; wherein the recommendation model is pre-trained using the historical reading behavior data of multiple users within the historical preset time period and the paper relationship map;

[0010] Based on the recommendation value of each of the papers, the papers are recommended to the target user.

[0011] Optionally, in the above-mentioned paper recommendation method, it is characterized in that the training method of the recommendation model includes:

[0012] Obtaining the paper relationship graph and historical reading behavior data of multiple sample users;

[0013] For a plurality of the sample users, the paper relationship graph and the historical reading behavior data of the sample users are respectively input into the pre-trained recommendation model, and the recommendation value of each paper in the paper relationship graph is obtained by the recommendation model;

[0014] Determine whether the recommended values ​​of each paper meet the preset expected values;

[0015] If the recommendation values ​​of each of the papers meet the preset expected values, the recommendation model is determined as a trained recommendation model;

[0016] If the recommendation values ​​of each of the papers do not meet the preset expected values, the parameters of the recommendation model are adjusted, and the process of inputting the paper relationship map and the historical reading behavior data of the sample users into the pre-trained recommendation model is returned to obtain the recommendation value of each of the papers in the paper relationship map through the recommendation model.

[0017] Optionally, in the above-mentioned method for recommending papers, the method for pre-constructing the paper relationship map includes:

[0018] Obtain complete background information on the target paper;

[0019] Extracting key information of the target paper from the complete background information of the target paper; wherein the key information includes entity information and the relationship between each of the entities;

[0020] Constructing a graph of the key information to obtain the paper relationship graph;

[0021] Identify the target paper as the paper currently to be processed;

[0022] For each cited paper of the currently processed paper, determining whether the cited paper meets the dissemination conditions;

[0023] If it is determined whether the cited paper meets the dissemination conditions, key information of the cited paper is obtained;

[0024] Expanding the paper relationship map using key information of the cited papers;

[0025] The cited paper is determined as the current paper to be processed, and the process returns to execute each cited paper for the current paper to be processed to determine whether the cited paper meets the propagation conditions, until there is no cited paper that meets the propagation conditions, and the final paper relationship map data is obtained.

[0026] Optionally, in the above-mentioned paper recommendation method, obtaining complete background information of the target paper includes:

[0027] Obtaining partial information of the paper to be searched uploaded by the user; wherein the partial information at least includes the paper title and some background information;

[0028] Find similar papers with titles close to the paper title from the corresponding titles of all papers;

[0029] Based on the partial background information, searching for a target paper that meets a preset condition from each of the similar papers; wherein the preset condition refers to background information that is completely consistent with the partial background information;

[0030] Extract complete background information corresponding to the target paper from the target paper.

[0031] Optionally, in the above-mentioned paper recommendation method, the recommendation model includes an input layer, a sequence extraction layer, a cross-aggregation layer, and an output layer. Inputting the paper relationship graph and the historical reading behavior data of the sample user into the pre-trained recommendation model, and obtaining the recommendation value of each paper in the paper relationship graph through the recommendation model, includes:

[0032] Performing feature processing on the paper relationship graph and the historical reading behavior data of the sample user through the input layer to obtain a first feature vector corresponding to the paper relationship graph and a second feature vector corresponding to the historical reading behavior data;

[0033] Performing feature processing on the second feature vector through the sequence extraction layer to obtain a feature sequence matrix;

[0034] Splicing the feature sequence matrix and the first feature vector corresponding to the paper relationship map through a sequence extraction layer to obtain a spliced ​​feature vector;

[0035] Perform multi-order crossover on the concatenated feature vectors through a crossover aggregation layer to obtain crossover sequence features;

[0036] The cross-sequence features are processed through the output layer to obtain the recommendation value of each paper in the paper relationship map.

[0037] The second aspect of the present application provides a device for recommending a paper, comprising:

[0038] A first acquisition unit is configured to acquire a pre-constructed paper relationship map and the target user's reading behavior data within a current preset time period; wherein the paper relationship map includes the relationship between multiple papers;

[0039] A first input unit is configured to input the paper relationship map and the reading behavior data within the current preset time period into a pre-trained recommendation model to obtain a recommendation value for each paper in the paper relationship map; wherein the recommendation model is pre-trained using the historical reading behavior data of multiple users within the historical preset time period and the paper relationship map;

[0040] A recommendation unit is used to recommend the paper to the target user based on the recommendation value of each paper.

