Paper recommendation method and system based on cross-community knowledge graph in hyperbolic space

Abstract

The invention relates to the technical field of thesis recommendation, in particular to a thesis recommendation method and system based on a cross-community knowledge graph in hyperbolic space, and the method comprises the steps: extracting an initial structure feature vector and an initial text feature vector of each thesis node of the knowledge graph from an input cross-community academic knowledge graph; performing hierarchical aggregation on the initial structural feature vector, performing similarity aggregation and upward aggregation on a hierarchical aggregation result, and performing feature fusion on the similarity aggregation result and the upward aggregation result to obtain a final structural feature vector; performing text aggregation operation on the initial text feature vector to obtain a final text feature vector; carrying out splicing operation on the final structure and text feature vectors to obtain a final feature vector of each node; and calculating the similarity between the feature vector of the query task of the to-be-recommended paper and the feature vector of each paper node in the cross-community academic knowledge graph, and outputting a paper recommendation result according to the similarity. And the performance of the thesis recommendation task is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of paper recommendation, and in particular to a paper recommendation method and system based on a cross-community knowledge graph in a hyperbolic space. Background Art

[0002] The purpose of paper recommendation is to recommend high-quality academic papers that are the same or related to their research direction to researchers. However, with the rapid development of scientific research, a large amount of literature materials (such as published papers) has brought serious information overload problems to researchers. This makes researchers sometimes need to spend a lot of time looking for papers in their own related directions. For beginners who have just entered a certain research field, it is even more difficult to retrieve high-quality papers through keywords on literature retrieval websites because they do not understand their own research direction, the large number of literature materials, and the uneven quality of literature. This makes them spend a lot of unnecessary time on paper screening. Therefore, how to analyze the massive literature data on the Internet and build a paper recommendation model to quickly recommend high-quality papers to researchers has become an important research issue in current paper recommendation.

[0003] Traditional paper recommendation methods are mainly based on query text and graph structure. Among them, the query text-based method mainly defines user features by using text information such as query text, user history, and user information. Similarly, the text features of the paper manuscript are extracted, and the similarity between the two is further calculated to recommend papers. However, sometimes the article you want to read cannot be retrieved well by query text alone, and once user information such as user preferences or history records is involved, it will face the problem of difficulty in obtaining user information. Therefore, relying solely on query text for paper recommendation has certain limitations. The paper recommendation method based on graph structure uses the meaningful relationship and semantics between one or more entities to better learn the relationship between nodes such as query authors, query texts, and papers. However, most of the current graph-based paper recommendation methods mainly learn the correlation between vertices and recommend papers by constructing heterogeneous graphs between users, query texts, papers, and paper authors, or recommend similar papers based on the citation link network of papers.

[0004] In recent years, the Knowledge Graph (KG) has demonstrated impressive capabilities in alleviating the cold start problem and improving the interpretability of recommendations, so many recommendation methods based on the Knowledge Graph have emerged. However, the inventors found that information related to paper recommendations may be hidden in different communities, and existing methods do not integrate this information, that is, lack cross-community knowledge fusion. In addition, it is not easy to accurately represent the nodes in the knowledge graph because of its complex topological structure. Especially in the case of cross-community knowledge graphs, feature aggregation in traditional Euclidean space is prone to node distortion. Summary of the invention

[0005] In order to address the shortcomings of the prior art, the present invention provides a paper recommendation method and system based on a cross-community knowledge graph in a hyperbolic space, which integrates paper information between communities with different structures and adopts a knowledge aggregation scheme in a hyperbolic space to reduce the node embedding distortion problem, thereby improving the performance of the paper recommendation task.

[0006] On the one hand, a paper recommendation method based on a cross-community knowledge graph in a hyperbolic space is provided, including: obtaining a paper query task to be recommended and a cross-community academic knowledge graph; inputting the paper query task to be recommended and the cross-community academic knowledge graph into a trained paper recommendation model, and the paper recommendation model outputs recommended papers; Among them, the trained paper recommendation model is used to extract the initial structural feature vector and initial text feature vector of each paper node in the input cross-community academic knowledge graph, perform hierarchical aggregation on the initial structural feature vector, perform similarity aggregation and upward aggregation on the hierarchical aggregation results respectively, and perform feature fusion on the similarity aggregation and upward aggregation results to obtain the final structural feature vector; perform text aggregation operation on the initial text feature vector to obtain the final text feature vector; perform splicing operation on the final structural feature vector and the final text feature vector to obtain the final feature vector of each node in the cross-community academic knowledge graph; calculate the similarity between the feature vector of the query task of the paper to be recommended and the feature vector of each paper node in the cross-community academic knowledge graph, and output the paper recommendation result according to the similarity.

