Scientific research cooperation prediction method and system based on translation mechanism

Abstract

The invention provides a scientific research cooperation prediction method and system based on a translation mechanism, and the method comprises the steps: constructing a scientific research cooperation network, taking author or mechanism entities as nodes of the scientific research cooperation network, and taking the cooperation and membership between the entities as edges of the scientific research cooperation network; generating a triple based on the contribution degree of the nodes in the scientific research cooperation network; extracting subject information implied by the entity to construct an entity vector; extracting theme information of a description document in the scientific research cooperation network in combination with a translation mechanism to construct a relation vector; and calculating the distance according to the entity vectors of the two entities and the relation vector between the two entities, and completing scientific research cooperation prediction. According to the method, the edge in the Coriolis cooperative network is regarded as a translation process between nodes, and the topic feature of the edge is learned through an LDA topic model and a deep auto-encoder. And aiming at the non-directivity of the scientific research cooperation network, the direction of the edge is determined based on the accumulated contribution degree of the authors, and then the non-interference triple set is generated, so that the prediction capability is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of electronic information technology, and in particular to a scientific research cooperation prediction method and system based on a translation mechanism. Background Art

[0002] Newman first introduced complex network analysis methods into scientific research cooperation networks, and discovered high-impact authors in the network by studying various characteristics of the network, such as clustering coefficient, centrality, cooperation intensity and average distance. Liben-Nowell et al. first proposed the link prediction problem of cooperation networks, using indicators such as Jaccard coefficient, common neighbors and SimRank to calculate the similarity between nodes, and then predict links based on similarity. Sun Xiaoling et al. proposed a cooperation network evolution model driven by interests and events based on network structure and node attributes, and used the model to generate network attributes and patterns for unsupervised link prediction, predicting possible future cooperation relationships. Liu Jing et al., based on the small world and scale-free characteristics of the network, measured the similarity of research interests and connection paths of researchers in the network through cosine distance and Katz indicators, and constructed a weighted prediction model based on link prediction. Wang Zhibing et al. constructed IDF and ICCR indicators based on the similarity of cooperative institution preferences by using the author node attribute characteristics, and weighted them with indicators such as AA, CN, Katz and LP based on the similarity of network topology structure, thus forming 8 prediction indicators of potential cooperative relationships, and finally used the indicators for link prediction. In addition, some researchers introduced multiple features such as authors, research institutions, keywords, references and journals to construct heterogeneous networks based on cooperative networks. Zhang Jinzhu et al. constructed an author-keyword bipartite network based on the association formed by cooperative relationships and keywords, extracted multiple paths on the network to characterize the cooperative relationship between authors, and formed multiple prediction indicators of cooperative relationships based on the calculation method of cooperative intensity. Luong NT et al. constructed an author-document-term-conference network on the DBLP dataset, and used multiple similarity calculation methods and random walk models to predict the potential cooperative relationship between authors. Zhang Jinzhu et al. first constructed a document-author bipartite network and projected it into a cooperative network, then extracted the second-order and third-order path features in the network, and finally used a logistic regression model to learn the contribution of different path features to cooperative relationships. Aike et al. comprehensively considered attributes such as papers, institutions and research interests, constructed sample characteristics from multiple dimensions such as title, number of publications, signature order and time, and proposed a scientific research cooperation potential prediction model based on ensemble learning.