[0041] Optionally, the device recommended in the above paper further includes:

[0042] A second acquisition unit is used to acquire the paper relationship map and historical reading behavior data of multiple sample users;

[0043] A second input unit is configured to input the paper relationship graph and the historical reading behavior data of the sample users into the pre-trained recommendation model for each of the sample users, and obtain a recommendation value for each of the papers in the paper relationship graph through the recommendation model;

[0044] The first judgment unit is used to judge whether the recommendation value of each paper meets the preset expected value;

[0045] A first determining unit is configured to determine the recommendation model as a trained recommendation model if the recommendation value of each of the papers meets the preset expected value;

[0046] An adjustment unit is used to adjust the parameters of the recommendation model if the recommendation values ​​of each of the papers do not meet the preset expected values, and return to the second input unit to execute the input of the paper relationship map and the historical reading behavior data of the sample users into the pre-trained recommendation model, and obtain the recommendation value of each of the papers in the paper relationship map through the recommendation model.

[0047] Optionally, the device recommended in the above paper further includes:

[0048] A third acquisition unit is used to obtain complete background information of the target paper;

[0049] A first extraction unit is configured to extract key information of the target paper from the complete background information of the target paper; wherein the key information includes entity information and the relationship between each of the entities;

[0050] A graph construction unit, configured to construct a graph for the key information to obtain the paper relationship graph;

[0051] The second determining unit is used to determine the target paper as the current paper to be processed;

[0052] A second judgment unit is used to judge whether each cited paper of the current paper to be processed meets the dissemination condition;

[0053] a fourth acquiring unit, configured to acquire key information of the cited paper if it is determined whether the cited paper meets the dissemination condition;

[0054] An expansion unit, configured to expand the paper relationship map using key information of the cited papers;

[0055] The third determination unit is used to determine the cited paper as the current paper to be processed, and return to the second judgment unit to execute each cited paper of the current paper to be processed to determine whether the cited paper meets the propagation condition, until there is no cited paper that meets the propagation condition, thereby obtaining the final paper relationship map data.

[0056] Optionally, in the device recommended in the above paper, the third obtaining unit includes:

[0057] A fifth acquiring unit is configured to acquire partial information of the paper to be searched uploaded by the user; wherein the partial information includes at least the paper title and partial background information;

[0058] The first search unit is used to search for similar papers with titles close to the paper title from the titles corresponding to all papers;

[0059] A second search unit is configured to search, based on the partial background information, for a target paper that meets a preset condition from among the similar papers; wherein the preset condition refers to background information that is completely consistent with the partial background information;

[0060] The second extraction unit is used to extract the complete background information corresponding to the target paper from the target paper.

[0061] Optionally, in the above-mentioned device for paper recommendation, the recommendation model includes an input layer, a sequence extraction layer, a cross-aggregation layer, and an output layer, and the second input unit includes:

[0062] A first processing unit is configured to perform feature processing on the paper relationship graph and the historical reading behavior data of the sample user through the input layer to obtain a first feature vector corresponding to the paper relationship graph and a second feature vector corresponding to the historical reading behavior data;

[0063] a second processing unit, configured to perform feature processing on the second eigenvector through the sequence extraction layer to obtain a feature sequence matrix;

[0064] A splicing unit, configured to splice the feature sequence matrix and the first feature vector corresponding to the paper relationship map through a sequence extraction layer to obtain a spliced ​​feature vector;

[0065] A multi-order crossover unit, configured to perform multi-order crossover on the concatenated feature vectors through a crossover aggregation layer to obtain crossover sequence features;

[0066] The third processing unit is used to process the cross-sequence features through the output layer to obtain a recommendation value for each of the papers in the paper relationship map.

[0067] A third aspect of the present application provides an electronic device, including:

[0068] memory and processor;

[0069] Wherein, the memory is used to store programs;

[0070] The processor is used to execute the program, and when the program is executed, it is specifically used to implement the method recommended by the paper as described in any one of the above.

[0071] In a fourth aspect, the present application provides a computer storage medium for storing a computer program, which, when executed, is used to implement the method for recommending papers as described in any one of the above.