[0007] On the other hand, a paper recommendation system based on a cross-community knowledge graph in a hyperbolic space is provided, including: an acquisition module, which is configured to: acquire a paper query task to be recommended and a cross-community academic knowledge graph; a recommendation module, which is configured to: input the paper query task to be recommended and the cross-community academic knowledge graph into a trained paper recommendation model, and the paper recommendation model outputs recommended papers; wherein the trained paper recommendation model is used to extract the initial structural feature vector and the initial text feature vector of each paper node in the input cross-community academic knowledge graph, perform hierarchical aggregation on the initial structural feature vector, perform similarity aggregation and upward aggregation on the hierarchical aggregation results respectively, perform feature fusion on the similarity aggregation and upward aggregation results, and obtain the final structural feature vector; perform text aggregation operation on the initial text feature vector to obtain the final text feature vector; perform splicing operation on the final structural feature vector and the final text feature vector to obtain the final feature vector of each node in the cross-community academic knowledge graph; calculate the similarity between the feature vector of the paper query task to be recommended and the feature vector of each paper node in the cross-community academic knowledge graph, and output the paper recommendation result according to the similarity.

[0008] The above technical solution has the following advantages or beneficial effects: (1) The present invention collects and analyzes paper-related information in academic communities with different structures to explore the correlation between paper information in different communities, thereby performing cross-community knowledge fusion to further learn the feature representation of papers and make better paper recommendations for specific tasks.

[0009] (2) This paper is based on the constructed cross-community academic knowledge graph data. It adopts the knowledge graph method to learn the structural features of task nodes and paper nodes, and adopts a hierarchical aggregation scheme for different structural communities to better learn the knowledge between different communities. It adopts LightGCN to learn the text feature representation of nodes, and further integrates the structural features and text features to obtain richer node feature representation, thereby improving the performance of the paper recommendation task.

[0010] (3) This paper adopts a knowledge propagation strategy based on hyperbolic space to solve the distortion problem of knowledge graph node aggregation in traditional Euclidean space. By taking advantage of the low distortion of hyperbolic space when embedding scale-free and hierarchical graphs, it can better learn the representation of knowledge graph nodes. In addition, it introduces an upward aggregation method for the hierarchical relationship between tasks, thereby capturing a more fine-grained representation of the upper-level tasks. This improves the performance of paper recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0012] Figure 1 This is a flow chart of the method of embodiment 1. DETAILED DESCRIPTION

[0013] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0014] Embodiment 1, as Figure 1 As shown, this embodiment provides a paper recommendation method based on a cross-community knowledge graph in a hyperbolic space, including: S101: obtaining a paper query task to be recommended and a cross-community academic knowledge graph; S102: inputting the paper query task to be recommended and the cross-community academic knowledge graph into a trained paper recommendation model, and the paper recommendation model outputs recommended papers; wherein the trained paper recommendation model is used to extract the initial structural feature vector and the initial text feature vector of each paper node in the input cross-community academic knowledge graph, hierarchically aggregate the initial structural feature vector, perform similarity aggregation and upward aggregation on the hierarchical aggregation results respectively, and perform feature fusion on the similarity aggregation and upward aggregation results to obtain a final structural feature vector; perform text aggregation operation on the initial text feature vector to obtain a final text feature vector; perform splicing operation on the final structural feature vector and the final text feature vector to obtain a final feature vector of each node in the cross-community academic knowledge graph; calculate the similarity between the feature vector of the paper query task to be recommended and the feature vector of each paper node in the cross-community academic knowledge graph, and output the paper recommendation result according to the similarity.

[0015] Furthermore, the S101: obtains the paper query task to be recommended and the cross-community academic knowledge graph, wherein the paper query task to be recommended, for example, is a machine learning task or a machine vision task.

[0016] Furthermore, the specific construction process of the cross-community academic knowledge graph includes: obtaining paper-related information from multiple academic communities, preprocessing the paper-related information, and constructing a cross-community academic knowledge graph based on the association relationship between papers in different academic communities.

[0017] During the construction process, each paper is regarded as a node of the knowledge graph, and each paper query task is also regarded as a node of the knowledge graph. If there is a relationship between papers, such as using the same data set, the paper nodes are connected to each other through the same data set as attribute nodes. If there is a relationship between the paper and the paper query task, there are interconnected edges between the paper nodes and the paper query task nodes.

[0018] The multiple academic communities refer to multiple paper databases, such as Wanfang Database and CNKI Database; the paper-related information includes: paper title, paper author, paper abstract and paper search terms. Based on the obtained paper attribute information between different academic communities, and through the associated entities between different academic communities (the paper attribute information shared by two academic communities), a cross-community academic knowledge graph is constructed. The academic community, for example, refers to the CSDN community or the GITHUB community, and the academic community refers to an academic website.

[0019] Furthermore, the input cross-community academic knowledge graph extracts the initial structural feature vector and the initial text feature vector of each paper node in the knowledge graph, wherein the extraction process of the initial structural feature vector includes: obtaining paper nodes, paper query task nodes and related paper attribute information from the constructed cross-community academic knowledge graph, and embedding them through the structural feature embedding layer. , using a linear layer as layer, and obtain the node structure feature vector in Euclidean space : ,in, is the representation of the corresponding node in the knowledge graph.