[0003] With the development and application of representation learning technology in natural language processing and complex networks, researchers have begun to introduce representation learning into scientific research cooperation prediction. Zhang Jinzhu et al. used the classic network representation learning method LINE to map the node adjacency information into the node vector, and calculated the similarity between authors based on the node vector. The two authors with higher similarity are more likely to form a cooperative relationship in the future. Yu Chuanming et al. used DeepWalk and Node2vec models to learn the overall topological structure characteristics of the network, and used LINE and SDNE models to learn the local adjacency structure characteristics of the network, and weighted the two features to obtain the vector representation of the node, and finally predicted the relationship based on the logistic regression model. Lin Yuan et al. integrated multiple co-occurrence features such as keyword co-occurrence, author co-occurrence, institution co-occurrence, author-keyword co-occurrence, etc. into the representation learning model, and measured the correlation strength between nodes through the Euclidean distance between node vectors, so as to predict potential cooperation objects. However, the above-mentioned prediction method based on representation learning has certain shortcomings. For example, it only considers node features and ignores the semantic information of text such as titles and abstracts on the edges, resulting in incomplete data analysis and insufficient prediction accuracy. In addition, the prediction task often takes the possibility of cooperation as the research goal, and lacks predictive guidance on the research topics of scientific research cooperation. Summary of the invention

[0004] Aiming at the problems existing in the prior art, a scientific research cooperation prediction method and system based on translation mechanism is provided. The first aspect of the present invention proposes a scientific research collaboration prediction method based on a translation mechanism, comprising: Construct a scientific research cooperation network, in which authors or institutional entities are used as nodes of the scientific research cooperation network, and the cooperation and affiliation between entities are used as edges in the scientific research cooperation network; Generate triples based on the contribution of nodes in the scientific research cooperation network; Extract the implicit subject information of the entity to construct the entity vector; Combined with the translation mechanism, the topic information of the description documents in the scientific research cooperation network is extracted to construct the relationship vector; The distance is calculated based on the entity vectors of two entities and the relationship vector between them to complete the scientific research cooperation prediction.

[0005] As a preferred solution, the specific method for building a scientific research cooperation network is: Perform text segmentation on scientific and technological literature to obtain a sentence set; 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; Taking entities as nodes and the cooperation and affiliation between entities as edges, a scientific research cooperation network is constructed.

[0006] As a preferred solution, the generation of valid triples based on the contribution of nodes in the scientific research cooperation network specifically includes: For edge , where the node u and v represents the author or institution, and edge e represents the cooperation or affiliation between nodes. E is an edge set; Computing Technology Literature d Midpoint u and nodes v Contribution , and then we get the node u and v Relative scientific literature d The contribution of is:

[0007]

[0008] If the node u and v If it appears in N scientific papers at the same time, the cumulative contribution is defined , ; like , then it means the node Contribute more to the relationship than , generating triples ; On the contrary, generate triples ; Generate triplets for all nodes and obtain the triplet set of the scientific research network.

[0009] As a preferred solution, the node u and nodes v Contribution , specifically:

[0010]

[0011] in, Indicates the frequency of occurrence of the entity corresponding to the node in the document.

[0012] As a preferred solution, extracting the topic information implied by the entity to construct the entity vector specifically includes: Input the entity description text into the LDA model, output the text-topic matrix and the topic-word matrix, and use the text-topic matrix to map the entity into a topic distribution vector; The deep autoencoder is used to model the entity-topic distribution features. The deep autoencoder maps the topic distribution vector to a low-dimensional vector representation space through N hidden layers to obtain the entity vector.

[0013] As a preferred solution, extracting the subject information of the description documents in the scientific research cooperation network to construct a relationship vector specifically includes: Establish translation relationships between entity vectors and relationship vectors based on the translation mechanism; The stochastic gradient descent algorithm is used to learn the relationship vectors corresponding to all edges in the scientific research cooperation network.

[0014] As a preferred solution, the translation relationship between the entity vector and the relationship vector is established based on the translation mechanism, specifically including: For any edge , the relationship vector corresponding to this edge Seen as The entity vector of To The translation of entity vectors is formalized as:

[0015] in, is the mapping matrix associated with the entity.

