Enterprise technology partner recommendation method and system based on time sequence heterogeneous network

Abstract

The invention discloses an enterprise technology cooperative partner recommendation method based on a time sequence heterogeneous network, and the method comprises the steps: obtaining mechanism data and patent data, and constructing the time sequence heterogeneous network fusing the semantic and structural relationship among mechanisms, patents and technical fields; generating a meta-path instance from the time sequence heterogeneous network in a random walk manner; screening a time constraint path by means of the Horkes process theory, and calculating the path influence between the node pairs according to the time, semantic and structural association of the time constraint path; the path influence is weighted by combining the embedding similarity, the structure attention mechanism and the time attention mechanism, and the semantic attention mechanism further adjusts the contribution of different edge types; and through positive and negative sample loss function optimization attention mechanism and node embedding, generating a recommendation list of potential technical partners of the target enterprise.

Description

Technical Field

[0001] The present invention relates to the field of matching and recommending potential technology partners for enterprises, and particularly to a method and system for recommending potential technology partners of a target enterprise based on a temporal heterogeneous network. Background Art

[0002] In the context of the current global competition intensification and the significant improvement of technological complexity, enterprises face many limitations in the process of technological innovation and R & D. The cost and risk of R & D have increased significantly, making it difficult for a single enterprise to bear the pressure of innovation. Therefore, seeking technology partners has become an important way to break through the technological innovation bottleneck of enterprises.

[0003] Technology partners can not only improve R & D efficiency through collaborative innovation, but also reduce the risks and uncertainties of enterprise technology development. However, enterprises also face dilemmas when seeking partners, such as information asymmetry of partners, difficulty in matching technology requirements and capabilities, high cost of trust establishment, etc. These factors seriously restrict the effective development of technological cooperation. In this context, studying a scientific and efficient method for recommending potential technology partners of enterprises is not only a practical need, but also provides strategic support for promoting the sustainable development of global technological cooperation.

[0004] "A multi-level fusion based decision support system for academic collaborator recommendation" published in Knowledge-Based Systems in June 2020 introduced a collaborator recommendation model DRACoR (Deep learning and Random walk based Academic Collaborator Recommender) based on multi-level fusion. Feature vectors were extracted from the author's paper title and abstract using LDA and Doc2Vec respectively to represent the author nodes, and the possibility of cooperation was predicted by calculating the vector similarity. Although the proposed model shows that the combination of models based on topic distribution and co-authorship network can significantly improve the effectiveness of academic cooperation. However, there may be many potential reasons behind the cooperation of two researchers. In addition, insufficient consideration was given to the personal profile information of the researchers.

[0005] "Collaborator recommendation in heterogeneous bibliographic networks using random walks", published in Information Retrieval Journal in August 2017, improved a random walk algorithm to retrieve relevant partners on a weighted heterogeneous bibliographic network. First, by removing the citing document nodes, a heterogeneous network with multiple types of nodes and links was constructed with a simplified network structure. Then, two importance measures were used to weight the edges in the network. Finally, we employed random walks to retrieve relevant authors and output an ordered recommendation list based on the ranking scores. Although the experiments verified the effectiveness of the method and its promising performance in collaborator prediction, there is still much work to be done in mining valuable information from author profiles for prediction.

[0006] "A Novel Approach to Enterprise Technical Collaboration: Recommending R&D Partners Through Technological Similarity and Complementarity", published in Journal of Informetrics in August 2024, proposed a method for collaborator recommendation by combining technological similarity and complementarity. A collaborator recommendation model enhanced by technological similarity and complementarity was introduced, a heterogeneous enterprise collaboration network was constructed, and a loss function was designed. The model effectively integrated the features of technological similarity and complementarity, enabling the neural network to capture and clarify the non-linear and multi-dimensional relationships in enterprise collaboration. Although this technique provides a novel approach for collaborator recommendation, emphasizes the necessity of multi-dimensional integration of technological characteristics, and provides valuable decision support for enterprises to select collaborators, there are still deficiencies in terms of interpretability.

[0007] Chinese Patent Application CN116304308A, published on February 16, 2023, provides a method for recommending R&D partners based on the technological innovation knowledge situation super network. A technological innovation knowledge situation model was constructed, the dimensions of the technological innovation knowledge situation were determined, multi-dimensional knowledge situation information was extracted from the technological innovation project data, the node features and hyper-edge features were determined, and an improved super network Bayesian inference method was adopted to generate the R&D partner recommendation results. This method can effectively improve the utilization efficiency of knowledge, effectively mine the deep relationships between knowledge contexts in the super network, and thus significantly improve the recommendation accuracy and efficiency of R&D partner recommendation. This method provides an idea for enterprise cooperation and recommendation, but it does not consider the time dimension and dynamic changes, and ignores the time dynamic factors of cooperation relationships.

[0008] In summary, the current research has proposed enterprise partner recommendation algorithms and systems based on homogeneous networks and heterogeneous networks, but each has certain limitations. Homogeneous networks usually only consider single types of nodes and edges. Although some research has introduced the attribute information of nodes, overall, their information expression ability is still limited. When migrating the partner recommendation method based on homogeneous networks to the field of enterprise technology partner recommendation, it is difficult to make full use of the diverse information between enterprises, thereby restricting the further improvement of the recommendation effect. Although dynamic heterogeneous networks can capture the dynamic influence between heterogeneous nodes, existing methods still have problems such as the loss of continuous dynamic information between snapshots of the cooperation network and the adverse impact of the time granularity of snapshot division on the results. Therefore, a dynamic embedding learning method applicable to cooperation networks is urgently needed, hoping that it can capture and reflect the dynamic changes of the network in real time, pay attention to the subtle changes of the network over time, and more accurately reflect the evolution process of enterprise cooperation relationships. Summary of the Invention

[0009] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a method for recommending potential technical partners of target enterprises based on a temporal heterogeneous network. Through the node embedding algorithm of the temporal heterogeneous network, it integrates the industrial and commercial information of institutions, retains the evolution process of the network, so as to improve the accuracy of partner recommendation within a future fine-grained time range.

[0010] According to one aspect of the specification of the present invention, there is provided a method for recommending enterprise technical partners based on a temporal heterogeneous network, including: Obtain institutional data and patent data, and construct a temporal heterogeneous network that integrates the semantic and structural relationships among institutions, patents, and technical fields; Randomly walk in the temporal heterogeneous network to generate meta-path instances; With the help of the Hawkes process theory, screen time-constrained paths, and calculate the path influence between node pairs according to the time, semantic, and structural associations of the time-constrained paths; The path influence is weighted by combining embedding similarity, structural attention mechanism, and time attention mechanism, and the semantic attention mechanism further adjusts the contributions of different edge types; Optimize the attention mechanism and node embedding through the positive and negative sample loss function to generate a recommendation list of potential technical partners of the target enterprise.

[0011] As a further technical solution, the construction of the temporal heterogeneous network further includes: Expand the relevant attributes of institutions and patents, and construct nodes in the temporal heterogeneous network based on the expanded data.

[0012] As a further technical solution, the generation of the meta-path instances includes: According to the essential motivation of institutional technical cooperation, four meta-paths are designed; Use the random walk method to extract meta-path instances by following the four meta-paths respectively.

[0013] As a further technical solution, the method further includes: Based on the extracted distance constraint edges, analyze and refine the influence of neighbor nodes into three aspects: semantics, time, and structure, and assign attention mechanisms to them respectively to assist in calculating node similarity.

[0014] As a further technical solution, the method further includes: Use the Euclidean distance to measure the attribute similarity between nodes, and process it through the exponential form of the softmax function to become the weights of each meta-path.

[0015] As a further technical solution, the method further includes: Adopt the form of negative sampling to consider positive and negative samples to design a loss function, combine the adaptive moment estimation in the gradient descent method to optimize the model, and analyze the recommendation quality of enterprise technical partners of the overall model.

[0016] According to one aspect of the specification of the present invention, a recommendation system for enterprise technical partners based on a temporal heterogeneous network is provided, including: A data collection module, used to obtain the institutional data and patent data required for constructing the temporal heterogeneous network, and through detailed node attribute extraction and transformation, provide rich and detailed representations for the entities in the network; A network construction module, used to describe the dynamic technical cooperation relationship between organizations, integrate the business information of institutions, so as to improve the accuracy of partner recommendation within a fine-grained time range; A prediction module, used to capture the continuous evolution between the main bodies in the enterprise technical cooperation network by using the Hawkes method, mine various potential associations between enterprise technical cooperations through the meta-path and random walk methods, and optimize the model parameters through various attention mechanisms; A recommendation module, used to optimize the attention mechanism and node embedding through the positive and negative sample loss function, and generate potential technical partners of the target enterprise.

[0017] According to one aspect of the specification of the present invention, a recommendation device for enterprise technical partners based on a temporal heterogeneous network is provided, including: at least one processor, at least one memory, and a communication interface; wherein, the processor, the memory, and the communication interface communicate with each other; the memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the above-mentioned recommendation method for enterprise technical partners based on a temporal heterogeneous network.

[0018] According to one aspect of the specification of the present invention, there is provided a non-transitory computer-readable storage medium storing computer instructions that cause a computer to execute the enterprise technology partner recommendation method based on a temporal heterogeneous network as described above.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention integrates a variety of technologies. According to the Hawkes process theory, a method is proposed that can effectively capture the continuous evolution between entities in the enterprise technology cooperation network. While considering the influence of the occurrence of neighboring events, it can avoid information loss caused by slice discretization. The present invention innovatively proposes a deep semantic extraction method applicable to the temporal heterogeneous network of enterprise technology cooperation, enhances the expressive ability of the network, more precisely mines potential cooperation opportunities between enterprises, improves the efficiency of enterprises in finding partners, promotes cooperation and communication between enterprises, and drives innovation and development in the industry.

[0020] 2. Based on the Hawkes process theory, the present invention proposes a method that can effectively capture the continuous evolution between entities in the enterprise technology cooperation network. While considering the influence of the occurrence of neighboring events, it can avoid information loss caused by slice discretization.

[0021] 3. The present invention innovatively proposes a deep semantic extraction method applicable to the temporal heterogeneous network of enterprise technology cooperation. Based on four types of technology cooperation association meta-paths and combined with the random walk method, it mines various potential associations between enterprise technology cooperations.

[0022] 4. When embedding nodes in the temporal heterogeneous network, the present invention designs an attention mechanism to optimize model parameters from three perspectives of structure, semantics, and time sequence for the proposed time constraint path and distance constraint edge, so as to improve the recommendation accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings used in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0024] Figure 1 It is a schematic flowchart of the enterprise technology partner recommendation method based on a temporal heterogeneous network provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of the temporal heterogeneous network of enterprise technology cooperation provided by an embodiment of the present invention; Figure 3Flowchart of the meta-path selection algorithm based on the Hawkes process provided by an embodiment of the present invention; Figure 4 Schematic diagram of an enterprise technology partner recommendation system based on a temporal heterogeneous network provided by an embodiment of the present invention; Figure 5 Schematic diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0025] The terms "including" and "having" and any variations thereof in the description and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] The block diagrams shown in the drawings are only functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices. The flowcharts shown in the drawings are only exemplary illustrations and do not necessarily include all the content and operations / steps, nor do they have to be executed in the described order. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to the actual situation.

[0027] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. In addition, the technical features in each embodiment or a single embodiment provided by the present invention can be combined with each other arbitrarily to form a new technical solution. This combination is not restricted by the order of steps and / or the pattern of structural composition, but must be based on what can be achieved by those of ordinary skill in the art. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the protection scope required by the present invention.

[0028] An embodiment of the present invention discloses an enterprise technology partner recommendation method based on a temporal heterogeneous network for matching and recommending potential technology partners of a target enterprise. The method specifically includes: S1: Collect and extract nodes and their attributes for constructing a temporal heterogeneous network. The data required for constructing the temporal heterogeneous network comes from two sources: organizational data and patent data. This step also includes expanding the node attributes of the collected organizations and patents.

[0029] S2: Use the expanded node attributes and their relationships to construct a temporal heterogeneous network, screen the classification number nodes of patents, and set a timestamp representing the specific time when the relationship is established for each edge.

[0030] S3: Use meta-paths to mine multiple technological cooperation paths between organizations in the temporal heterogeneous network. And incorporate the influence of time-proximal event pairs on node edges based on the Hawkes process. Calculate the embedding similarity between nodes using the Euclidean distance, and refine the influence of neighbor nodes into three aspects: semantics, time, and structure, and assign attention mechanisms to them respectively to assist in the calculation of node similarity.

[0031] S4: Sort the similarities between the target organization and other organizations from large to small to generate a recommended list of technological partners for the target organization.

[0032] As a preferred embodiment, as Figure 1 shown, the example of the present invention is described from four aspects: data collection and node attribute extraction, temporal heterogeneous network construction, node embedding, and potential enterprise partner recommendation.

[0033] In the example of the present invention, the data collection and node attribute extraction aspect includes data collection, node attribute extraction, and expansion, which are used to extract and expand the node attributes of institutional data and patent data. Specifically, in this aspect, relevant patent data and institutional data in a specific industry field are selected to expand the relevant attributes of institutions and patents. The organizational data includes the organization name, while the patent data covers the patent names held by these organizations and the corresponding technical fields (IPC classification numbers). For institutional attributes, the industrial and commercial information of enterprises is collected, including the province where the enterprise is located, the prefecture-level city where it is located, and the business scope. For institutions such as universities and research institutes, their dominant discipline information is used to replace the industrial and commercial information.

[0034] In the data collection step of the embodiment of the present invention, data collection and analysis are carried out based on the patent data within a certain industry. Patent retrieval is based on the classification of emerging strategic industries, and the specific industry is the new energy vehicle industry. The data comes from the Patsnap database, and the patent types are limited to "invention applications" and "authorized inventions". The legal status is limited to "authorized" and "published", and the receiving office is China. The patent data content includes the publication number, application number, title, abstract, current assignee, application date, IPC classification number, and patents cited within three years. The geographical location and business scope of the patent assignee come from the Qichacha database.

[0035] In the node attribute extraction and expansion step of the embodiments of the present invention, a variety of natural language processing technologies are adopted. For organizational attributes, business information of organizations is collected, including the province where the company is located, the prefecture-level city where it is located, the business scope, etc. For organizations such as universities and research institutes, the business information is replaced with information on their advantageous disciplines. For provinces and prefecture-level cities, Word2Vec is directly used to convert them into d1-dimensional vectors. For the business information of each institution, first, TextRank is used to extract keywords from the text, and then Word2Vec is used to generate d2-dimensional vectors. For patent attributes, the abstract text of each patent is collected, and BERT is used to classify the coarse-grained technical field to which the patent belongs. Finally, the classification result of each patent is represented as a D (D = d1 + d2)-dimensional vector. Through the extraction and conversion of such node attributes, rich and detailed representations of the entities in the network are provided, laying a solid foundation for the subsequent stages of the model.

[0036] In the embodiments of the present invention, the construction of the temporal heterogeneous network and node embedding are used to construct the temporal heterogeneous network and node embedding based on the data obtained from the data collection and node attribute extraction and expansion links. As Figure 2 shown, specifically, the nodes in the temporal heterogeneous network constructed based on this data include organizations (represented by organization names), patents (represented by patent application numbers), and technical fields (represented by IPC classification numbers). For the multiple IPC classification numbers involved in each patent, a "three-three combination" is performed, that is, any three IPCs are randomly and non-repeatedly combined from all the IPC classification numbers of each patent, and only the combinations that appear more than 2 times in all patents are retained as valid IPC classification number nodes. The edges in the temporal heterogeneous network are all undirected edges, and there is only one edge between any two nodes. Among them, the edge between an organization and a patent reflects the publication behavior of the patent, the edge between a patent and a technical field represents their subordinate relationship, and the edge between patents depicts the citation relationship between patents. Each edge is attached with a timestamp t indicating the specific time when the relationship is established, providing a basis for the analysis of the subsequent dynamic evolution of the network.

[0037] In the embodiments of the present invention, after constructing the temporal heterogeneous network in this link, a node embedding method is proposed. First, based on the research on inter-organizational technical cooperation, 4 meta-paths are designed, and multiple technical cooperation paths between organizations in the temporal heterogeneous network are mined using the meta-paths. Next, while considering the influence of the semantics and structure of the paths, the influence of time-proximal events based on the Hawkes process is combined. Through the action of three designed attention mechanisms and loss functions, the vector representation of the nodes in the temporal heterogeneous network is optimized in the low-dimensional space.

[0038] In the embodiments of the present invention, when designing the meta - paths, four meta - paths are designed according to the essential motivation of institutional technical cooperation. The first meta - path is: Organization - Patent - Organization (OPO). This meta - path captures the direct cooperation relationship between organizations, that is, the close connection formed by organizations through jointly applying for patents. Jointly applying for patents usually means that organizations have in - depth cooperation in technological R & D, resource investment, and innovation achievements, so the possibility of their future continued cooperation is also high. The second meta - path is: Organization - Patent - Patent - Organization (OPPO). This meta - path focuses on exploring the citation relationship between the patents held by organizations. Even if there is no direct cooperation between two organizations, if the patent of one organization is cited by the patent of another organization, it indicates that the technical achievements of the former have had a positive impact on the latter, thus forming an indirect cooperation relationship in terms of technology dissemination and innovation evolution. The third meta - path is: Organization - Patent - Technical Field - Patent - Organization (OPIPO). This meta - path focuses on the technical fields to which the patents held by organizations belong, and reveals the potential cooperation possibilities between organizations through the commonalities of technical fields. This not only shows that organizations have similarities in market positioning and R & D strategies, but also indicates potential cross - institutional cooperation opportunities in the future. The fourth meta - path is: Organization - Patent - Organization - Patent - Organization (OPOPOP). This meta - path reflects the network structure characteristics of triadic closure, that is, the multilateral cooperation relationship formed through an intermediary organization. In reality, two organizations may not directly jointly apply for patents, but they may respectively cooperate with a third - party organization, thus forming a connection in resource sharing.

[0039] To obtain specific semantic information and temporal information in the temporal heterogeneous network, we use the random - walk method to extract meta - path instances respectively following the four meta - paths. Oi represents the organization, Pi represents the patent, and Ti represents the technical field. Meta - path instances of different colors correspond to Figure 3 the different meta - paths shown on the right. For each node, we set a parameter m to control the number of random walks starting from this node.

[0040] In practical applications, generating meta - path instances only based on the meta - path length is limited for node embedding. Therefore, a parameter n is set to control the length of the meta - path instances. Assuming the length of the meta - path is k, based on the meta - path instance generated for the first time, using the end - point node as the starting node to conduct random walks again, and so on for (n - 1) times, to form a meta - path instance with a length of (k - 1)*n + 1.

[0041] Considering that both patent nodes and enterprise nodes have node attributes, this paper uses the concatenation of the d1-dimensional vector extracted in the previous step and the d2-dimensional vector as the initialization vector for the enterprise node, the extracted D-dimensional vector as the vector for the patent node, and a randomly initialized D-dimensional vector as the vector for the technology field node. After completing the initialization of the node embedding, this paper iteratively optimizes the vector through a subsequent attention mechanism.

[0042] For a temporally heterogeneous network with edge r i Any two nodes and , we record the time when it occurs as t j We iterate over all the Meta-path instances, filter all meta-path instances where the maximum time of the edges is less than or equal to t j Meta-path instances. Arrange the edges of these meta-path instances from near to far according to their average time, and take the first n meta-path instances as time-constrained paths. In the time-constrained path, set the nodes and Subsequent analysis is performed for the intermediate window length w, and the retained time-constrained paths are called distance-constrained edges.

[0043] Whether two nodes form an edge is influenced not only by the node's own attributes but also by the attributes of adjacent nodes. In this step, the present invention analyzes this influence based on the distance-constrained edges extracted in the previous step, distilling the influence of adjacent nodes into three dimensions: semantic, temporal, and structural. An attention mechanism is assigned to each dimension to assist in calculating node similarity.

[0044] In the embodiment of the present invention, the Euclidean distance is used to calculate the attribute similarity between nodes. The Euclidean distance is a node similarity measurement method that shows greater stability in high-dimensional data. and ,set up and The vectors are represented as and Then, their node embedding similarity is defined as:

[0045] where β p,d and β q,d Respectively and The d-th dimension element of ES p,q The larger the value, the and The greater the possibility of forming an edge between them.

[0046] In the embodiments of the present invention, in this step, a semantic attention mechanism, a temporal attention mechanism, and a structural attention mechanism are proposed to conduct a comprehensive analysis from different perspectives, including the influence of various edge attributes, the influence of temporally proximal events, and the influence of topological structures.

[0047] For the temporal influence, a parameter τ(t−tj) is specified to represent the time difference between the occurrence time t i (excluding the target edge r) of each edge r in the distance-constrained edges j and the occurrence time t of r. Then, the temporal influence of r i on r is calculated, expressed as the product of the node embedding similarity and τ(t−tj). Therefore, the temporal influence of r i on r is defined as:

[0048] where S is the set of all edges that satisfy the distance constraint except the target edge r, modeled using an exponential function and defined as . Here, is a trainable parameter related to the node, used to adjust the temporal decay effect. