Super-network-based scientific research cooperation recommendation method and system

Abstract

The invention provides a hyper-network-based scientific research cooperation recommendation method and system, and the method comprises the steps: obtaining a multivariate relationship between entities from field-related scientific and technical literatures, taking the entities as nodes, taking the multivariate relationship as hyperedges, and building time labels on the hyperedges to form a multi-dimensional time sequence hyper-network; measuring the structural similarity between the nodes in the multi-dimensional time sequence super network at different times; measuring the attribute similarity among the nodes in the multi-dimensional time sequence super network at different times; and fusing the structural similarity and the attribute similarity to calculate a cooperation prediction score between authors, and determining whether to recommend the authors to perform scientific research cooperation or not according to the cooperation prediction score. According to the method, objects with cooperation potential can be recommended for scientific research personnel, selection and configuration of group members can be assisted, academic cooperation groups can be constructed, and the method plays an important role in potential cooperation relationship mining and hotspot research subject discovery.

Description

Technical Field

[0001] The present invention relates to the field of intelligent recommendation, and in particular to a method and system for recommending scientific research cooperation based on a hypernetwork. Background Art

[0002] Scientific research cooperation recommendation can be divided into three categories: content-based recommendation, relationship network-based recommendation, and hybrid recommendation. Content-based recommendation usually uses topic models and vector models to mine the academic texts produced by authors, and matches and recommends authors with similar content. Rosen et al. introduced author-document relationship information to construct an author topic model for research such as author clustering and similar author recommendation. Liu Ping et al. constructed an author-keyword network and divided the community, and then calculated the topic similarity of authors in the same community based on the LDA model to achieve scientific research cooperation recommendation. Pu Shanshan et al. used the author topic model to extract the knowledge structure characteristics of the author, and then clustered the authors based on the characteristics, recommending the authors with the highest academic influence in each cluster to form an academic complementary team. Li Zhong et al. used a sparse vector model to represent the text content of the author's paper, and achieved better recommendation results than the LDA model. Kong et al. constructed a topic-aware vector model, aiming to use the author's topic information and cooperation information at the same time in scientific research cooperation recommendation.

[0003] Recommendations based on relationship networks usually use cooperation, citation, and co-occurrence relationships to build a relationship network, and then use link prediction technology to evaluate the possibility of two non-adjacent nodes (authors) on the network to generate links (cooperation) in the future. Yu Chuanming et al. combined the Node2vec and LINE models to simultaneously learn the node location characteristics and topological structure characteristics of the scientific research cooperation network, and realized cooperation recommendation through a logistic regression model. Zhou et al. used authors, papers, and keywords as nodes, and cooperation, citation, and inclusion relationships as edges to build a semantic link network, and used semantic paths to realize cooperation recommendation. Lin Yuan et al. extracted information such as authors, institutions, and keywords from scientific literature to build a co-occurrence network, and used network representation learning technology to mine potential collaborators and institutions for authors.

[0004] Content-based recommendation ignores the association between authors, while relational network-based recommendation does not consider the attribute characteristics of the nodes themselves. Hybrid recommendation is to combine content-based recommendation with relational network-based recommendation to make its recommendation effect higher than single recommendation. Wang Zhibing et al. defined the node institution attribute similarity index and network structure similarity index on the scientific research cooperation network, and weighted the two types of indicators to predict the probability of cooperation between scholars. Zhang Xin et al. used the Node2vec model and the author topic model to obtain the structural characteristics of the scientific research cooperation network and the topic characteristics of the author, respectively, and then linearly fused these two characteristics to calculate the author similarity. Xiong Huixiang et al. used scholar feature words and scholar co-citation relationship to calculate the combined similarity to achieve personalized recommendation for scholars. Du et al. used attribute network embedding technology to build a cooperation recommendation model, which can simultaneously utilize the structural characteristics of the network and the multi-dimensional attribute characteristics of scholars. Pradhan et al. used topic models and Doc2vec to extract the academic characteristics of authors, and then used the feature-guided random walk model on the cooperation network to achieve scientific research cooperation recommendation. The above methods all use ordinary graph structures when constructing relationship networks, which cannot represent the complex multivariate relationships between authors, between documents, and between authors and documents. Summary of the invention

[0005] Aiming at the problems existing in the prior art, a method and system for recommending scientific research cooperation based on a hypernetwork is provided, which recommends objects with cooperation potential to scientific researchers by constructing a hypernetwork.