[0020] The node structure feature vector obtained in the Euclidean space is mapped to the Lorentz manifold, and the negative curvature is of The Lorentz manifold is denoted by ,make and represents the Euclidean embedding and the transformed hyperbolic embedding, is mapped as follows: ;in, For the tangent space dimensional vector, Defined as The origin in is used as a reference vector to perform tangent space operations, for The second norm of is the exponential mapping function, and are the hyperbolic sine and cosine functions respectively.

[0021] It should be understood that based on the node structure feature vector in the Euclidean space, the exponential mapping function Map it to the Lorenz manifold to better disseminate knowledge across hierarchical data such as community knowledge graphs.

[0022] The reason why the Lorentz model is chosen is that it is a typical equivalent model that can well describe hyperbolic geometry and has unique simplicity and numerical stability. of The Lorentz manifold is denoted by , with vector centered The dimensional Euclidean tangent space is denoted by . yes It is a local first-order approximation to the Lorentz manifold, which is useful for performing graph convolution operations in hyperbolic space.

[0023] Furthermore, the hyperbolic space and the tangent space are mapped by exponential and logarithmic mapping to convert.

[0024] Given , .

[0025] Through index mapping Will Mapping to hyperbolic space, and vice versa, can be done by logarithmic mapping Will Mapped into tangent space centered at .

[0026] ; ; in, for The norm of is the curvature-aware inner product in hyperbolic space, Hyperbolic space middle The measured distance between two points is calculated as follows: ; .

[0027] Furthermore, the input cross-community academic knowledge graph extracts the initial structural feature vector and the initial text feature vector of each node of the knowledge graph, wherein the initial text feature vector is extracted using the text pre-training model BGE, which can convert words, phrases or entire sentences into vector form, which is used to extract the initial text feature vector of each node of the knowledge graph.

[0028] It should be understood that BGE is an open source Chinese and English semantic vector model. The BGE model is pre-trained on Chinese and English datasets using the RetroMAE method. The English datasets are from Pile, Wikipedia, and msmarco, while the Chinese dataset is from wudao. 24 A100s (40G) were used for training. Furthermore, the hierarchical aggregation of the initial structural feature vector includes: (1) by considering the triple The structural information of the nodes is used to learn the attention weights of the nodes and edges between the nodes connected by the relationship : ;in is the relationship weight matrix, and They are and Embedded in hyperbolic space, and using The function maps it to the tangent space. Then, by using the softmax function, The connected edges are further normalized to obtain : .

[0029] (2) Based on the neighbor node weight calculation formula, firstly, the knowledge propagation process of neighbor node information is carried out on the task node, and For task nodes exist The first-order neighbor set in the task node is composed only of paper nodes. The embedding of neighbor information can be defined as follows: ;in, For the corresponding side The normalized attention weights of and They are and Embedding in hyperbolic space;.

[0030] (3) Similarly, for the paper node Perform similar neighbor information aggregation operations, and let For thesis node The first-order task node neighbor set, For thesis node exist The first-order other entity neighbor set in the paper node The embedding of neighbor information is defined as: .

[0031] (4) In order to further obtain higher-order neighbor information in the cross-community academic knowledge graph, multi-hop operations are performed in the knowledge graph to conduct Neighbor sampling, let represent The The set of entity neighbors that are jumped; then, through Medium Entity Neighbors to update Entities in Embedding ; in, For the corresponding side The normalized attention weights of and Entity and Embedding representation in hyperbolic space.

[0032] (5) There may be associated entities in , defined as ,in is a collection of related entities; Entities in When embedding neighbor information, first consider whether there is an associated entity in its neighbor set. ; If there is an associated entity, first Related entities Perform neighbor aggregation and then obtain Related entities of medium and high-level knowledge Put into To further Neighborhood information embedding learning; exist Related entities When acquiring high-order neighbor information, different neighbor sampling sizes and aggregation layers are used; represent middle No. The set of neighbors of the entity that jumped off; .

[0033] (6) After obtaining the multi-hop neighbor information embedding, it is used to update the embedding representation of the task node and the paper node. The Sum Aggregator method is used to sum the embedding of the target node itself and the neighbor information embedding, perform a nonlinear transformation, and then perform nonlinear activation in the hyperbolic space to obtain the updated embedding representation.

[0034] ; in and are trainable weights and biases defined in the relevant tangent space, and is a linear transformation in hyperbolic space, is a hyperbolic activation function, which is defined as: ; ; ; in, , and is an exponential function, and is a logarithmic function, is the hyperbolic tangent activation function; So far, we have obtained For thesis node The final structure embedding representation of .