[0016] As a preferred solution, the stochastic gradient descent algorithm is used to learn the relationship vectors corresponding to all edges in the scientific research cooperation network, specifically including: Based on the description document set of scientific and technological literature, the LDA topic model is trained to generate a text-topic matrix, and the text-topic matrix is ​​used to map entities to topic distribution vectors; based on the deep autoencoder, the topic distribution vector is converted into an encoding matrix to obtain the initial entity vector; In each iteration, a small batch of triplets is randomly selected from the triple set S generated by the scientific research cooperation network , where each positive triplet generates a corresponding negative triplet set ; In a pair of positive and negative triplet training, calculate the gradient and update the relationship vector; During the learning process, the objective function is set as:

[0017] in, return x and 0, whichever is greater. is the maximum separation distance between positive and negative triplets, For vector The L1 or L2 distance between For vector The L1 or L2 distance between them; A research collaboration network G Generate a set of positive triples, In u and v Randomly replace the node in the node set with another different node to obtain a negative triplet ,Right now:

[0018] in, It is the entity set of scientific research cooperation network.

[0019] As a preferred solution, the distance is calculated based on the entity vectors of the two entities and the relationship vector between the two entities, specifically including: The distance function is defined as:

[0020] in, For vector The L1 or L2 distance between Representation Node With Node The edge, For Node The entity vector of For Node The entity vector of For edge The corresponding relationship vector; like is less than the preset threshold, it means the node u and nodes v Possibility of scientific research cooperation.

[0021] The second aspect of the present invention proposes a scientific research cooperation prediction system based on a translation mechanism, including a memory and a processor, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the method described in the first aspect.

[0022] Compared with the prior art, the beneficial effects of the above technical solution are as follows: the present invention regards the edges in the scientific cooperation network as the translation process between nodes, and learns the topic features of the edges through the LDA topic model and the deep autoencoder. In view of the undirected nature of the scientific research cooperation network, the direction of the edge is determined based on the author's cumulative contribution, and then a set of non-interference triples is generated, which greatly improves the effect and prediction ability of the representation vector. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Flowchart of the scientific research collaboration prediction method based on translation mechanism.

[0024] Figure 2 It is a schematic diagram of constructing entity vectors in one embodiment of the present invention. DETAILED DESCRIPTION

[0025] 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.

[0026] At present, scientific research cooperation prediction is usually carried out on scientific research cooperation networks. Based on complex network theory, the cooperation prediction problem is converted into a link prediction problem, and the possibility of future cooperation between researchers is calculated. To a certain extent, good results have been achieved, but there are still the following problems: (1) Most existing prediction methods construct prediction indicators based on the overall topological structure of the collaboration network, the local adjacency structure, or the connectivity paths between nodes. When the scale of the network expands, the complexity of the indicator calculation increases exponentially, making it difficult to adapt to the prediction of scientific research collaboration in a big data environment. On the other hand, since the scientific research collaboration network has a power-law distribution, most researchers or research institutions are in the long-tail range, so the current prediction methods mainly target highly productive researchers or institutions with more co-authors, ignoring researchers or institutions with average or low output in the marginal areas of the network.

[0027] (2) Existing prediction methods for scientific research collaboration networks ignore the rich textual semantic information on nodes. Therefore, how to use textual information to automatically extract effective features of the collaboration between researchers and institutions has become an urgent problem to be solved.

[0028] Based on this, the embodiment of the present invention proposes a scientific research cooperation prediction method based on a translation mechanism. For the first problem, the present invention uniformly maps the nodes and edges in the scientific research cooperation network to a low-dimensional representation space, so that each researcher or institution corresponds to a representation vector to solve the long-tail distribution problem; at the same time, in the prediction, the complex indicator calculation is converted into a calculation between low-dimensional vectors, which significantly improves the calculation efficiency and is more suitable for scientific research cooperation prediction in a big data environment. For the second problem, the present invention first uses the LDA topic model to obtain the topic distribution characteristics in the author or institution description text, then constructs the entity vector based on the deep autoencoder, and finally uses the learned relationship vector to form a representation vector to predict the possibility of cooperation between researchers or institutions.

[0029] Please refer to Figure 1 The scientific research collaboration prediction method based on translation mechanism specifically includes: Step 1: Build a scientific research cooperation network.

[0030] In this embodiment, a large language model is mainly used to extract entities such as authors and institutions from field-related scientific papers and patents, and the entities are used as nodes, and the cooperation and affiliation between entities are used as edges to form a scientific research cooperation network. The specific process includes: 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.

[0031] 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.

[0032] 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 ).

[0033] 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.

[0034] After two rounds of question and answer, entities and their relationships can be acquired.

[0035] Therefore, we take entities as nodes and the cooperation and affiliation between entities as edges to construct a scientific research cooperation network. In order to more accurately describe the scientific research cooperation network, relevant definitions and symbols are given here to facilitate subsequent use and description. The network is denoted as , where V is the node set and E is the edge set. , where the node u and v represents the author or institution, and edge e represents the cooperation or affiliation between nodes.

[0036] Step 2: Generate triples based on the contribution of nodes in the scientific research cooperation network.

[0037] The scientific research cooperation network constructed is an undirected network, and the relationship between nodes is reflexive, that is, Two triplets will be generated that interfere with each other and Based on this, this embodiment proposes a method for generating valid triples based on node contribution to eliminate interference between triplets.

[0038] In this embodiment, the following assumptions are made about nodes and contributions: (1) Nodes with different frequencies of occurrence have different contributions to scientific literature.

[0039] (2) The number of times two nodes co-occur in scientific literature is different, and the corresponding relationship is also different. The relationship is described by the accumulation of contribution.

[0040] Therefore, scientific literature is defined Midpoint u and nodes v Contribution :

[0041]

[0042] in, Indicates the frequency of occurrence of the entity corresponding to the node in the document.

[0043] So Node u and v Relative scientific literature d The contribution of If the node u and v exist N If they appear in the same scientific literature, the cumulative contribution is defined .if , description node Contributes more to the relationship than the node , so the triple is generated ; On the contrary, generate triples .in, , .

[0044] Generate triplets for all nodes and obtain the triplet set of the scientific research network.

[0045] Step 3: Extract the implicit topic information of the entity to construct the entity vector.

[0046] In this embodiment, the LDA (Latent Dirichlet Allocation) model is used to construct entity vectors. After the entity description text is input into LDA, the text-topic matrix is ​​output. and the topic-word matrix ,use The node / entity v can be mapped to a topic distribution vector But the topic distribution vector The dimension of is the same as the number of topics and the value of each dimension (topic probability) is non-negative. Directly using it as an entity vector will affect the flexibility and computability of the vector representation.

[0047] This embodiment further uses a stacked autoencoder (SAE) to model entity-topic distribution features. Based on this, an entity vector is generated in an unsupervised way, and its structure is as follows Figure 2 As shown, the autoencoder can Convert it into a feature vector of any dimension so that the topic probability distribution feature is mapped to a point in the Euclidean space, which is convenient for the subsequent calculation of the similarity between entities using the distance metric function.

[0048] Please continue to refer to Figure 2 , the topic distribution vector After being input to the encoder, the encoder distributes the entity-topic probability through N hidden layers. Mapped into a low-dimensional vector representation space to get the encoding (that is, get the entity vector ), the decoder also reconstructs the encoding into entity topic probability distribution through N hidden layers. The specific expression is as follows:

[0049]

[0050] The deep autoencoder training objective can be minimized and The network parameters are learned by layer-by-layer training so that the output encoding contains the characteristics of the original entity-topic distribution.

[0051] Step 4: Combine the translation mechanism to extract the topic information of the documents describing the scientific research cooperation network and construct a relationship vector.

[0052] Based on the translation model TransE, this embodiment assumes that the nodes and edges in the scientific research cooperation network G have a translation relationship in the representation space. , the relationship vector As a vector from the entity To entity vector The translation mechanism can be formally expressed as:

[0053] in, is the mapping matrix associated with the entity.

[0054] In order to make the triple To match the translation mechanism as much as possible, this embodiment uses the distance function to calculate the vector The L1 or L2 distance between them is as follows:

[0055] Among them, L1 refers to L1 regularization, which is the sum of the absolute values ​​of each element in the vector; L2 refers to L2 regularization, which is the 1 / 2 power of the sum of the squares of each element; the closer the calculated distance is, the closer the triple is. With formula The higher the degree of match.