[0072] The present application provides a method for recommending papers, which obtains the user's reading behavior data and the graph data of paper relationships to achieve accurate recommendation of each paper to the user. By obtaining a pre-constructed paper relationship graph and the target user's reading behavior data within the current preset time period, wherein the paper relationship graph includes the relationship between multiple papers, and then inputting the paper relationship graph and the reading behavior data within the current preset time period into a pre-trained recommendation model, the recommendation value of each paper in the paper relationship graph is finally obtained. Since the recommendation model is pre-trained using the historical reading behavior data and paper relationship graph of multiple users within the historical preset time period. Therefore, based on the recommendation value of each paper, the paper is recommended to the target user. Therefore, the paper recommendation to the user is no longer achieved by calculating the similarity between each two papers, but based on the user's reading behavior data and the graph data of the paper relationship, each paper is accurately recommended to the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0074] Figure 1 A flowchart of a method for recommending a paper provided in an embodiment of the present application;

[0075] Figure 2 A flowchart of a method for pre-constructing a paper relationship map provided in an embodiment of the present application;

[0076] Figure 3 A flowchart of a method for obtaining complete background information of a paper provided in an embodiment of the present application;

[0077] Figure 4 A flowchart of a training method for a recommendation model provided in an embodiment of the present application;

[0078] Figure 5 A schematic diagram of a network structure of a recommendation model provided in an embodiment of the present application;

[0079] Figure 6 A flowchart of a method for obtaining a recommendation value of a paper provided in another embodiment of the present application;

[0080] Figure 7 A schematic structural diagram of a device recommended in a paper provided as another embodiment of the present application;

[0081] Figure 8 A schematic structural diagram of an electronic device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0082] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0083] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0084] This application embodiment provides a method for recommending papers, such as Figure 1 As shown, the specific steps include:

[0085] S101. Obtain a pre-built paper relationship map and target user's reading behavior data within a current preset time period.

[0086] The paper relationship graph includes the relationships between multiple papers, such as the citation relationships between multiple papers. For example, "Research Progress in Video-Based Face Recognition" may cite 12 papers, and these 12 citing papers may also cite other papers. The citation relationships between these papers can then be constructed to obtain a knowledge graph of paper citation relationships. The paper relationship graph can also include information about the papers themselves, such as their authors and writing dates.

[0087] It should be noted that the preset time period refers to a time period whose length is a preset threshold. Optionally, the preset threshold can be 12 hours, or other thresholds, which can be set according to needs.

[0088] The reading behavior data within the current preset time period may include click data and process data of reading papers within the corresponding time period. The click data may include the time and number of times a user reads a paper within the preset time period. For example, on March 18, the user spent two hours reading "Progress in Video-Based Face Recognition Research" and read it three times. The process data may include the time a user stays on a single page and the page number when reading a paper within the preset time period. For example, on March 17, the user stayed on page 15 of the paper while reading "Progress in Video-Based Face Recognition Research."

[0089] Optionally, a point-of-use method can be used to record user reading behavior data. That is, an interface for recording user reading behavior data can be reserved in the system. When the user's reading time and number of readings of a certain paper reach the preset expectations, the front-end will send a report that meets the requirements to the back-end through the interface provided by the system. Then, based on the user's login information, the user's reading behavior data of the paper is stored as key information in the user's corresponding record file in the form of a timestamp. Therefore, the corresponding implementation method of step S101 can obtain the target user's reading behavior data from the target user's corresponding record file.

[0090] Optionally, the present application embodiment provides a method for pre-constructing a paper relationship map, such as Figure 2 As shown, the following steps are included:

[0091] S201. Obtain complete background information of the target paper.

[0092] It should be noted that the target paper refers to the starting paper for which the paper relationship map is pre-constructed.

[0093] Optionally, the system can pre-cache each paper that the user has read locally, and then use the DOI of the target paper to detect whether there is a target file with the DOI of the target paper as the file name in the local cached files. If it is detected that there is a target file with the DOI of the target paper as the file name in the local cached files, the complete background information of the target paper is obtained from the target file. If it is detected that there is no target file with the DOI of the target paper as the file name in the local cached files, a search request can be sent to the paper database based on the DOI of the target paper. Correspondingly, the paper database will retrieve the paper feedback in the form of an xml file, and then the xml file needs to be parsed to obtain the search results of the target paper, so that the complete background information of the target paper can be obtained. It should also be noted that the DOI of the target paper refers to the paper identifier of the target paper.

[0094] Optionally, in another embodiment of the present application, the target paper refers to the paper to be searched uploaded by the user, so another specific implementation of step S201 is as follows: Figure 3 As shown, the following steps are included:

[0095] S301. Obtain partial information of the paper to be searched uploaded by the user.

[0096] It should be noted that since most of the paper information provided by major paper platforms is incomplete, the information of the papers to be verified uploaded by users is also very limited. Therefore, in the embodiment of this application, only partial information of the papers to be searched uploaded by users can be obtained.

[0097] Part of the information includes at least the paper title and some background information. It should also be noted that some background information may include the paper author, publisher, and publication date, etc.

[0098] S302. Search for similar papers with a title close to the paper title from the corresponding titles of all papers.

[0099] S303. Based on some background information, search for a target paper that meets preset conditions from similar papers.

[0100] The preset condition refers to background information that is completely consistent with the partial background information.

[0101] It should be noted that the paper titles uploaded by users may be incomplete or contain errors. Therefore, it is also necessary to screen out the target paper with completely consistent background information from the papers with similar titles based on the partial background information of the paper to be found. Therefore, the target paper is the paper that the user is actually looking for.

[0102] S304. Extract complete background information corresponding to the target paper from the target paper.

[0103] Specifically, the complete background information corresponding to the target paper is relevant information for writing the target paper, for example, the complete background information includes the title of the target paper, the author, the paper identifier and other relevant information.

[0104] S202. Extract key information of the target paper from the complete background information of the target paper.

[0105] Among them, key information includes entity information and the relationship between entities.

[0106] It's important to note that the complete background information includes a wealth of information about the target paper, but not all of it can be used to construct a knowledge graph. Therefore, it's necessary to extract data from the target paper's complete background information that can be used to build a knowledge graph. For example, the target paper's complete background information includes three key pieces of information: "paper title," "author," and "paper identifier," all of which are critical for building a knowledge graph.

[0107] S203. Construct a graph of key information to obtain a paper relationship graph.

[0108] It should be noted that the graph is constructed using key information. For example, a triplet method can be adopted, where the triplet consists of two entities and a relationship. The entity can be a paper or an author, and the relationship is the connection between the two entities, such as <Paper A, citation, Paper B> or <Paper A, author, author name>, etc., thereby obtaining a paper relationship graph.

[0109] S204. Determine the target paper as the current paper to be processed.

[0110] S205. For each cited paper of the current paper to be processed, determine whether the cited paper meets the dissemination conditions.

[0111] Optionally, the dissemination conditions can be set according to the user's needs. Of course, other conditions can also be set, and the specific settings can be based on needs. Therefore, if the user's need is to disseminate the cited paper, it is determined that the cited paper meets the dissemination conditions, and step S206 is executed. If the user's need is not to disseminate the cited paper, it is determined that the cited paper does not meet the dissemination conditions, and the paper relationship map in step S203 can be directly obtained.

[0112] S206. Obtain key information of cited papers.

[0113] S207. Use the key information of cited papers to expand the paper relationship map.

[0114] It should be noted that expansion refers to adding the key information of the cited papers to the paper relationship map, that is, connecting the papers with citation relationships.

[0115] S208. Determine the cited paper as the current paper to be processed.

[0116] It should be noted that after the key information of a citing paper has been used to expand the paper relationship graph, the citing paper must be identified as the current paper to be processed to avoid repeated expansion of the citing paper's key information, which would affect the graph expansion of the next citing paper. Furthermore, the process must return to step S205 until no more citing papers meet the propagation criteria, thus obtaining the final paper relationship graph data.

[0117] S102: Input the paper relationship map and the reading behavior data within the current preset time period into a pre-trained recommendation model to obtain a recommendation value for each paper in the paper relationship map.

[0118] Among them, the recommendation model is pre-trained using the historical reading behavior data of multiple users within a preset historical time period and the paper relationship map.

[0119] It should be noted that the historical preset time period refers to a time period in the past whose length is the historical preset threshold.

[0120] Optionally, the historical preset threshold may be 8 hours. Of course, it may also be other threshold times, which can be set specifically according to needs.

[0121] Optionally, the present application embodiment provides a training method for a recommendation model, such as Figure 4 As shown, the following steps are included:

[0122] S401. Obtain a paper relationship map and historical reading behavior data of multiple sample users.

[0123] It should be noted that the specific implementation of step S401 can refer to step S101 accordingly, and will not be repeated here.

[0124] S402: For multiple sample users, the paper relationship map and the historical reading behavior data of the sample users are respectively input into a pre-trained recommendation model, and the recommendation value of each paper in the paper relationship map is obtained through the recommendation model.