[0035] It should be understood that the interactive nodes in the cross-community academic knowledge graph are feature embedded to obtain the node structure feature vector in the Euclidean space, and based on the obtained node structure feature vector in the Euclidean space, it is mapped to the hyperbolic space through an exponential mapping function to better propagate the knowledge of hierarchical data. The structural feature vector mapped to the hyperbolic space is subjected to a multi-hop knowledge propagation process to aggregate the information of neighbor nodes. Specifically, it is mapped to the corresponding tangent space through a logarithmic mapping function, and according to The triples are used to calculate the weights of neighbor nodes.

[0036] It should be understood that based on the structural feature vector in the hyperbolic space, a multi-hop knowledge propagation process is carried out to aggregate the information of multi-hop neighbor nodes. A hierarchical aggregation scheme is designed for academic communities with different structures. Different hop numbers and neighbor sampling sizes are used to better learn the paper knowledge between different communities, and different aggregation strategies are adopted for the distribution of task nodes and paper nodes in the cross-community knowledge graph.

[0037] It should be understood that before neighbor node information is propagated, the attention weights of neighbor nodes must first be calculated in the tangent space. Depending on the connection relationship between the neighbor nodes and the target node, the contribution to the target node will also be different. Therefore, it is very important to distinguish different relationships by learning different attention weights.

[0038] Based on the structure embedding of the obtained task nodes , a task similarity graph is constructed according to the number of papers that interact with each other, and on this basis, similarity aggregation operations of task nodes are performed. At the same time, a task hierarchy graph is constructed according to the hierarchical relationship between tasks, and on this basis, upward aggregation operations of task nodes are performed.

[0039] Furthermore, the hierarchical aggregation results are respectively subjected to similarity aggregation and upward aggregation, wherein the specific steps of similarity aggregation include: firstly constructing a task similarity graph according to the number of common interactive papers between tasks, defining For task nodes With task nodes The number of papers that interact with each other, for each task node , only consider the top candidates with the largest number of interactive papers task nodes, if the task Belongs to the task node Before Tasks , then the weight of its edge is , otherwise 0.

[0040] .

[0041] The softmax function is used to calculate the weights of the adjacent tasks for the aggregation task nodes, so as to better aggregate the adjacent task nodes with the most identical number of interactive papers, where: For the task Before Neighbor task nodes.

[0042] .

[0043] It should be understood that after obtaining the structural feature vector in the hyperbolic space, the similarity aggregation between task nodes is performed according to the number of papers that interact together between task nodes. The more papers that interact together, the more similar the tasks are.

[0044] Furthermore, the hierarchical aggregation results are respectively subjected to similarity aggregation and upward aggregation, wherein the specific step of upward aggregation includes: constructing a task hierarchy graph according to the superior-subordinate relationship between task nodes, if the task node For task nodes The weight of this edge is 1. If the task node Not a task node The subtask of is 0, that is, the edge does not exist: ; Based on the task hierarchy diagram, the task representation after upward aggregation is obtained : ; in, For the task A collection of subtasks.

[0045] For example, object recognition is a subtask of computer vision, so there is an edge from object recognition to computer vision, but the recommendation system does not belong to the subtask of computer vision, so there is no edge.

[0046] It should be understood that, according to the superior-subordinate relationship between tasks, the information aggregation operation of the subtasks is performed on the superior task to capture more fine-grained representation information.

[0047] Furthermore, the similarity aggregation and upward aggregation results are subjected to feature fusion to obtain a final structural feature vector, including: Based on the obtained task nodes of and , perform a weighted fusion operation on the tangent space to obtain a dual representation containing similarity information and superior-subordinate information, and the attention weight Initialized to 0.1; ; For task nodes Embedded in its own structure and neighbor task node information embedding Sum and perform nonlinear transformation, then perform nonlinear activation in hyperbolic space to obtain the final structural embedding representation of the task node : ; in, and are trainable weights and biases defined in the relevant tangent space, and is a linear transformation in hyperbolic space, is a hyperbolic activation function.

[0048] It should be understood that feature fusion performs feature weighted fusion on the similarity aggregation results and the upward aggregation results to obtain the final structural feature vector, and further concatenates the final node structural feature vector and text feature vector to obtain the final feature embedding vector of the node.

[0049] It should be understood that the structural embedding vector obtained in the Euclidean space is projected onto the Lorentz manifold through the hyperbolic space embedding layer, and the multi-hop knowledge propagation operation of the knowledge graph nodes is performed in the tangent space of the Lorentz manifold, in which the Sum Aggregator method is used to aggregate multi-hop neighbor information. At the same time, a hierarchical aggregation scheme is proposed for different structures of the community, and different hop numbers and neighbor sampling sizes are adopted for communities with different structures to better learn the paper knowledge in different communities.

[0050] The paper node structure vector obtained after multi-hop propagation is used as its final structure embedding vector, and the task node structure embedding vector obtained needs to further perform similarity aggregation and upward aggregation operations under the task isomorphic graph, and then perform weighted fusion operations on it, and the task node structure embedding vector after further weighted fusion is used as its final structure embedding vector.