[0006] The first aspect of the present invention proposes a method for recommending scientific research cooperation based on a hypernetwork, comprising: Obtain multi-dimensional relationships between entities from relevant scientific and technological literature in the field, use entities as nodes, multi-dimensional relationships as hyperedges, and establish time labels on the edges to form a multi-dimensional time series hypernetwork; Structural similarity measurement between nodes in multidimensional temporal hypernetworks at different times; Attribute similarity measurement between nodes in multidimensional temporal hypernetworks at different times; For two authors who do not have a cooperative relationship, the cooperation prediction score between the authors is calculated by integrating structural similarity and attribute similarity, and whether to recommend the authors for scientific research cooperation is determined based on the cooperation prediction score.

[0007] As a preferred solution, the specific method for forming the multidimensional temporal hypernetwork includes: Perform text segmentation on massive scientific and technological documents to obtain sentence sets; Given a list of entity types, use a large language model to determine the matching probability between sentence text features and different entity types, remove entity types with probabilities less than a preset threshold, and obtain the entity list actually contained in the scientific literature; Given a list of multiple relationship types, use a large language model to determine and obtain the corresponding relationship types of entities in the entity list; Different entities are represented as nodes, multivariate relationships are represented as edges, and time is used as the feature of the edge to establish author subnetworks, subject subnetworks, and institution subnetworks respectively. The subnetworks are sliced ​​according to the time window, and hyperedges are established between nodes of different subnetworks in the same slice to construct a multidimensional temporal hypernetwork.

[0008] As a preferred solution, in the multidimensional temporal hypernetwork, the correlation matrix H Defined as The matrix whose elements are defined as follows:

[0009] node v With super edge e If the matrix H The corresponding element value is 1, otherwise the corresponding element value is 0; At this time, the node v's overtime at time x Defined as:

[0010]

[0011] in, To represent the hyperedge e Is it in time x Appear, For super edge e time tag.

[0012] As a preferred solution, in the multidimensional time series hypernetwork, the nodes v On the super edge e The weight on is defined as:

[0013]

[0014] in, Is a node In scientific literature The weight of the node is assigned by the frequency of the node appearing in the literature. Is a node In scientific literature d The order of frequency of occurrence in It is a document d The number of nodes in is the document set corresponding to the hyperedge e.

[0015] As a preferred solution, the structural similarity measurement method between nodes in the multidimensional temporal hypernetwork at different times includes: For any time point, the weight of the hyperedge corresponding to the node is added to the process of calculating the common neighbor index and resource allocation index between nodes; The entropy weight method is used to determine the weights between the common neighbor index and the resource allocation index to obtain the structural similarity between the nodes at that point in time.

[0016] As a preferred solution, the common neighbor index calculation method is:

[0017] The resource allocation indicator calculation method is:

[0018] in, For Node x and nodes y The common neighbor index at time t, For Node x and nodes y Resource allocation index at time t, parameter a Used to adjust the influence of the node weight in the hyperedge on the similarity value. , HCN and HRA degenerate into unweighted forms if , the metric is equivalent to the ordinary weight metric; Is a node x and nodes z The weight at time t is defined as follows:

[0019] in, is a hyperedge set, each hyperedge in the set e All included x and z Two nodes, Represents a hyperedge e Is it in time t Appear, For Node x On the super edge e The weight on For Node z On the super edge e The weight on is the weight of node z and node y at time t, with the same definition.

[0020] As a preferred solution, the specific calculation method of the structural similarity is:

[0021] in, For Node x and nodes y Structural similarity at time t; EWM stands for entropy weight method, which measures the internal differences of each indicator data and determines the weight through information entropy.

