Scientific research cooperation relation prediction method and device based on Horkes point process

Abstract

The invention discloses a scientific research cooperation relationship prediction method and device based on a Horkes point process, and the method comprises the steps: obtaining a scientific research cooperation information set, and constructing a scientific research cooperation network; constructing an interaction loss function based on the representation vector of the two-hop historical partner node set and the representation vector of the scientific researchers; obtaining a model total loss function in combination with the interaction loss, and training and optimizing the dynamic graph neural network by using the total loss function; obtaining representation vectors of all scientific research workers in the scientific research cooperation network by adopting the dynamic graph neural network after training optimization; and calculating a similarity score between different scientific research workers in the scientific research cooperation network, and predicting a future cooperation object according to the similarity score. The representation vector of the scientific researchers, which is obtained by the method provided by the invention, effectively contains local topological information and historical interaction information of a certain scientific researchers in the scientific research collaborative network, and when the representation vector is used for calculating the similarity score, the prediction result is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of cooperation prediction, and in particular to a method and device for predicting scientific research cooperation relationships based on a Hawkes point process. Background Art

[0002] In recent years, with the continuous deepening of scientific research and the increasing complexity of cooperation networks, for a researcher, finding suitable partners among many researchers can effectively improve scientific research efficiency and save a lot of time. In addition, multidisciplinary and multi-field cooperation can help generate new ideas and innovations and promote scientific research breakthroughs. For a cooperative relationship prediction model, link prediction tasks are usually used for simulation. In a link prediction task, the final prediction result is affected by the representation vector of the scientific research collaboration network graph composed of researchers. The more information the representation vector of the researcher node in the scientific research collaboration network carries, the higher the link prediction accuracy. When representing the scientific research collaboration network graph composed of researchers and their cooperation relationships, such as TGAT, TREND, HTNE and other dynamic graph representation methods, they can well aggregate the information of the neighboring worker nodes around the researchers in the scientific research collaboration network to obtain the representation vector of the researchers.

[0003] However, there are two problems in the above-mentioned graph representation method: (i) The topological structure around a certain researcher node in the scientific research collaboration network changes over time without considering the impact of the information carried in this change process on the future cooperation behavior of the researcher; (ii) When applying dynamic graph neural networks to aggregate the historical cooperation information of a certain researcher, the impact of the information carried in the cooperation on the future cooperation behavior of the researcher is not considered. In view of this, how to consider the incentive effect of historical cooperation and carry more abundant historical cooperation information in the representation vector of the researcher to improve the accuracy of the prediction results has become a technical problem that needs to be solved at present. Summary of the invention

[0004] In this embodiment, a method, device, electronic device and storage medium for predicting scientific research cooperation relationships based on the Hawkes point process are provided to consider the incentive effect of historical cooperation in the representation vector of scientific researchers and carry richer historical cooperation information to improve the accuracy of prediction results.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting scientific research cooperation relationships based on a Hawkes point process, the method comprising: Obtain scientific research cooperation information collection and build a scientific research collaboration network; Selecting a number of elements from the scientific research collaboration network to construct a training set, and obtaining positive and negative samples for any scientific researcher in the training set; Select the researchers as the starting node, perform biased time-order walk, and obtain the k two-hop historical collaborator node sets of the starting node. The representation vector of the two-hop historical collaborator node set is obtained based on the embedding method of the Hawkes point process. The historical cooperation information of the researchers is aggregated using a dynamic graph neural network to obtain the representation vector of the researchers. The interactive loss function is constructed based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher; The dynamic loss function is constructed by using the change in the number of collaborators of scientific researchers within a certain period of time. The total loss function of the model is obtained by combining the interaction loss. The total loss function is used to train and optimize the dynamic graph neural network. A dynamic graph neural network after training and optimization is used to obtain representation vectors of all scientific researchers in the scientific research collaboration network; similarity scores between different scientific researchers in the scientific research collaboration network are calculated, and future cooperation partners are predicted based on the similarity scores.

[0006] In an optional embodiment, obtaining a scientific research cooperation information set and building a scientific research cooperation network includes: Get scientific research cooperation information collection , where each triple represents a piece of cooperation information, Representing scientific researchers i and researchers j exist Cooperation is generated at all times; Constructing a scientific research collaboration network based on the cooperation information ,in, Represents a collection of scientific researchers. For cooperation collection, is a timestamp collection, Represents the adjacency matrix of the scientific research collaboration network.