[0125] Optionally, the recommendation model can be constructed based on the neural network mechanism of Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM) and Gate Recurrent Unit (GRU). Therefore, in an embodiment of the present application, the recommendation model is trained based on the neural network mechanism to obtain the training results.

[0126] Optionally, in another embodiment of the present application, Figure 5 The recommendation model shown may include an input layer, a sequence extraction layer, a cross aggregation layer, and an output layer. Accordingly, in the embodiment of the present application, a specific implementation of step S402 is as follows: Figure 6 As shown, the following steps are included:

[0127] S601. Perform feature processing on the paper relationship map and the historical reading behavior data of the sample user through the input layer to obtain a first feature vector corresponding to the paper relationship map and a second feature vector corresponding to the historical reading behavior data.

[0128] Specifically, the paper relationship map and the historical reading behavior data of sample users are processed into vector form through the input layer, that is, each data item is represented by a corresponding numerical value.

[0129] S602: Perform feature processing on the second eigenvector through a sequence extraction layer to obtain a feature sequence matrix.

[0130] It's important to note that the sample user's historical reading behavior data also includes the sequential relationships between papers. Therefore, the sequence extraction layer can be used to extract features from the second eigenvector, extracting the sequential relationships between papers from the second eigenvector to generate a feature sequence matrix. For example, if a user first reads Paper A, then Paper B, and finally Paper C, the time series relationship between Paper A, Paper B, and Paper C can be derived based on the chronological order of reading.

[0131] S603. Concatenate the feature sequence matrix and the first feature vector corresponding to the paper relationship map through the sequence extraction layer to obtain a concatenated feature vector.

[0132] It should be noted that by splicing the correlation products between the paper relationship graphs onto the feature sequence matrix through the sequence extraction layer, the sequence relationship between the papers can be effectively improved and the impact of noise on serialization extraction can be reduced.

[0133] S604: Perform multi-order crossover on the concatenated feature vectors through a cross aggregation layer to obtain cross sequence features.

[0134] Specifically, the purpose of processing the concatenated feature vectors through the cross-aggregation layer is to obtain subsequence relationships in the paper sequence. For example, there are five papers in the sequence relationship: Paper A → Paper B → Paper C → Paper D → Paper E. Among them, Paper A and Paper C belong to the same field, and Paper B, Paper D, and Paper E belong to another field. Therefore, the cross-aggregation layer can be used to split the sequence relationship into Paper A → Paper C and Paper B → Paper D → Paper E, obtaining two subsequence relationships. It should also be noted that the concatenated vectors can be cross-linked at multiple levels by calculating formulas, such as:

[0135] The calculation formula is (a+b+c) 2 =a 2 +b 2 +c 2 +2ab+2ac+2bc

[0136] After moving the item, you can get Here, a, b, and c all represent the papers read by the user, and one letter corresponds to one paper. This allows the model to achieve linear computational complexity.

[0137] S605. Process the cross-sequence features through the output layer to obtain the recommendation value of each paper in the paper relationship map.

[0138] S403. Determine whether the recommendation value of each paper meets the preset expected value.

[0139] Optionally, the recommendation value of each paper in the paper relationship map obtained by the recommendation model in step S403 may not meet the user's needs for reading these papers, so the recommendation value of each paper needs to be further judged. Therefore, after obtaining the recommendation value of each paper in the paper relationship map, it will be judged whether the recommendation value of each paper meets the preset expected value. If the recommendation value of each paper meets the preset expected value, it means that the recommendation value of each paper meets the user's needs for reading papers, and then step S404 is executed. If the recommendation value of each paper does not meet the preset expected value, it means that the recommendation model needs to be iteratively trained, and then step S405 is executed.

[0140] S404: Determine the recommendation model as the trained recommendation model.

[0141] S405: Adjust the parameters of the recommendation model.

[0142] It should be noted that, when the recommendation values ​​of each paper do not meet the preset expected values, it is necessary to return to step S402 and execute until the recommendation values ​​of each paper meet the preset expected values.

[0143] S103. Recommend papers to target users based on the recommendation value of each paper.

[0144] Optionally, each paper can be ranked in descending order of recommendation value and recommended to the user in that order. Alternatively, only the top N papers can be recommended to the user in that order. Of course, other recommendation strategies can also be used based on recommendation value, and the specific ones can be set according to needs.

[0145] The present application provides a method for recommending papers, which obtains the user's reading behavior data and the graph data of paper relationships to achieve accurate recommendation of each paper to the user. By obtaining a pre-constructed paper relationship graph and the target user's reading behavior data within the current preset time period, wherein the paper relationship graph includes the relationship between multiple papers, and then inputting the paper relationship graph and the reading behavior data within the current preset time period into a pre-trained recommendation model, the recommendation value of each paper in the paper relationship graph is finally obtained. Since the recommendation model is pre-trained using the historical reading behavior data and paper relationship graph of multiple users within the historical preset time period. Therefore, based on the recommendation value of each paper, the paper is recommended to the target user. Therefore, the paper recommendation to the user is no longer achieved by calculating the similarity between each two papers, but based on the user's reading behavior data and the graph data of the paper relationship, each paper is accurately recommended to the user.

[0146] Another embodiment of the present application provides a device for recommending papers, such as Figure 7 As shown, including:

[0147] The first acquisition unit 701 is used to acquire a pre-built paper relationship map and the target user's reading behavior data within a current preset time period.

[0148] Among them, the paper relationship map includes the relationship between multiple papers.

[0149] The first input unit 702 is used to input the paper relationship map and the reading behavior data in the current preset time period into a pre-trained recommendation model to obtain a recommendation value for each paper in the paper relationship map.

[0150] Among them, the recommendation model is pre-trained using the historical reading behavior data of multiple users within a preset historical time period and the paper relationship map.

[0151] The recommendation unit 703 is used to recommend papers to target users based on the recommendation value of each paper.

[0152] It should be noted that the specific working process of the above-mentioned units in the embodiment of the present application can refer to steps S101 to S103 in the above-mentioned method embodiment, and will not be repeated here.

[0153] Optionally, in another embodiment of the present application, the following units are further included:

[0154] The second acquisition unit is used to obtain the paper relationship map and historical reading behavior data of multiple sample users.

[0155] The second input unit is used to input the paper relationship map and the historical reading behavior data of the sample users into a pre-trained recommendation model for multiple sample users, and obtain the recommendation value of each paper in the paper relationship map through the recommendation model.

[0156] The first judgment unit is used to judge whether the recommendation value of each paper meets the preset expected value.

[0157] The first determining unit is configured to determine the recommendation model as a trained recommendation model if the recommendation values ​​of each paper meet a preset expected value.

[0158] The adjustment unit is used to adjust the parameters of the recommendation model if the recommendation values ​​of each paper do not meet the preset expected values, return to the second input unit to input the paper relationship map and the historical reading behavior data of the sample user into the pre-trained recommendation model, and obtain the recommendation value of each paper in the paper relationship map through the recommendation model.

[0159] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the corresponding steps in the above method embodiments, and will not be repeated here.

[0160] Optionally, in another embodiment of the present application, the following units are further included:

[0161] The third acquisition unit is used to obtain complete background information of the target paper.

[0162] The first extraction unit is used to extract the key information of the target paper from the complete background information of the target paper.

[0163] Among them, key information includes entity information and the relationship between entities.

[0164] The graph construction unit is used to construct a graph of key information to obtain a paper relationship graph.

[0165] The second determining unit is used to determine the target paper as the current paper to be processed.

[0166] The second judgment unit is used to judge whether each cited paper of the current paper to be processed meets the dissemination conditions.

[0167] The fourth acquisition unit is used to obtain key information of the cited paper if it is determined whether the cited paper meets the dissemination conditions.

[0168] The extension unit is used to expand the paper relationship map using the key information of the cited papers.

[0169] The third determination unit is used to determine the cited paper as the current paper to be processed, and return to the second judgment unit to execute each cited paper of the current paper to be processed to determine whether the cited paper meets the propagation conditions, until there is no cited paper that meets the propagation conditions, and obtain the final paper relationship map data.

[0170] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the corresponding steps in the above method embodiments, and will not be repeated here.

[0171] Optionally, in another embodiment of the present application, the third acquiring unit includes:

[0172] The fifth acquisition unit is used to acquire partial information of the paper to be searched uploaded by the user.

[0173] Among them, some information includes at least the title of the paper and some background information.

[0174] The first search unit is used to search for similar papers with a title close to the paper title from the titles corresponding to all papers.

[0175] The second search unit is used to search for a target paper that meets preset conditions from various similar papers based on partial background information.