[0022] As a preferred solution, the attribute similarity measurement method between nodes in the multidimensional temporal hypernetwork at different times includes: At any time point, the dynamic subject model is used to map the node into a vector form, which is then input into the variational autoencoder to generate the node's attribute feature vector in an unsupervised manner; The Euclidean distance between the attribute feature vectors of the nodes is calculated to obtain the attribute similarity between the nodes at that time point.

[0023] As a preferred solution, the cooperation prediction score calculation method is:

[0024]

[0025]

[0026] in, For Author x, y The cooperation prediction score of and Respectively indicate the author x, y Overall structural similarity and overall attribute similarity, , Respectively indicate the author x, y The corresponding author nodes, subject nodes and institution nodes in the multidimensional temporal hypernetwork, the balance parameter Used to adjust the weight of structural similarity value and similarity value. Respectively indicate the author x With the author y The structural similarity between the author node, topic node and structure node, Respectively indicate the author x With the author y The attribute similarity between the author node, topic node and structure node.

[0027] The second aspect of the present invention proposes a scientific research cooperation recommendation system based on a hypernetwork, comprising a computer system and a computer program, and when the computer program is run by the computer system, the method described in the first aspect is executed. Compared with the prior art, the beneficial effects of adopting the above technical solution are: (1) The present invention can recommend potential partners for scientific researchers, assist in the selection and configuration of group members, build academic cooperation groups, and play an important role in mining potential cooperative relationships and discovering hot research topics.

[0028] (2) The present invention proposes to model the multivariate relationships between objects based on the scientific research cooperation hypernetwork to avoid information loss caused by the conversion of multivariate relationships into binary relationships.

[0029] (3) The present invention uses both structural similarity features and attribute similarity features on the scientific research cooperation supernetwork to achieve hybrid recommendation of scientific research cooperation, which not only avoids the problem of sparse explicit cooperation relationships, but also avoids the problem of too concentrated and lack of diversity in implicit cooperation recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of the hypernetwork-based scientific research collaboration recommendation method proposed in the present invention.

[0031] Figure 2 Schematic diagram of a multi-dimensional temporal hypernetwork constructed in one embodiment of the present invention.

[0032] Figure 3 The figure is a schematic diagram of generating an attribute feature vector in one embodiment of the present invention. DETAILED DESCRIPTION

[0033] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar modules or modules with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be construed as limitations on the present application. On the contrary, the embodiments of the present application include all changes, modifications and equivalents that fall within the spirit and connotation of the appended claims.

[0034] In the era of big data where information overload is becoming more and more serious, how to help researchers find the information they want from massive academic data has become a difficult problem that needs to be solved urgently. Search engines can assist scientific research users in screening and filtering information in the ocean of information, but search engines can only passively obey input instructions and cannot actively recommend content of interest to users. The scientific research cooperation recommendation method proposed in the embodiment of the present invention can provide users with personalized recommendation services while solving the problem of information overload, and can provide reference and suggestions for the selection and configuration of team members. Please refer to Figure 1 , the specific plan is as follows: Step 1: Construct a multi-dimensional temporal hypernetwork.

[0035] In this embodiment, a large prediction model is used to obtain multivariate relationships between entities from field-related scientific and technological literature, with entities as nodes and multivariate relationships as hyperedges, and time tags are established on the edges to form a multidimensional temporal hypernetwork. In one embodiment, entities such as authors, topics, and institutions can be obtained from scientific and technological literature. Constructing a multidimensional temporal hypernetwork mainly includes two parts: entity recognition and multivariate relationship extraction and hypernetwork construction. The following describes these two parts one by one.

[0036] (1) Entity recognition and multi-relation extraction Scientific and technological literature d Perform text segmentation to obtain sentence sets S , and then for each sentence Construct prompts to obtain entities and multi-party relations through two-round question-answering with a large language model.

[0037] First round of questions and answers: Given a list of entity types, use a large language model to determine the matching probability between sentence text features and different entity types. The prompt is constructed as follows: Given a list of entity types: [entity type 1, entity type 2, ...] Question: What entity types may be contained in the sentence "The text..." and give the probability.

[0038] Eliminate entity types with probability less than the preset threshold and obtain the entity list actually contained in the scientific literature (entity type s1 , entity type s2 , ..., entity type s ).

[0039] Second round of questions and answers: Given a list of multi-relationship types, use the large language model to determine and obtain the corresponding relationship types of entities in the entity list. The prompt construction is as follows: The given list of multi-relationship types is: [relationship type 1, relationship type 2, ...] Question: "Text..." Determine the above sentence (entity type s1, entity type s2, ..., entity type sn) corresponding to the relationship type.

[0040] After two rounds of question and answer, the entities and multiple relationships between entities can be obtained.

[0041] (2) Construction of multi-dimensional temporal hypernetwork Please refer to Figure 2 , perform entity recognition and multi-relationship extraction on massive scientific and technological literature, represent different entities as nodes, multi-relationships as edges, and use time as the feature of the edge to establish author subnets, subject subnets, and institution subnets respectively. Then slice the subnets according to the time window, establish hyperedges between nodes of different subnets in the same slice, and construct a multi-dimensional time series hypernetwork.

[0042] In this embodiment, the multidimensional temporal hypernetwork is defined as ,in, is a finite set of nodes. n is the number of nodes; is a set of node attributes, Representation Node The corresponding text attributes; is a superedge set, is the number of hyperedges; represents a hyperedge, m is the number of nodes in the hyperedge, ; It is a super edge e The time tag, Is a node v On the super edge e The weights on the hypernetwork. The hypernetwork corresponds to the association matrix H is a The matrix whose elements are defined as follows:

[0043] node v With super edge e If the matrix H The corresponding element value in is 1, otherwise the corresponding element value is 0. In the hypernetwork, the node v In time x The salvation Defined as:

[0044]

[0045] in, To represent the hyperedge e Is it in time x Appear, For super edge e time tag.

[0046] Each hyperedge contains multiple nodes. Different nodes often contribute differently to the hyperedge, so it is necessary to reasonably assign weights to the nodes. v On the super edge e The weight calculation formula on is: (1) (2) in, Is a node In the literature The weight of the node is assigned by the frequency of the node appearing in the literature. Is a node In the paper dThe order of frequency of occurrence in It is a document d The number of nodes in is the document set corresponding to the hyperedge e. It can be seen that formula (2) can effectively assign weights to nodes, which meets the following three principles: 1) In the paper d The more times a node appears in the graph, the greater its weight; 2) Each node is given a reasonable weight; 3) v When the ranking remains unchanged, the larger the number of nodes, the more nodes v The smaller the weight.

[0047] Step 2: Measure the structural similarity between nodes in the multidimensional temporal hypernetwork at different times.

[0048] Traditional structural similarity metrics are implemented by studying the topological structure of ordinary networks and have achieved high accuracy on public datasets, such as Common Neighbor (CN) and Resource Allocation (RA). Among them, the CN metric considers that two nodes x and y The more common neighbors the two nodes have, the more similar they are. The definition is shown in formula (3): (3) The RA indicator considers that the similarity between two nodes is determined by a node x Pass to another node y The resource volume is determined by the amount of resources, and its definition is shown in formula (4): (4) Is a node x The set of neighbor nodes of Neighbor node z The degree of the node x The neighbors are the medium for transferring resources. Each medium also has one unit of resources and distributes it equally to its neighbors. Finally, the node y The amount of resources received is the node x and nodes y The similarity value of .

[0049] The quality of structural similarity metrics depends on whether they can accurately capture the structural characteristics of the network. The CN and RA metrics only consider the binary relationship between nodes in ordinary graphs, and ignore the impact of the weight characteristics of nodes and edges on similarity. In fact, the rational use of weight information in hypernetworks can effectively improve the accuracy of link prediction. Therefore, this embodiment adds the weight of the hyperedge corresponding to the node to the process of calculating the common neighbor index and resource allocation index between nodes, making it suitable for multi-dimensional temporal hypernetworks to achieve structural similarity measurement. Specifically, this embodiment proposes a specific definition method for the common neighbor index HCN and resource allocation index HRA of a multi-dimensional temporal hypernetwork, as shown in formulas (5) and (6): (5) (6) Among them, the parameters a It is used to adjust the influence of the weight of the node in the hyperedge on the similarity value. , HCN and HRA degenerate into unweighted forms; if , the metric is equivalent to the ordinary weight metric. is the weight of the corresponding hyperedge between node x and node z, which is defined as shown in formula (7): (7) Among them, the super edge set , each hyperedge e in the set contains two nodes, x and z. Represents a hyperedge e Is it in time t Appear, For Node x On the super edge e The weight on For Node z On the super edge e The weight on is the weight of node z and node y at time t, with the same definition.

[0050] HCN measures the similarity of the neighbor structures of hypernetwork nodes, while HRA calculates the amount of resources transferred between two hypernetwork nodes. The larger the values ​​of these two indicators, the more similar the local structures of the nodes are. According to the principle of clustering, two nodes with similar local structures have a greater probability of generating edges.

[0051] In this embodiment, the HCN and HRA indicators obtained by the entropy weight method are used to measure the structural similarity between nodes. Specifically, the structural similarity between node x and node y at time t is calculated as follows:

[0052] Among them, EWM (the entropy weight method) represents the entropy weight method, which measures the internal differences of each indicator data and determines the weights through information entropy. It should be noted that the structural similarity measurement is calculated within the subnet, that is, the measurement is performed between nodes in the author subnet, between nodes in the subject subnet, and between nodes in the institutional subnet.

[0053] Step 3: Measure the attribute similarity between nodes in the multidimensional temporal hypernetwork at different times.

[0054] When building a relationship network, we usually face the problem of sparse network structure caused by missing literature. It is difficult to obtain accurate cooperation recommendation results by relying solely on the topological structure. Existing recommendation research shows that node attributes can improve the accuracy of the recommendation model on the one hand, and alleviate the cold start and data sparsity problems on the other hand. Nodes in multidimensional time series hypernetworks carry a large number of text attributes. Semantic analysis can effectively mine the implicit relationship between nodes to make up for the shortcomings of topological structure similarity measurement. Topic model is a key technology in text semantic analysis and is widely used in scientific research cooperation prediction, field hotspot detection and community discovery. Dynamic Topic Model (DTM) introduces information about the evolution of topics over time based on LDA (Latent Dirichlet Allocation).

[0055] In this embodiment, a dynamic topic model is used to construct the attribute features of nodes, and the attribute similarity calculation is converted into the distance calculation between vectors. The output of DTM is the text-topic matrix , Topic-Word Matrix and time-theme matrix , using the matrix and You can set the time t Node at time v Mapping to vector form:

[0056] The dimension of is the same as the number of topics and the value of each dimension is non-negative. Directly using it as the attribute feature vector of the node will affect the flexibility and computability of the vector representation.

[0057] Furthermore, this embodiment uses a stacked autoencoder (SAE) to model node features and generate attribute feature vectors in an unsupervised manner. Figure 3 As shown, is the input of the encoder, and the output is the variational distribution , is the latent vector, are the parameters of the inference network. The input of the decoder is , the output is the probability distribution p , are the parameters of the generated network. Input into the encoder, then the potential vector z is the constructed attribute feature vector.

[0058] After obtaining the attribute feature vector of the node, this embodiment uses the Euclidean distance to evaluate the node x and y The attribute similarity between them is defined as shown in formula (10):

[0059] in, is the node at time t x, y The similarity of attributes between them.

[0060] It should be noted that the attribute similarity measure is calculated within the subnet, that is, between nodes in the author subnet, between nodes in the topic subnet, and between nodes in the institution subnet.

[0061] Step 4: Recommend scientific research cooperation.

[0062] In scientific research activities, the cooperative relationships between authors are divided into two categories: explicit relationships and implicit relationships. Among them, explicit relationships are relationships established by authors through co-publishing documents, and implicit relationships refer to potential cooperative relationships formed when some authors do not directly cooperate due to factors such as region and time, but have similar research interests and topics. Traditional cooperative recommendation algorithms usually use explicit or implicit relationships alone to build recommendation models. The former faces the problem of sparse explicit cooperative relationships, while the latter will lead to over-concentration and lack of diversity in recommendation results, resulting in poor recommendation effects.

[0063] The structural similarity index and attribute similarity index calculated in this embodiment respectively represent the explicit relationship and invisible relationship in the multidimensional temporal hypernetwork. In order to capture the characteristics of these two types of cooperative relationships in cooperative recommendation, the two metrics are combined to predict the author x With the author y In time t The possibility of cooperation, the prediction score calculation formula is as follows:

[0064]

[0065]

[0066] in, For Author x, y The cooperation prediction score of and Respectively indicate the author x, y Overall structural similarity and overall attribute similarity, , Respectively indicate the author x, y The corresponding author nodes, subject nodes and institution nodes in the multidimensional temporal hypernetwork, the balance parameter Used to adjust the weight of structural similarity value and similarity value. The smaller the absolute value of , the greater the influence of the structural similarity of the hypernetwork on the recommendation results; The larger the absolute value of , the stronger the dominant role of attribute similarity between nodes in cooperative recommendation; They represent the author at time node t x With the author y The structural similarity between the author node, topic node and structure node, They represent the author at time node t x With the author y The attribute similarity between the author node, topic node and structure node.

[0067] For two authors who have no cooperative relationship x, y , the calculated prediction score The larger the value, the more x and author y The greater the probability of a cooperative relationship; when the predicted score is greater than the threshold to be recommended, the author is recommended x and author y Conduct scientific research cooperation.

[0068] This embodiment also proposes a scientific research cooperation recommendation system based on a hypernetwork, including a computer system and a computer program. When the computer program is run by the computer system, the aforementioned scientific research cooperation recommendation method based on a hypernetwork is executed.

[0069] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer readable medium, and the computer program includes a program code for executing the method shown in the flowchart.

[0070] It should be noted that the computer-readable medium shown in the embodiment of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0071] The flowchart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operations of the system, method and computer program product according to various embodiments of the present application. Wherein, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0072] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. The names of these units do not constitute limitations on the units themselves in some cases.

[0073] As another aspect, the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the hypernetwork-based scientific research collaboration recommendation method described in the above embodiment.

[0074] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the hypernetwork-based scientific research cooperation recommendation method described in the above embodiment.

[0075] It should be noted that, although several modules or units of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0076] Through the description of the above implementation methods, it is easy for those skilled in the art to understand that the example implementation methods described here can be implemented by software, or by combining software with necessary hardware. Therefore, the technical solution according to the implementation method of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation method of the present application.

[0077] For those skilled in the art, the specific meanings of the above terms in the present invention can be understood in specific situations; the drawings in the embodiments are used to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0078] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for recommending scientific research cooperation based on a hypernetwork, characterized in that: include: Obtain multi-dimensional relationships between entities from relevant scientific and technological literature in the field, use entities as nodes, multi-dimensional relationships as hyperedges, and establish time labels on the edges to form a multi-dimensional time series hypernetwork; Structural similarity measurement between nodes in multidimensional temporal hypernetworks at different times; Attribute similarity measurement between nodes in multidimensional temporal hypernetworks at different times; For two authors who do not have a cooperative relationship, the cooperation prediction score between the authors is calculated by integrating structural similarity and attribute similarity, and whether to recommend the authors for scientific research cooperation is determined based on the cooperation prediction score.

2. The method for recommending scientific research cooperation based on a hypernetwork according to claim 1, characterized in that: The specific method for forming the multidimensional temporal hypernetwork includes: Perform text segmentation on massive scientific and technological documents to obtain sentence sets; Given a list of entity types, use a large language model to determine the matching probability between sentence text features and different entity types, remove entity types with probabilities less than a preset threshold, and obtain the entity list actually contained in the scientific literature; Given a list of multiple relationship types, use a large language model to determine and obtain the corresponding relationship types of entities in the entity list; Different entities are represented as nodes, multivariate relationships are represented as edges, and time is used as the feature of the edge to establish author subnetworks, subject subnetworks, and institution subnetworks respectively. The subnetworks are sliced ​​according to the time window, and hyperedges are established between nodes of different subnetworks in the same slice to construct a multidimensional temporal hypernetwork.