[0056] After establishing the translation relationship, this embodiment uses a stochastic gradient descent algorithm to learn the relationship vectors corresponding to all edges in the scientific research cooperation network, and uses dropout to prevent overfitting, specifically including: Based on the description document set of scientific and technological literature, the LDA topic model is trained to generate a text-topic matrix, and the text-topic matrix is ​​used to map entities to topic distribution vectors; based on the deep autoencoder, the topic distribution vector is converted into an encoding matrix to obtain the initial entity vector; In each iteration, a small batch of triplets is randomly selected from the triple set S generated by the scientific research cooperation network , where each positive triplet generates a corresponding negative triplet set ; In a pair of positive and negative triplet training, the gradient is calculated and the representation vector is updated, where the representation vector includes the entity vector and the relationship vector.

[0057] In the iterative learning process, in order to enhance the discrimination of the representation vector and maximize the interval distance between positive and negative samples in the process of learning the relationship vector, the objective function is set as follows in this embodiment:

[0058] in, Indicates return x and 0, whichever is greater. is the maximum separation distance between positive and negative triplets, For vector The L1 or L2 distance between For vector The L1 or L2 distance between them; A research collaboration network G Generate a set of positive triples, In u and v Randomly replace the node in the node set with another different node to obtain a negative triplet ,Right now:

[0059] in, It is the entity set of scientific research cooperation network.

[0060] Table 1 SCPM training algorithm

[0061] Through the SCPM training algorithm recorded in Table 1, the entity vector and relationship vector can be updated based on stochastic gradient descent.

[0062] Step 5: Prediction of scientific research cooperation.

[0063] After representing the author entities and relationships in the scientific research cooperation network as a real-valued vector of fixed length, scientific research cooperation prediction is transformed into a calculation problem between vectors. For nodes u and v that have no previous relationship, the distance is calculated using the distance function:

[0064] in, For vector The L1 or L2 distance between Representation Node With Node The edge, For Node The entity vector of For Node The entity vector of For edge The corresponding relationship vector; like is less than the preset threshold, it means the node u and nodes v There is the possibility of scientific research cooperation and the formation of relationship e.

[0065] This embodiment also proposes a scientific research cooperation prediction model SCPM constructed based on the aforementioned scientific research cooperation prediction method based on the translation mechanism, through which cooperation prediction between authors or institutions can be realized.

[0066] This embodiment also provides a scientific research cooperation prediction system based on a translation mechanism, including a memory and a processor, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the aforementioned scientific research cooperation prediction method based on the translation mechanism.

[0067] 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.

[0068] 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.

[0069] 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.

[0070] 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, in some cases, constitute limitations on the units themselves.

[0071] As another aspect, the present application also provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the scientific research collaboration prediction method based on the translation mechanism described in the above embodiment.

[0072] 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 scientific research collaboration prediction method based on the translation mechanism described in the above embodiment.

[0073] 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.

[0074] 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.

[0075] 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.

[0076] 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 scientific research collaboration prediction method based on translation mechanism, characterized in that: include: Construct a scientific research cooperation network, in which authors or institutional entities are used as nodes of the scientific research cooperation network, and the cooperation and affiliation between entities are used as edges in the scientific research cooperation network; Generate triples based on the contribution of nodes in the scientific research cooperation network; Extract the implicit subject information of the entity to construct the entity vector; Combined with the translation mechanism, the topic information of the description documents in the scientific research cooperation network is extracted to construct the relationship vector; The distance is calculated based on the entity vectors of two entities and the relationship vector between them to complete the scientific research cooperation prediction.

2. The scientific research cooperation prediction method based on translation mechanism according to claim 1, characterized in that: The specific method of building a scientific research cooperation network is: Perform text segmentation on scientific and technological literature to obtain a sentence set; 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; Taking entities as nodes and the cooperation and affiliation between entities as edges, a scientific research cooperation network is constructed.

3. The scientific research cooperation prediction method based on translation mechanism according to claim 1, characterized in that: The generation of valid triples based on the contribution of nodes in the scientific research cooperation network specifically includes: For edge , where the node u and v represents the author or institution, and edge e represents the cooperation or affiliation between nodes. E is an edge set; Computing Technology Literature d Midpoint u and nodes v Contribution , and then we get the node u and v Relative scientific literature d The contribution of is: If the node u and v If it appears in N scientific papers at the same time, the cumulative contribution is defined , ; like , then it means the node Contribute more to the relationship than , generating triples ; On the contrary, generate triples ; Generate triplets for all nodes and obtain the triplet set of the scientific research network.

4. The scientific research cooperation prediction method based on translation mechanism according to claim 3 is characterized in that: The node u and nodes v Contribution , specifically: in, Indicates the frequency of occurrence of the entity corresponding to the node in the document.

5. The scientific research cooperation prediction method based on translation mechanism according to claim 1, characterized in that: The extracting of the topic information implied by the entity to construct the entity vector specifically includes: Input the entity description text into the LDA model, output the text-topic matrix and the topic-word matrix, and use the text-topic matrix to map the entity into a topic distribution vector; The deep autoencoder is used to model the entity-topic distribution features. The deep autoencoder maps the topic distribution vector to a low-dimensional vector representation space through N hidden layers to obtain the entity vector.

6. The scientific research cooperation prediction method based on translation mechanism according to claim 1, characterized in that: The step of extracting the topic information of the document describing the scientific research cooperation network and constructing the relationship vector specifically includes: Establish translation relationships between entity vectors and relationship vectors based on the translation mechanism; The stochastic gradient descent algorithm is used to learn the relationship vectors corresponding to all edges in the scientific research cooperation network.

7. The scientific research cooperation prediction method based on translation mechanism according to claim 6, characterized in that: The translation relationship between the entity vector and the relationship vector is established based on the translation mechanism, specifically including: For any edge , the relationship vector corresponding to this edge Seen as The entity vector of To The translation of entity vectors is formalized as: in, is the mapping matrix associated with the entity.

8. The scientific research cooperation prediction method based on translation mechanism according to claim 7, characterized in that: The stochastic gradient descent algorithm is used to learn the relationship vectors corresponding to all edges in the scientific research cooperation network, specifically including: Based on the description document set of scientific and technological literature, the LDA topic model is trained to generate a text-topic matrix, and the text-topic matrix is ​​used to map entities to topic distribution vectors; based on the deep autoencoder, the topic distribution vector is converted into an encoding matrix to obtain the initial entity vector; In each iteration, a small batch of triplets is randomly selected from the triple set S generated by the scientific research cooperation network , where each positive triplet generates a corresponding negative triplet set ; In a pair of positive and negative triplet training, calculate the gradient and update the relationship vector; During the learning process, the objective function is set as: in, return x and 0, whichever is greater. is the maximum separation distance between positive and negative triplets, For vector The L1 or L2 distance between For vector The L1 or L2 distance between them; A research collaboration network G Generate a set of positive triples, In u and v Randomly replace the node in the node set with another different node to obtain a negative triplet ,Right now: in, It is the entity set of scientific research cooperation network.

9. The scientific research cooperation prediction method based on translation mechanism according to claim 1, characterized in that: The calculating of the distance according to the entity vectors of the two entities and the relationship vector between the two entities specifically includes: The distance function is defined as: in, For vector The L1 or L2 distance between Representation Node With Node The edge, For Node The entity vector of For Node The entity vector of For edge The corresponding relationship vector; like is less than the preset threshold, it means the node u and nodes v There is the possibility of scientific research cooperation.

10. A scientific research cooperation prediction system based on translation mechanism, characterized in that: The invention comprises a memory and a processor, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the method according to any one of claims 1 to 9.

Images