[0176] The preset condition refers to background information that is completely consistent with the partial background information.

[0177] The second extraction unit is used to extract the complete background information corresponding to the target paper from the target paper.

[0178] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the corresponding steps in the above method embodiments, and will not be repeated here.

[0179] Optionally, in another embodiment of the present application, the recommendation model includes an input layer, a sequence extraction layer, a cross-aggregation layer, and an output layer, and the second input unit includes:

[0180] The first processing unit is used to perform feature processing on the paper relationship map and the historical reading behavior data of the sample user through the input layer to obtain a first feature vector corresponding to the paper relationship map and a second feature vector corresponding to the historical reading behavior data.

[0181] The second processing unit is configured to perform feature processing on the second eigenvector through a sequence extraction layer to obtain a feature sequence matrix.

[0182] The splicing unit is used to splice the feature sequence matrix and the first feature vector corresponding to the paper relationship map through the sequence extraction layer to obtain a spliced ​​feature vector.

[0183] The multi-order cross unit is used to perform multi-order crossover on the concatenated feature vector through the cross aggregation layer to obtain cross sequence features.

[0184] The third processing unit is used to process the cross-sequence features through the output layer to obtain the recommendation value of each paper in the paper relationship map.

[0185] It should be noted that the specific working process of each unit provided in the above embodiments of the present application can refer to the corresponding steps in the above method embodiments, and will not be repeated here.

[0186] Another embodiment of the present application provides an electronic device, such as Figure 8 As shown, including:

[0187] Memory 801 and processor 802 .

[0188] The memory 801 is used to store programs.

[0189] The processor 802 is used to execute a program. When the program is executed, it is specifically used to implement the method for recommending papers provided in any of the above embodiments.

[0190] Another embodiment of the present application provides a computer storage medium for storing a computer program. When the computer program is executed, it is used to implement the method for paper recommendation provided in any of the above embodiments.

[0191] Computer storage media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0192] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0193] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for recommending papers, characterized in that: include: Obtaining a pre-constructed paper relationship map and the target user's reading behavior data within a current preset time period; wherein the paper relationship map includes the relationships between multiple papers, and the reading behavior data includes click data and process data of reading papers within the preset time period, wherein the click data includes the time and number of times the user reads the paper within the preset time period, and the process data includes the time the user stays on a single page when reading the paper within the preset time period and the page number on which the user stays; Inputting the paper relationship map and the reading behavior data within the current preset time period into a pre-trained recommendation model to obtain a recommendation value for each paper in the paper relationship map; wherein the recommendation model is pre-trained using the historical reading behavior data of multiple users within the historical preset time period and the paper relationship map; Recommending the papers to the target user based on the recommendation value of each paper; The pre-built method of the paper relationship map includes: Obtain complete background information on the target paper; Extracting key information of the target paper from the complete background information of the target paper; wherein the key information includes entity information and the relationship between each of the entities; Constructing a graph of the key information to obtain the paper relationship graph; Identify the target paper as the paper currently to be processed; For each cited paper of the currently processed paper, determining whether the cited paper meets the dissemination conditions; If it is determined whether the cited paper meets the dissemination conditions, key information of the cited paper is obtained; Expanding the paper relationship map using key information of the cited papers; The cited paper is determined as the current paper to be processed, and the process returns to execute each cited paper for the current paper to be processed to determine whether the cited paper meets the propagation conditions, until there is no cited paper that meets the propagation conditions, and the final paper relationship map data is obtained.

2. The method according to claim 1, characterized in that The training method of the recommendation model includes: Obtaining the paper relationship graph and historical reading behavior data of multiple sample users; For a plurality of the sample users, the paper relationship graph and the historical reading behavior data of the sample users are respectively input into the pre-trained recommendation model, and the recommendation value of each paper in the paper relationship graph is obtained by the recommendation model; Determine whether the recommended values ​​of each paper meet the preset expected values; If the recommendation values ​​of each of the papers meet the preset expected values, the recommendation model is determined as a trained recommendation model; If the recommendation values ​​of each of the papers do not meet the preset expected values, the parameters of the recommendation model are adjusted, and the process of inputting the paper relationship map and the historical reading behavior data of the sample users into the pre-trained recommendation model is returned to obtain the recommendation value of each of the papers in the paper relationship map through the recommendation model.

3. The method according to claim 1, characterized in that The complete background information of the target paper is obtained, including: Obtaining partial information of the paper to be searched uploaded by the user; wherein the partial information at least includes the paper title and some background information; Find similar papers with titles close to the paper title from the corresponding titles of all papers; Based on the partial background information, searching for a target paper that meets a preset condition from each of the similar papers; wherein the preset condition refers to background information that is completely consistent with the partial background information; Extract complete background information corresponding to the target paper from the target paper.

4. The method according to claim 2, characterized in that The recommendation model includes an input layer, a sequence extraction layer, a cross-aggregation layer, and an output layer. The paper relationship graph and the historical reading behavior data of the sample users are input into the pre-trained recommendation model, and the recommendation value of each paper in the paper relationship graph is obtained by the recommendation model, including: Performing feature processing on the paper relationship graph and the historical reading behavior data of the sample user through the input layer to obtain a first feature vector corresponding to the paper relationship graph and a second feature vector corresponding to the historical reading behavior data; Performing feature processing on the second feature vector through the sequence extraction layer to obtain a feature sequence matrix; Splicing the feature sequence matrix and the first feature vector corresponding to the paper relationship map through a sequence extraction layer to obtain a spliced ​​feature vector; Perform multi-order crossover on the concatenated feature vectors through a crossover aggregation layer to obtain crossover sequence features; The cross-sequence features are processed through the output layer to obtain the recommendation value of each paper in the paper relationship map.

5. A device for recommending papers, characterized in that: include: A first acquisition unit is configured to acquire a pre-constructed paper relationship map and the target user's reading behavior data within a current preset time period; wherein the paper relationship map includes the relationships between multiple papers, and the reading behavior data includes click data and process data of reading papers within the preset time period, wherein the click data includes the time and number of times the user reads the papers within the preset time period, and the process data includes the time the user stays on a single page when reading the papers within the preset time period and the page number on which the user stays; A first input unit is configured to input the paper relationship map and the reading behavior data within the current preset time period into a pre-trained recommendation model to obtain a recommendation value for each paper in the paper relationship map; wherein the recommendation model is pre-trained using the historical reading behavior data of multiple users within the historical preset time period and the paper relationship map; A recommendation unit, configured to recommend the papers to the target user based on the recommendation value of each paper; The third acquisition unit is used to obtain complete background information of the target paper; A first extraction unit is configured to extract key information of the target paper from the complete background information of the target paper; wherein the key information includes entity information and the relationship between each of the entities; A graph construction unit, configured to construct a graph for the key information to obtain the paper relationship graph; The second determining unit is used to determine the target paper as the current paper to be processed; A second judgment unit is used to judge whether each cited paper of the current paper to be processed meets the dissemination condition; a fourth acquiring unit, configured to acquire key information of the cited paper if it is determined whether the cited paper meets the dissemination condition; An expansion unit, configured to expand the paper relationship map using key information of the cited papers; The third determination unit is used to determine the cited paper as the current paper to be processed, and return to execute each cited paper for the current paper to be processed to determine whether the cited paper meets the propagation condition, until there is no cited paper that meets the propagation condition, and obtain the final paper relationship map data.

6. The device according to claim 5, characterized in that Also includes: A second acquisition unit is used to acquire the paper relationship map and historical reading behavior data of multiple sample users; A second input unit is configured to input the paper relationship graph and the historical reading behavior data of the sample users into the pre-trained recommendation model for each of the sample users, and obtain a recommendation value for each of the papers in the paper relationship graph through the recommendation model; The first judgment unit is used to judge whether the recommendation value of each paper meets the preset expected value; A first determining unit is configured to determine the recommendation model as a trained recommendation model if the recommendation value of each of the papers meets the preset expected value; An adjustment unit is used to adjust the parameters of the recommendation model if the recommendation values ​​of each of the papers do not meet the preset expected values, and return to executing the step of inputting the paper relationship map and the historical reading behavior data of the sample users into the pre-trained recommendation model, and obtaining the recommendation value of each of the papers in the paper relationship map through the recommendation model.

7. An electronic device, characterized in that: include: memory and processor; Wherein, the memory is used to store programs; The processor is used to execute the program, and when the program is executed, it is specifically used to implement the method for paper recommendation according to any one of claims 1 to 4.

8. A computer storage medium, characterized in that Used to store a computer program, which, when executed, is used to implement the method for paper recommendation as described in any one of claims 1 to 4.

Images