[0051] Furthermore, the text aggregation operation is performed on the initial text feature vector to obtain the final text feature vector, including: using LightGCN to To encode: ; ; in, and Task nodes and paper nodes The set of one-hop neighbors of and Initialized by text pre-training model; Used to normalize the text features learned in each layer to avoid continuous expansion of the scale.

[0052] Furthermore, according to the aggregation process, after L layers of data propagation, each node will obtain L+1 text feature vectors. The final text embedding representation is obtained by weighted summing the obtained L+1 text feature vector representations: ; .

[0053] It should be understood that although the use of knowledge graphs can effectively retain structural information, it lacks entity description information. In order to better improve the recommendation effect, this embodiment takes into account text information such as task descriptions and paper abstracts, and constructs a task-paper interaction graph. Text aggregation is performed in the network to better learn the interactive relationship between task nodes and paper nodes. The LightGCN method simplifies graph convolution operations by removing feature transformation and nonlinear activation modules to improve recommendation performance.

[0054] It should be understood that the text embedding vectors such as the task description and paper abstract obtained at the beginning are subjected to text embedding operations. Specifically, the initial text feature vector of the node is first obtained through the text pre-training model, and then the LightGCN method is used to perform multiple neighbor aggregations of text information in the task-paper interaction graph. The so-called task-paper interaction graph includes task nodes and paper nodes. If there is an association relationship between the task and the paper, there are interconnected edges between the corresponding task nodes and the paper nodes. After L layers of data propagation, the representation obtained by each GCN layer is weighted summed to obtain the final text embedding representation of the node.

[0055] The obtained initial node text feature vector is used to perform text feature aggregation operation on the constructed task-paper interaction graph using the LightGCN method to obtain the final text feature vector, so as to better learn the semantic relationship between interaction nodes.

[0056] Furthermore, the final structural feature vector and the final text feature vector are concatenated to obtain the final feature vector of each node in the cross-community academic knowledge graph, including: And task node structure embedding representation , and text embedding representations of paper nodes and task nodes , , the structure embedding representation and the text embedding representation are concatenated through the concat operation to better utilize these two different information, thereby obtaining the final node feature embedding vector.

[0057] Since the current structure embedding representation is in the hyperbolic space, it is first mapped to the tangent space and then concatenated.

[0058] . .

[0059] in, This is the concat operation.

[0060] Furthermore, the method also includes: at the same time, using the trained paper recommendation model to extract the to-be-recommended paper query task feature vector from the to-be-recommended paper query task node.

[0061] Furthermore, the calculation of the similarity between the feature vector of the task of the paper to be recommended and the feature vector of each paper node in the cross-community academic knowledge graph includes: after obtaining the final feature representation of the task node and the paper node, calculating the matching score between them by using the L2 distance in the matching model: ;in, is the sigmoid function, The higher the score, the better the task. Recommended Papers The greater the possibility.

[0062] After obtaining the final node structure embedding vector and text embedding vector, in order to better utilize these two different information, the structure embedding vector in the hyperbolic space is mapped to the tangent space, and then a concat operation is performed on it to obtain the final representation vectors of the task node and the paper node; based on the final feature embedding vectors of the task node and the paper node, the matching score between them is calculated by constructing a softmax scoring function.

[0063] Furthermore, S102: calculates the similarity between the feature vector of the query task of the paper to be recommended and the feature vector of each node in the cross-community academic knowledge graph, outputs the paper recommendation results based on the similarity, and outputs the top N papers with the highest similarity as recommended papers.

[0064] Furthermore, S102: inputs the paper query task to be recommended and the cross-community academic knowledge graph into the trained paper recommendation model, and the paper recommendation model outputs the recommended paper, wherein the training process of the trained paper recommendation model includes: constructing a training set, the training set is a cross-community academic knowledge graph of known paper recommendation results; the cross-community academic knowledge graph includes: paper query task nodes and paper nodes; if there is an association relationship between the paper query task node and the paper node, there are interconnected edges between the two; the cross-community academic knowledge graph and the known paper query task are used as input values ​​of the paper recommendation model, and the known paper recommendation results are used as output values ​​of the paper recommendation model. When the total loss function value of the paper recommendation model no longer decreases, the training is stopped to obtain the trained paper recommendation model.

[0065] Using cross entropy as the loss function, let For task nodes A collection of orthogonal interaction papers, let is a collection of papers with the same number of negative interactions, and each training process and All samples are randomly sampled, so the loss function is defined as follows: ; in, is the cross entropy loss, is the set of trainable parameters and feature embeddings in the model, is a hyperparameter that controls the size of the L2 regularizer.

[0066] The click-through rate (CTR) of the test samples is predicted and the results are compared with the actual situation. This embodiment uses accuracy (ACC) and area under the curve (AUC) as evaluation indicators. The comparison results are shown in Table 1.

[0067] Table 1 Accuracy and AUC of the paper CTR prediction task

[0068] And the Top-K paper recommendation prediction is performed on the test samples, and the results are compared with the actual situation. This embodiment uses Precision@K, Recall@K and ndcg@K as evaluation indicators.

[0069] Based on the results in Table 1, it can be seen that the recommendation performance of the paper recommendation model proposed in this embodiment is better than other recommendation methods.

[0070] At the same time, in order to determine whether the cross-community knowledge graph can effectively improve the recommendation performance, this embodiment further implements a cross-community knowledge graph effectiveness verification experiment for the recommendation method using the knowledge graph. The results are shown in Table 2.

[0071] Table 2 Study on the effectiveness of cross-community knowledge graph

[0072] The results in Table 2 show that the paper recommendation method based on the cross-community knowledge graph proposed in this embodiment plays a certain positive role in improving the recommendation performance.

[0073] After obtaining the final representation vectors of the task and paper nodes, the matching score between them is calculated by using the L2 distance in the matching model. The matching score is calculated by constructing a softmax function. The learning parameters of the model are trained using the back propagation algorithm by calculating the loss function of the output value of the softmax function to complete the model training.

[0074] The loss error between the output value of the softmax scoring function and the label value is used, and the back propagation algorithm is used to train the learning parameters of the entire model to complete the model training.

[0075] After the model is trained, it is tested using data that did not participate in the training process. The model test results are compared with the actual situation, and the model is optimized and adjusted to continuously optimize the weight data in the model and continuously improve the accuracy of paper recommendations.

[0076] Exemplarily, a large amount of semi-structured paper data from the PapersWithCode platform and the Github platform is first collected, and the acquired massive paper data is preprocessed, including data cleaning, data definition and storage. Specifically, a total of 4496 tasks, 64179 papers, and 111809 interactions were collected from the open academic platform PapersWithCode. And data cleaning was performed on it, including data deduplication, missing value completion, etc., and finally 1288 tasks, 28978 papers and 76605 interactions were retained as data samples. At the same time, text information such as task descriptions and paper abstracts were collected to further learn the semantic relationship between tasks and papers. Then, based on the 34612 code attribute entities contained in the retained papers, other paper information related to them was collected as core entities in the Github community. And it is stored in a relational database.

[0077] The massive amount of paper data obtained is normalized, and Protege is used to perform the ontology modeling task of the cross-community academic knowledge graph, including defining the relevant entities in the paper recommendation and the object attributes and numerical attributes of each type of entity. Then, according to the defined ontology model, the relational data is converted into a triple format with the help of mapping rules to complete the construction task of the cross-community academic knowledge graph. Finally, the Neo4j graph database is used to store the data of the cross-community academic knowledge graph. As shown in Table 3, this is an example of the statistical data set of the paper, where The main knowledge graph constructed for the paper information in the PapersWithCode platform, Auxiliary knowledge graph built for paper information in the Github platform, It is a cross-community academic knowledge graph formed by the association of two academic communities. If a paper belongs to a task, it is said that there is an interaction relationship between them, and the number of interactions is the number of interactions between task nodes and paper nodes in the dataset.

[0078] Table 3 Basic statistical information of the paper dataset

[0079] Extract various attribute information of papers from massive paper data obtained from academic communities with different structures; based on the obtained paper attribute information, construct a cross-community academic knowledge graph through the associated entities between different communities; and obtain the structural embedding vector and text embedding vector of the node through the embedding layer and text pre-training model respectively for the interactive nodes in the cross-community academic knowledge graph.

[0080] The obtained structural embedding vector is mapped to the hyperbolic space through an exponential mapping function, and the knowledge propagation operation is performed through a hierarchical aggregation method. In addition, the task nodes are further similarly aggregated and upwardly aggregated on the constructed isomorphic graph, and the final structural feature vector is obtained through feature fusion operation.

[0081] The obtained text embedding vector is subjected to text aggregation operation on the constructed interaction graph through the LightGCN method to obtain the final text feature vector; the structural feature vector and the text feature vector are concatenated to obtain the final node feature representation vector.

[0082] After obtaining the final representation vectors of the task and paper nodes, the matching score between them is calculated by using the L2 distance most commonly used in the matching model. The matching score is calculated by constructing a softmax function. The loss function of the output value of the softmax function is calculated, and the back propagation algorithm is used to train the learning parameters of the model to complete the training of the model.

[0083] This paper fully considers and integrates the paper information between academic communities with different structures, and proposes a hierarchical aggregation scheme in hyperbolic space for the problem of learning node representations in large and complex cross-community knowledge graphs, which further effectively improves the paper recommendation performance.

[0084] Embodiment 2: This embodiment provides a paper recommendation system based on a cross-community knowledge graph in a hyperbolic space, including: an acquisition module, which is configured to: acquire a paper query task to be recommended and a cross-community academic knowledge graph; a recommendation module, which is configured to: input the paper query task to be recommended and the cross-community academic knowledge graph into a trained paper recommendation model, and the paper recommendation model outputs recommended papers; Among them, the trained paper recommendation model is used to extract the initial structural feature vector and initial text feature vector of each paper node in the input cross-community academic knowledge graph, perform hierarchical aggregation on the initial structural feature vector, perform similarity aggregation and upward aggregation on the hierarchical aggregation results respectively, and perform feature fusion on the similarity aggregation and upward aggregation results to obtain the final structural feature vector; perform text aggregation operation on the initial text feature vector to obtain the final text feature vector; perform splicing operation on the final structural feature vector and the final text feature vector to obtain the final feature vector of each node in the cross-community academic knowledge graph; calculate the similarity between the feature vector of the query task of the paper to be recommended and the feature vector of each paper node in the cross-community academic knowledge graph, and output the paper recommendation result according to the similarity.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A paper recommendation method based on cross-community knowledge graph in hyperbolic space, characterized by: include: Obtain the query task of papers to be recommended and the cross-community academic knowledge graph; Input the query task of the papers to be recommended and the cross-community academic knowledge graph into the trained paper recommendation model, and the paper recommendation model outputs the recommended papers; Among them, the trained paper recommendation model is used to extract the initial structural feature vector and initial text feature vector of each paper node in the input cross-community academic knowledge graph, perform hierarchical aggregation on the initial structural feature vector, perform similarity aggregation and upward aggregation on the hierarchical aggregation results respectively, and perform feature fusion on the similarity aggregation and upward aggregation results to obtain the final structural feature vector; perform text aggregation operation on the initial text feature vector to obtain the final text feature vector; perform splicing operation on the final structural feature vector and the final text feature vector to obtain the final feature vector of each node in the cross-community academic knowledge graph; calculate the similarity between the feature vector of the query task of the paper to be recommended and the feature vector of each paper node in the cross-community academic knowledge graph, and output the paper recommendation result according to the similarity.

2. The paper recommendation method based on cross-community knowledge graph in hyperbolic space as claimed in claim 1 is characterized in that: The input cross-community academic knowledge graph is used to extract the initial structural feature vector and the initial text feature vector of each paper node in the knowledge graph, wherein the extraction process of the initial structural feature vector includes: The paper nodes, paper query task nodes and related paper attribute information are obtained from the constructed cross-community academic knowledge graph, and the structural feature embedding layer , using a linear layer as layer, and obtain the node structure feature vector in Euclidean space : ； in, is the representation of the corresponding node in the knowledge graph; The node structure feature vector obtained in the Euclidean space is mapped to the Lorentz manifold, and the negative curvature is of The Lorentz manifold is denoted by ,make and represents the Euclidean embedding and the transformed hyperbolic embedding, is mapped as follows: ； in, For the tangent space dimensional vector, Defined as The origin in is used as a reference vector to perform tangent space operations, for The second norm of is the exponential mapping function, and are the hyperbolic sine and cosine functions respectively.

3. The paper recommendation method based on cross-community knowledge graph in hyperbolic space as claimed in claim 1 is characterized in that: The step of performing hierarchical aggregation on the initial structural feature vector includes: (1) By considering the triple The structural information of the nodes is used to learn the attention weights of the nodes and edges between the nodes connected by the relationship : ； in, is the relationship weight matrix, and They are and Embedded in hyperbolic space, and using The function maps it into the tangent space; Then, by using the softmax function, The connected edges are further normalized to obtain : ； (2) Based on the neighbor node weight calculation formula, firstly, the knowledge propagation process of neighbor node information is carried out on the task node, and For task nodes exist The first-order neighbor set in the task node is composed only of paper nodes; therefore, The embedding of neighbor information can be defined as follows: ； in, For the corresponding side The normalized attention weights of and They are and Embedding in hyperbolic space; (3) Similarly, for the paper node Perform similar neighbor information aggregation operations, and let For thesis node The first-order task node neighbor set, For thesis node exist The first-order other entity neighbor set in the paper node The embedding of neighbor information is defined as: ； (4) Perform multi-hop operations in the knowledge graph to conduct Neighbor sampling, let represent The The set of entity neighbors that are jumped; then, through Medium Entity Neighbors to update Entities in Embedding ； in, For the corresponding side The normalized attention weights of and Entity and Embedding representation in hyperbolic space; (5) There may be associated entities in , defined as ,in is a collection of related entities; Entities in When embedding neighbor information, first consider whether there is an associated entity in its neighbor set. ; If there is an associated entity, first Related entities Perform neighbor aggregation and then obtain Related entities of medium and high-level knowledge Put into To further Neighborhood information embedding learning; exist Related entities When acquiring high-order neighbor information, different neighbor sampling sizes and aggregation layers are used; represent middle No. The set of neighbors of the entity that jumped off; ； (6) After obtaining the multi-hop neighbor information embedding, it is used to update the embedding representation of the task node and the paper node; the updated embedding representation is obtained by summing the embedding of the target node itself and the neighbor information embedding, performing a nonlinear transformation, and then performing a nonlinear activation in the hyperbolic space: ； in and are trainable weights and biases defined in the relevant tangent space, and is a linear transformation in hyperbolic space, is a hyperbolic activation function, which is defined as: ； ； ； in, , and is an exponential function, and is a logarithmic function, is the hyperbolic tangent activation function; So far, we have obtained For thesis node The final structure embedding representation of .

4. The paper recommendation method based on cross-community knowledge graph in hyperbolic space as claimed in claim 1 is characterized in that: The hierarchical aggregation results are respectively subjected to similarity aggregation and upward aggregation, wherein the specific steps of similarity aggregation include: First, we construct a task similarity graph based on the number of common interactive papers between tasks and define For task nodes With task nodes The number of papers that interact with each other, for each task node , only consider the top candidates with the largest number of interactive papers task nodes, if the task Belongs to the task node Before Tasks , then the weight of its edge is , otherwise 0: ； Use the softmax function to calculate the weights of the neighboring tasks for the aggregation task node: ； in, For the task Before Neighbor task nodes.

5. The paper recommendation method based on cross-community knowledge graph in hyperbolic space as claimed in claim 1 is characterized in that: The hierarchical aggregation results are respectively subjected to similarity aggregation and upward aggregation, wherein the specific steps of upward aggregation include: Construct a task hierarchy diagram based on the hierarchical relationship between task nodes. For task nodes The weight of this edge is 1. If the task node Not a task node The subtask of is 0, that is, the edge does not exist: ； Based on the task hierarchy diagram, the task representation after upward aggregation is obtained : ； in, For the task A collection of subtasks.

6. The paper recommendation method based on cross-community knowledge graph in hyperbolic space as claimed in claim 1 is characterized in that: The similarity aggregation and upward aggregation results are subjected to feature fusion to obtain the final structural feature vector, including: Based on the obtained task nodes of and , perform a weighted fusion operation on the tangent space to obtain a dual representation containing similarity information and superior-subordinate information, and the attention weight Initialized to 0.1; ； For task nodes Embedded in its own structure and neighbor task node information embedding Sum and perform nonlinear transformation, then perform nonlinear activation in hyperbolic space to obtain the final structural embedding representation of the task node : ； in, and are trainable weights and biases defined in the relevant tangent space, and is a linear transformation in hyperbolic space, is a hyperbolic activation function.

7. The paper recommendation method based on cross-community knowledge graph in hyperbolic space as claimed in claim 1 is characterized in that: The text aggregation operation is performed on the initial text feature vector to obtain the final text feature vector, including: right To encode: ； ； in, and Task nodes and paper nodes The set of one-hop neighbors of and Initialized by text pre-training model; Used to normalize the text features learned in each layer; According to the aggregation process, after L layers of data propagation, each node will obtain L+1 text feature vectors. The final text embedding representation is obtained by weighted summing the obtained L+1 text feature vector representations: ； 。 8. The paper recommendation method based on cross-community knowledge graph in hyperbolic space as claimed in claim 1 is characterized in that: The final structural feature vector and the final text feature vector are concatenated to obtain the final feature vector of each node in the cross-community academic knowledge graph, including: Embedding representation based on paper node structure and task node structure embedding representation , and text embedding representations of paper nodes and task nodes , , the structural embedding representation and the text embedding representation are concatenated through the concat operation to obtain the final node feature embedding vector. Since the current structural embedding representation is in the hyperbolic space, it is first mapped to the tangent space and then concatenated: ； ； in, This is the concat operation.

9. The paper recommendation method based on cross-community knowledge graph in hyperbolic space as claimed in claim 1 is characterized in that: The calculation of the similarity between the feature vector of the query task of the paper to be recommended and the feature vector of each paper node in the cross-community academic knowledge graph includes: After obtaining the final feature representation of the task node and the paper node, the matching score between them is calculated by using the L2 distance in the matching model: ； in, is the sigmoid function, The higher the score, the better the task. Recommended Papers The greater the possibility.

10. A paper recommendation system based on a cross-community knowledge graph in a hyperbolic space, characterized by: include: An acquisition module is configured to: acquire a query task of papers to be recommended and a cross-community academic knowledge graph; The recommendation module is configured to: input the query task of the paper to be recommended and the cross-community academic knowledge graph into the trained paper recommendation model, and the paper recommendation model outputs the recommended paper; Among them, the trained paper recommendation model is used to extract the initial structural feature vector and initial text feature vector of each paper node in the input cross-community academic knowledge graph, perform hierarchical aggregation on the initial structural feature vector, perform similarity aggregation and upward aggregation on the hierarchical aggregation results respectively, and perform feature fusion on the similarity aggregation and upward aggregation results to obtain the final structural feature vector; perform text aggregation operation on the initial text feature vector to obtain the final text feature vector; perform splicing operation on the final structural feature vector and the final text feature vector to obtain the final feature vector of each node in the cross-community academic knowledge graph; calculate the similarity between the feature vector of the query task of the paper to be recommended and the feature vector of each paper node in the cross-community academic knowledge graph, and output the paper recommendation result according to the similarity.

Images