[0007] In an optional embodiment, a plurality of elements are selected from the scientific research collaboration network to construct a training set, and for any scientific researcher in the training set, obtaining the positive and negative samples thereof includes: Select S elements from the set of scientific researchers to form a training set; Get any researcher in the training set i The set of collaborators before time t; Choose to work with this researcher i Researchers who have collaborated j As a positive sample, not with the researcher i Collaborative researchers l as negative samples.

[0008] In an optional embodiment, a scientific researcher is selected as a starting node, and a biased time-series walk is performed to obtain a set of k two-hop historical collaborator nodes of the starting node, and a representation vector of the two-hop historical collaborator node set is obtained based on an embedding method of a Hawkes point process; the historical cooperation information of the scientific researcher is aggregated using a dynamic graph neural network, and the representation vector of the scientific researcher is obtained, including: Researchers i , Researchers j , Researchers l As the starting node, a biased time sequence walk is performed to obtain the scientific research workers i , Researchers j , Researchers l The k two-hop historical collaborator node sets are obtained by using the embedding method based on Hawkes point process. i , scientific researchers j , scientific researchers l The representation vector of the set of k two-hop historical sampling partner nodes; Using dynamic graph neural network to study scientific researchers i , scientific researchers j , Researchers l The representation vectors of historical collaborators that meet the time conditions are aggregated to obtain the scientific researcher i , scientific researchers j , scientific researchers l The representation vector of .

[0009] In an optional embodiment, constructing an interactive loss function based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher includes: According to researchers i , scientific researchers j The representation vector of the k two-hop historical sampling collaborator node set and the scientific researcher i , scientific researchers j The representation vector constructs the positive sample interaction loss function; According to researchers i , scientific researchers l The representation vector of the k two-hop historical sampling collaborator node set and the scientific researcher i , scientific researchers l The representation vector constructs the negative sample interaction loss function; The interaction loss function is obtained by adding the positive sample interaction loss function and the negative sample interaction loss function.

[0010] In an optional embodiment, a dynamic loss function is constructed by using the change in the number of collaborators of a scientific researcher within a certain period of time, and a total loss function of the model is obtained by combining the interaction loss. The total loss function is used to train and optimize the dynamic graph neural network, including: According to researchers i The change in the number of partners over a certain period of time constructs a dynamic loss; The dynamic loss is combined with the interaction loss to obtain the total loss function of the model, and the model training is optimized to obtain a dynamic graph neural network after training and optimization.

[0011] In an optional embodiment, a trained and optimized dynamic graph neural network is used to obtain representation vectors of all scientific researchers in the scientific research collaboration network; similarity scores between different scientific researchers in the scientific research collaboration network are calculated, and future cooperation partners are predicted based on the similarity scores, including: The similarity scores between different researchers in the scientific research collaboration network are calculated and used as a mapping of the future cooperation probability of researchers in the real world. For any researcher, the researcher with the highest similarity score is selected as the most likely future cooperation partner.

[0012] Compared with the prior art, the scientific research cooperation relationship prediction method based on Hawkes point process of the present invention has the following beneficial effects: By extracting the historical sampling collaborator set of a certain researcher in the scientific research collaboration network, and using the encoding method based on the Hawkes process to obtain the representation vector of the historical sampling collaborator; using the dynamic graph neural network enhanced by the Hawkes point process, the historical collaborator representation vectors are aggregated to obtain the representation vector of the researcher. The representation vector of the researcher obtained by the method proposed in the present invention effectively contains the local topological information and historical interaction information of a certain researcher in the scientific research collaboration network. When the representation vector is used to calculate the similarity score, the prediction result is more accurate; at the same time, the conditional probability intensity in the Hawkes point process is constructed by using the historical sampling set representation vector of the researcher and the self-representation vector obtained by the dynamic graph neural network, and the dynamic graph neural network parameters are optimized, which effectively quantifies the incentive effect of the historical cooperation of a certain scientific research collaborator on the future cooperative behavior of the researcher.

[0013] In a second aspect, an embodiment of the present invention provides a scientific research cooperation relationship prediction device based on Hawkes point process, comprising: Construction module, used to obtain scientific research cooperation information collection and build scientific research collaboration network; A sampling module, used to select a number of elements from the scientific research collaboration network to construct a training set, and obtain positive and negative samples for any scientific researcher in the training set; The calculation module is used to select a researcher as the starting node, perform biased time-series walks, obtain a set of k two-hop historical collaborator nodes of the starting node, and obtain the representation vector of the two-hop historical collaborator node set based on the embedding method of the Hawkes point process; use a dynamic graph neural network to aggregate the historical cooperation information of the researcher to obtain the representation vector of the researcher; and construct an interactive loss function based on the representation vector of the two-hop historical collaborator node set and the representation vector of the researcher; The loss training module is used to construct a dynamic loss function using the changes in the number of collaborators of scientific researchers within a certain period of time, and combine the interaction loss to obtain the total loss function of the model. The total loss function is used to train and optimize the dynamic graph neural network. The prediction module is used to obtain the representation vectors of all scientific researchers in the scientific research collaboration network using a trained and optimized dynamic graph neural network; calculate the similarity scores between different scientific researchers in the scientific research collaboration network, and predict future cooperation partners based on the similarity scores.

[0014] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a processor, a communication interface, a memory and a bus, wherein the processor, the communication interface and the memory communicate with each other through the bus, and the processor can call logical instructions in the memory to execute the steps of the method provided in the first aspect.

[0015] In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for predicting scientific research cooperation relationships based on the Hawkes point process as described in the first aspect.

[0016] Compared with the prior art, the beneficial effects of the scientific research cooperation relationship prediction device, electronic device and storage medium based on Hawkes point process of the present invention are the same as the scientific research cooperation relationship prediction method based on Hawkes point process described in the first aspect, so they will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 Flow chart of a method for predicting scientific research cooperation relationships based on Hawkes point process in an embodiment of the present invention; Figure 2A schematic diagram of a method for obtaining a historical sampling set representation vector of scientific researchers in a scientific research collaboration network provided by an embodiment of the present invention; Figure 3 A schematic diagram of a method for obtaining a representation vector of a scientific researcher in a scientific research collaboration network provided by an embodiment of the present invention; Figure 4 It is a structural block diagram of a scientific research cooperation relationship prediction device based on Hawkes point process in an embodiment of the present invention; Figure 5 4 is a structural block diagram of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0020] Unless otherwise defined, the technical terms or scientific terms involved in this application shall have the general meaning understood by people with general skills in the technical field to which this application belongs. The words "one", "a", "the", "these" and the like in this application do not indicate a quantitative limitation, and they may be singular or plural. The terms "include", "comprise", "have" and any variants thereof involved in this application are intended to cover non-exclusive inclusions; for example, a process, method and system, product or device comprising a series of steps or modules (units) is not limited to the listed steps or modules (units), but may include unlisted steps or modules (units), or may include other steps or modules (units) inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" may mean: A exists alone, A and B exist at the same time, and B exists alone. Generally, the character " / " indicates that the objects associated with each other are in an "or" relationship. The terms "first", "second", "third", etc. in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.

[0021] In an embodiment of the present invention, a method for predicting scientific research cooperation relationships based on Hawkes point process is provided. Figure 1 is a flow chart of the method for predicting scientific research cooperation relationship based on Hawkes point process of the present invention, such as Figure 1 As shown, the process includes the following steps: S1. Obtain scientific research cooperation information collection and build a scientific research collaboration network; Obtaining scientific research cooperation information collection and building a scientific research collaboration network include: Combination Figure 2 and Figure 3 As shown, obtain the scientific research cooperation information collection , where each triple represents a piece of cooperation information. For example, Representing scientific researchers i and researchers j exist Cooperation is generated at all times; Constructing a scientific research collaboration network based on the cooperation information ,in, Represents a collection of scientific researchers. For cooperation collection, is a timestamp collection, represents the adjacency matrix of the scientific research collaboration network, n is the number of scientific researchers, if the number of scientific researchers is i and researchers j If there has been cooperation, X [ i,j ] is 1 if yes, otherwise it is 0.

[0022] S2. Selecting a number of elements from the scientific research collaboration network to construct a training set, and obtaining positive and negative samples for any scientific researcher in the training set; Specifically, a number of elements are selected from the scientific research collaboration network to construct a training set, and for any scientific researcher in the training set, the positive and negative samples thereof are obtained including: Select S elements from the set of scientific researchers to form a training set; Get any researcher in the training set i The set of collaborators before time t; Choose to work with this researcher i Researchers who have collaborated j As a positive sample, not with the researcher i Collaborative researchers l as negative samples.

[0023] For example, from the scientific research collaboration network G A collection of scientific researchers Select S elements from the set K (training set) as training samples for the model (dynamic graph neural network).

[0024] For any researcher , obtain its end to t Before the moment, collaborators gathered , select researchers with whom you have collaborated As positive samples, researchers who have not collaborated with them As negative samples; Specifically, from the probability function Randomly select a researcher l As a negative sample, the probability distribution function is as follows: , Among them, deg(v) represents the total number of collaborators of a certain researcher v in the scientific research collaboration network. This function indicates that the selection probability of the researcher v is proportional to the 3 / 4 power of its degree; According to the probability function Calculate the selection probability of each node; then normalize the probabilities of these nodes so that their sum is 1; then generate a random number between 0 and 1 to represent a position in the probability distribution; starting from the first node, accumulate the normalized probability of each node until a certain node is reached, and the cumulative probability is greater than or equal to the random number, then select the node as the negative sample node.

[0025] As mentioned above, given a researcher i , researchers who have interacted with it j As positive samples, researchers who have not interacted with them l as negative samples.

[0026] S3. Select a researcher as the starting node, perform biased time-series walk, obtain a set of k two-hop historical collaborator nodes of the starting node, and obtain the representation vector of the two-hop historical collaborator node set based on the embedding method of the Hawkes point process; use a dynamic graph neural network to aggregate the historical cooperation information of the researcher to obtain the representation vector of the researcher; Select the researchers as the starting node, perform biased time-series walks, and obtain the k two-hop historical collaborator node sets of the starting node. The representation vector of the two-hop historical collaborator node set is obtained based on the embedding method of the Hawkes point process. Use the dynamic graph neural network to aggregate the historical cooperation information of the researchers, and the representation vector of the researchers is obtained, including: Researchers i , scientific researchers j , scientific researchers l As the starting node, a biased time sequence walk is performed to obtain the scientific research workers i , scientific researchers j , scientific researchers l The k two-hop historical collaborator node sets are obtained by using the embedding method based on Hawkes point process. i , scientific researchers j , scientific researchers lThe representation vector of the set of k two-hop historical sampling partner nodes; Using dynamic graph neural network to study scientific researchers i , scientific researchers j , scientific researchers l The representation vectors of historical collaborators that meet the time conditions are aggregated to obtain the scientific researcher i , scientific researchers j , scientific researchers l The representation vector of .

[0027] Research worker nodes As the starting node, a biased time sequence walk is performed to obtain The set of k (k defaults to 10) two-hop historical partner nodes ; The nodes are obtained by using the embedding method based on Hawkes point process The representation vector of the set of k two-hop historical sampling partner nodes ; (The following operation takes the i node as an example, and the acquisition method of other node sets is the same) Specifically, the sampling method is as follows (the sampling method of the one-hop and two-hop historical partners is the same, but the historical cooperation time of the first hop must be less than the current time, and the historical cooperation time of the second hop must be less than the historical cooperation time of the first hop): , , in, For manually set parameters (default value: =1e-6, = 1.0), controlling the temporal and spatial bias of sampling, Representing scientific researchers i With researchers a the time at which the interaction occurred; For a node at time The degree, With researchers i The set of scientific researchers who have collaborated before time t (including time t); exp is an exponential function used to calculate the probability of a certain scientific researcher being selected; the numerator is the sampling probability of the current node, and the denominator is the sum of the sampling probabilities of all collaborators.

[0028] Based on the Hawkes weighting method, different weights are assigned to collaborators generated at different times. The weight calculation is shown in the following formula: , in, For the current moment, Moments for other researchers to collaborate with researcher i; is the weight parameter, which is used to adjust the influence of time difference; The function is an activation function, which is often used to map input elements to values ​​between 0 and 1, representing complex nonlinear relationships; For researchers i A historical sampling of collaborators , and perform weighted processing, as shown in the following formula: , in , , For researchers i The weights corresponding to the historical collaborators of is an element in the historical sampling collaborator set, and is the number of the scientific researcher; Indicates multiplication; Represented as a linear layer function in a neural network, it performs a linear transformation on the input elements, usually represented as a matrix multiplication; Represents a weighted sampling set; After weighting the elements in the k two-hop sampling sets, the scientific research work i The k two-hop historical sampling collaborator sets are encoded as follows: , , Obtaining researchers through linear functions i The representation vectors of k two-hop historical sampling partners ;in It represents the processing function for the elements in the historical sampling collaborator set of researcher i. If the element is greater than >2.5, it is set to 3, and other sizes remain unchanged.

[0029] Dynamic graph neural network enhanced by Hawkes point process for scientific researchers The representation vectors of the direct historical collaborators that meet the time condition (i.e., the historical cooperation time must be less than the current time) are aggregated to obtain the scientific research workers The representation vector (The following operations are based on i Node as an example): At any time t, the d-dimensional time representation vector obtained by the time encoding function is as shown below: , in For the parameters that need to be trained, simulate the fluctuation characteristics of time; It is a sine and cosine function, which can well capture the periodic characteristics of scientific research cooperation; The representation vector of the moment is concatenated with the initial representation vector of the historical collaborator X[k] obtained from the adjacency matrix X of the scientific research collaboration network to obtain the scientific researcher i The historical cooperation representation vector is as follows: , The effects of cooperation at different time points on the future cooperation behavior of researchers are different, and the incentive effect decreases over time; therefore, the softmax function is introduced to calculate the Hawkes weight coefficient, as shown in the following formula: , in, is a trainable parameter that controls the relative importance of cooperation at different time points. is the timestamp of historical cooperation; through softmax( ) function is normalized to finally obtain the time weight coefficient ; Based on the above weight coefficient and the above historical cooperation representation vector, the scientific researcher i Each historical cooperation representation vector is weighted by Hawkes to represent the historical interaction, as shown in the following formula: , Researchers i Each weighted historical collaboration representation vector is aggregated using a dynamic graph neural network to obtain the scientific researchers i The representation vector As follows: , in, represents the d-dimensional space representation of historical cooperation, represents the trainable neural network parameters, is the time weight; It is also an activation function, and the Sigmoid function is usually used.

[0030] The interactive loss function is constructed based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher; Specifically, the interactive loss function is constructed based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher, including: According to researchers i , scientific researchers j The representation vector of the k two-hop historical sampling collaborator node set and the scientific researcher i , scientific researchersj The representation vector constructs the positive sample interaction loss function; According to researchers i , Researchers l The representation vector of the k two-hop historical sampling collaborator node set and the scientific researcher i , Researchers l The representation vector constructs the negative sample interaction loss function; The interaction loss function is obtained by adding the positive sample interaction loss function and the negative sample interaction loss function.

[0031] More specifically, researchers i , Researchers l The k two-hop historical sampling collaborator node set representation vector and scientific researchers i Node, scientific researcher l The representation vector of the node constructs the negative sample interaction loss (the following operation takes the calculation of the interaction loss of the positive sample pair as an example, and the calculation process of the negative sample is the same): The influence of historical incentives on the conditional probability intensity during the Hawkes point calculation is shown in the following formula: , in For researchers i The representation vector of the historical sampling set before (including) time t, For researchers i The representation vector of the historical sampling set before (including) time t; is the fully connected layer function, is the parameter vector of the fully connected layer; finally, the incentive effect generated by historical cooperation is obtained through the activation function; The basic intensity in the time conditional probability intensity of the Hawkes point process is calculated using the representation vectors of researchers i and j, as shown in the following formula: , in, represents the representation vector of researcher i at time t, is the representation vector of researcher j at time t; is the fully connected layer, is the fully connected layer parameter, which maps the node interaction features to the excitation intensity by inputting the vector of squared difference of parameter features; Represents the operation of subtracting the elements of the input vector and taking the square; Based on the data obtained above, construct a i , scientific researchers j The conditional probability strength of the generated link is shown in the following formula: , in is a model parameter that adjusts the weight ratio of historical incentives to basic strength; Based on the above conditional probability strength, the interaction loss is calculated, and the calculation formula is shown as follows: , Where Q is the number of negative samples, ensuring the adaptability of the model to low-degree nodes; the conditional probability strength of negative samples as well as Function can make the model easier to solve and ensure that the global optimal solution is found.

[0032] S4. Use the change in the number of collaborators of scientific researchers within a certain period of time to construct a dynamic loss function, combine the interaction loss to obtain the total loss function of the model, and use the total loss function to train and optimize the dynamic graph neural network; The dynamic loss function is constructed by using the change in the number of collaborators of scientific researchers within a certain period of time, and the total loss function of the model is obtained by combining the interaction loss. The total loss function is used to train and optimize the dynamic graph neural network, including: According to researchers i The change in the number of partners over a certain period of time constructs a dynamic loss; The dynamic loss is combined with the interaction loss to obtain the total loss function of the model, and the model training is optimized to obtain a dynamic graph neural network after training and optimization.

[0033] Using scientific researchers i The change in the number of collaborators within a certain period of time constructs a dynamic loss, and the two losses are added together as the total loss function; Researchers i The representation vector of is input into the node dynamic estimator, and the computing researchers i Dynamic loss , as shown below: , in, For researchers i The representation vector obtained by the dynamic graph neural network, Indicates that at a certain moment, researchers i How many people do you collaborate with? It is a type of artificial neural network with a linear transformation function (linear function) and a nonlinear activation function (such as the Sigmoid function) to model complex nonlinear relationships. Based on scientific researchers i Dynamic loss and interaction loss, construct global loss: , in Dynamic loss for scientific researchers, is the interaction loss.

[0034] S5. Use the trained and optimized dynamic graph neural network to obtain the representation vectors of all scientific researchers in the scientific research collaboration network; calculate the similarity scores between different scientific researchers in the scientific research collaboration network, and predict future cooperation partners based on the similarity scores.

[0035] The representation vectors of all researchers in the scientific research collaboration network are obtained by using a trained and optimized dynamic graph neural network; the similarity scores between different researchers in the scientific research collaboration network are calculated, and the future cooperation partners are predicted based on the similarity scores, including: The similarity scores between different researchers in the scientific research collaboration network are calculated and used as a mapping of the future cooperation probability of researchers in the real world. For any researcher, the researcher with the highest similarity score is selected as the most likely future cooperation partner.

[0036] For any researcher , repeat the above operations to reduce the global loss and obtain the optimal dynamic graph neural network parameters.

[0037] Through the trained dynamic graph neural network, researchers can i , the final representation vector of the researcher p is as follows: , Represents the historical collaboration set of scientific researchers.

[0038] The similarity scores between different scientific researcher nodes in the scientific research collaboration network are calculated as a mapping of the probability of cooperation between scientific researchers in the real world. The higher the score, the higher the probability of future cooperation, as shown in the following formula (taking sample pair i, p as an example): , , Represent the representation vectors of researcher i and researcher p obtained through the trained dynamic graph neural network; subtract the input vectors and square them modulo; is the probability of cooperation between the two at time t.

[0039] The present invention extracts the set of historical sampling collaborators of a certain researcher in the scientific research collaboration network, and uses the encoding method based on the Hawkes process to obtain the representation vector of the historical sampling collaborators; adopts the dynamic graph neural network based on the Hawkes point process enhancement to aggregate the representation vectors of the historical collaborators to obtain the representation vector of the researcher. The representation vector of the researcher obtained by the method proposed in the present invention effectively contains the local topological information and historical interaction information of a certain researcher in the scientific research collaboration network. When the representation vector is used to calculate the similarity score, the prediction result is more accurate; at the same time, the conditional probability intensity in the Hawkes point process is constructed using the representation vector of the historical sampling set of the researcher and the self-representation vector obtained by the dynamic graph neural network, and the dynamic graph neural network parameters are optimized, which effectively quantifies the incentive effect of the historical cooperation of a certain scientific research collaborator on the future cooperative behavior of the researcher.

[0040] The embodiment of the present invention also provides a device for predicting scientific research cooperation relationships based on the Hawkes point process, which is used to implement the above method embodiment, and the description has been made no further. The terms "module", "unit", "subunit", etc. used below can implement a combination of software and / or hardware of predetermined functions. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware or a combination of software and hardware is also possible and conceived.

[0041] like Figure 4 As shown, Figure 4 : is a structural block diagram of a scientific research cooperation relationship prediction device based on Hawkes point process in the present invention, the device comprises: Building module 101, used to obtain a scientific research cooperation information set and build a scientific research cooperation network; The sampling module 102 is used to select a number of elements from the scientific research collaboration network to construct a training set, and obtain positive and negative samples for any scientific researcher in the training set; The calculation module 103 is used to select a scientific researcher as a starting node, perform biased time-series walk, obtain a set of k two-hop historical collaborator nodes of the starting node, obtain a representation vector of the two-hop historical collaborator node set based on the embedding method of the Hawkes point process; use a dynamic graph neural network to aggregate the historical cooperation information of the scientific researcher to obtain the representation vector of the scientific researcher; and construct an interactive loss function based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher; The loss training module 104 is used to construct a dynamic loss function by using the change in the number of collaborators of the scientific researcher within a certain period of time, obtain the total loss function of the model by combining the interaction loss, and use the total loss function to train and optimize the dynamic graph neural network; The prediction module 105 is used to obtain the representation vectors of all scientific researchers in the scientific research collaboration network using the trained and optimized dynamic graph neural network; calculate the similarity scores between different scientific researchers in the scientific research collaboration network, and predict future cooperation partners based on the similarity scores.

[0042] Figure 5 A structural block diagram of an electronic device provided by an embodiment of the present invention, such as Figure 5 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the following method: Obtain scientific research cooperation information collection and build a scientific research collaboration network; Selecting a number of elements from the scientific research collaboration network to construct a training set, and obtaining positive and negative samples for any scientific researcher in the training set; Select the researchers as the starting node, perform biased time-order walk, and obtain the k two-hop historical collaborator node sets of the starting node. The representation vector of the two-hop historical collaborator node set is obtained based on the embedding method of the Hawkes point process. The historical cooperation information of the researchers is aggregated using a dynamic graph neural network to obtain the representation vector of the researchers. The interactive loss function is constructed based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher; The dynamic loss function is constructed by using the change in the number of collaborators of scientific researchers within a certain period of time. The total loss function of the model is obtained by combining the interaction loss. The total loss function is used to train and optimize the dynamic graph neural network. A dynamic graph neural network after training and optimization is used to obtain representation vectors of all scientific researchers in the scientific research collaboration network; similarity scores between different scientific researchers in the scientific research collaboration network are calculated, and future cooperation partners are predicted based on the similarity scores.

[0043] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0044] An embodiment of the present invention further provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method provided in each of the above embodiments is implemented, for example, including: Obtain scientific research cooperation information collection and build a scientific research collaboration network; Selecting a number of elements from the scientific research collaboration network to construct a training set, and obtaining positive and negative samples for any scientific researcher in the training set; Select the researchers as the starting node, perform biased time-order walk, and obtain the k two-hop historical collaborator node sets of the starting node. The representation vector of the two-hop historical collaborator node set is obtained based on the embedding method of the Hawkes point process. The historical cooperation information of the researchers is aggregated using a dynamic graph neural network to obtain the representation vector of the researchers. The interactive loss function is constructed based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher; The dynamic loss function is constructed by using the change in the number of collaborators of scientific researchers within a certain period of time. The total loss function of the model is obtained by combining the interaction loss. The total loss function is used to train and optimize the dynamic graph neural network. A dynamic graph neural network after training and optimization is used to obtain representation vectors of all scientific researchers in the scientific research collaboration network; similarity scores between different scientific researchers in the scientific research collaboration network are calculated, and future cooperation partners are predicted based on the similarity scores.

[0045] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting scientific research cooperation relationships based on Hawkes point process, characterized in that: The scientific research cooperation relationship prediction method comprises: Obtain scientific research cooperation information collection and build a scientific research collaboration network; Selecting a number of elements from the scientific research collaboration network to construct a training set, and obtaining positive and negative samples for any scientific researcher in the training set; Select the researchers as the starting node, perform biased time-order walk, and obtain the k two-hop historical collaborator node sets of the starting node. The representation vector of the two-hop historical collaborator node set is obtained based on the embedding method of the Hawkes point process. The historical cooperation information of the researchers is aggregated using a dynamic graph neural network to obtain the representation vector of the researchers. The interactive loss function is constructed based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher; The dynamic loss function is constructed by using the change in the number of collaborators of scientific researchers within a certain period of time. The total loss function of the model is obtained by combining the interaction loss. The total loss function is used to train and optimize the dynamic graph neural network. A dynamic graph neural network after training and optimization is used to obtain representation vectors of all scientific researchers in the scientific research collaboration network; similarity scores between different scientific researchers in the scientific research collaboration network are calculated, and future cooperation partners are predicted based on the similarity scores.

2. The method for predicting scientific research cooperation relationships based on Hawkes point process according to claim 1, characterized in that: Obtaining scientific research cooperation information collection and building a scientific research collaboration network include: Get scientific research cooperation information collection , where each triple represents a piece of cooperation information, Representing scientific researchers i and researchers j exist Cooperation is generated at all times; Constructing a scientific research collaboration network based on the cooperation information ,in, Represents a collection of scientific researchers. For cooperation collection, is a timestamp collection, Represents the adjacency matrix of the scientific research collaboration network.

3. The method for predicting scientific research cooperation relationships based on Hawkes point process according to claim 2, characterized in that: Selecting a number of elements from the scientific research collaboration network to construct a training set, and obtaining positive and negative samples for any scientific researcher in the training set includes: Select S elements from the set of scientific researchers to form a training set; Get any researcher in the training set i The set of collaborators before time t; Choose to work with this researcher i Researchers who have collaborated j As a positive sample, not with the researcher i Collaborative researchers l as negative samples.

4. The method for predicting scientific research cooperation relationships based on Hawkes point process according to claim 3, characterized in that: Select the researchers as the starting node, perform biased time walk, and obtain the k two-hop historical collaborator node sets of the starting node. The representation vector of the two-hop historical collaborator node set is obtained based on the embedding method of Hawkes point process. The historical cooperation information of scientific researchers is aggregated using dynamic graph neural networks, and the representation vectors of scientific researchers are obtained, including: Researchers i , Researchers j , scientific researchers l As the starting node, a biased time sequence walk is performed to obtain the scientific research workers i , scientific researchers j , scientific researchers l The k two-hop historical collaborator node sets are obtained by using the embedding method based on Hawkes point process. i , Researchers j , Researchers l The representation vector of the set of k two-hop historical sampling partner nodes; Using dynamic graph neural network to study scientific researchers i , Researchers j , Researchers l The representation vectors of historical collaborators that meet the time conditions are aggregated to obtain the scientific researcher i , Researchers j , Researchers l The representation vector of .

5. The method for predicting scientific research cooperation relationships based on Hawkes point process according to claim 4, characterized in that: The interactive loss function constructed based on the representation vector of the two-hop historical collaborator node set and the representation vector of the scientific researcher includes: According to researchers i , Researchers j The representation vector of the k two-hop historical sampling collaborator node set and the scientific researcher i , Researchers j The representation vector constructs the positive sample interaction loss function; According to researchers i , Researchers l The representation vector of the k two-hop historical sampling collaborator node set and the scientific researcher i , Researchers l The representation vector constructs the negative sample interaction loss function; The interaction loss function is obtained by adding the positive sample interaction loss function and the negative sample interaction loss function.

6. The method for predicting scientific research cooperation relationships based on Hawkes point process according to claim 5, characterized in that: The dynamic loss function is constructed by using the change in the number of collaborators of scientific researchers within a certain period of time, and the total loss function of the model is obtained by combining the interaction loss. The total loss function is used to train and optimize the dynamic graph neural network, including: According to researchers i The change in the number of partners over a certain period of time constructs a dynamic loss; The dynamic loss is combined with the interaction loss to obtain the total loss function of the model, and the model training is optimized to obtain a dynamic graph neural network after training and optimization.

7. The method for predicting scientific research cooperation relationships based on Hawkes point process according to claim 1, characterized in that: The representation vectors of all researchers in the scientific research collaboration network are obtained by using a dynamic graph neural network after training and optimization; Calculate the similarity scores between different researchers in the scientific research collaboration network, and predict future cooperation partners based on the similarity scores, including: The similarity scores between different researchers in the scientific research collaboration network are calculated and used as a mapping of the future cooperation probability of researchers in the real world. For any researcher, the researcher with the highest similarity score is selected as the most likely future cooperation partner.

8. A scientific research cooperation relationship prediction device based on Hawkes point process, characterized in that: include: Construction module, used to obtain scientific research cooperation information collection and build scientific research collaboration network; A sampling module, used to select a number of elements from the scientific research collaboration network to construct a training set, and obtain positive and negative samples for any scientific researcher in the training set; The calculation module is used to select a researcher as the starting node, perform biased time-series walks, obtain a set of k two-hop historical collaborator nodes of the starting node, and obtain the representation vector of the two-hop historical collaborator node set based on the embedding method of the Hawkes point process; use a dynamic graph neural network to aggregate the historical cooperation information of the researcher to obtain the representation vector of the researcher; and construct an interactive loss function based on the representation vector of the two-hop historical collaborator node set and the representation vector of the researcher; The loss training module is used to construct a dynamic loss function using the changes in the number of collaborators of scientific researchers within a certain period of time, and combine the interaction loss to obtain the total loss function of the model. The total loss function is used to train and optimize the dynamic graph neural network. The prediction module is used to obtain the representation vectors of all scientific researchers in the scientific research collaboration network using a trained and optimized dynamic graph neural network; calculate the similarity scores between different scientific researchers in the scientific research collaboration network, and predict future cooperation partners based on the similarity scores.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method for predicting scientific research cooperation relationships based on the Hawkes point process as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting scientific research cooperation relationships based on the Hawkes point process as described in any one of claims 1 to 7 are implemented.

Images