3. The method for recommending scientific research cooperation based on a hypernetwork according to claim 2, characterized in that: In the multidimensional temporal hypernetwork, the association matrix H Defined as The matrix whose elements are defined as follows: node v With super edge e If the matrix H The corresponding element value is 1, otherwise the corresponding element value is 0; At this point, the node v In time x The salvation Defined as: in, To represent the hyperedge e Is it in time x Appear, For super edge e time tag.

4. The method for recommending scientific research cooperation based on a hypernetwork according to claim 3, characterized in that: In the multidimensional temporal hypernetwork, the weight of node v on hyperedge e is defined as: in, Is a node In scientific literature The weight of the node is assigned by the frequency of the node appearing in the literature. Is a node In scientific literature d The order of frequency of occurrence in It is a document d The number of nodes in is the document set corresponding to the hyperedge e.

5. The method for recommending scientific research cooperation based on a hypernetwork according to claim 1 or 2, characterized in that: The structural similarity measurement method between nodes in the multidimensional temporal hypernetwork at different times includes: For any time point, the weight of the hyperedge corresponding to the node is added to the process of calculating the common neighbor index and resource allocation index between nodes; The entropy weight method is used to determine the weights between the common neighbor index and the resource allocation index to obtain the structural similarity between the nodes at that point in time.

6. The method for recommending scientific research cooperation based on a hypernetwork according to claim 5, characterized in that: The common neighbor index calculation method is: The resource allocation indicator calculation method is: in, For Node x and nodes y In time t The common neighbor index of For Node x and nodes y Resource allocation index at time t, parameter a Used to adjust the influence of the node weight in the hyperedge on the similarity value. , HCN and HRA degenerate into unweighted forms if , the metric is equivalent to the ordinary weight metric; is the weight of node x and node z at time t, defined as follows: in, is a hyperedge set, each hyperedge in the set e All included x and z Two nodes, Represents a hyperedge e Is it in time t Appear, For Node x On the super edge e The weight on For Node z On the super edge e The weight on is the weight of node z and node y at time t, with the same definition.

7. The method for recommending scientific research cooperation based on a hypernetwork according to claim 6, characterized in that: The specific calculation method of the structural similarity is: in, For Node x and nodes y Structural similarity at time t; EWM stands for entropy weight method, which measures the internal differences of each indicator data and determines the weight through information entropy.

8. The method for recommending scientific research cooperation based on a hypernetwork according to claim 1 or 2, characterized in that: The attribute similarity measurement method between nodes in the multidimensional temporal hypernetwork at different times includes: At any time point, the dynamic subject model is used to map the node into a vector form, which is then input into the variational autoencoder to generate the node's attribute feature vector in an unsupervised manner; The Euclidean distance between the attribute feature vectors of the nodes is calculated to obtain the attribute similarity between the nodes at that time point.

9. The method for recommending scientific research cooperation based on a hypernetwork according to claim 8, characterized in that: The cooperative prediction score calculation method is: in, For Author x , y The cooperation prediction score of and Respectively indicate the author x , y Overall structural similarity and overall attribute similarity, , Respectively indicate the author x , y The corresponding author nodes, subject nodes and institution nodes in the multidimensional temporal hypernetwork, the balance parameter Used to adjust the weight of structural similarity value and similarity value. They represent the author at time node t x With the author y The structural similarity between the author node, topic node and structure node, They represent the author at time node t x With the author y The attribute similarity between the author node, topic node and structure node.

10. A scientific research cooperation recommendation system based on a hypernetwork, characterized in that: The method comprises a computer system and a computer program. When the computer program is run by the computer system, the method according to any one of claims 1 to 9 is executed.

Images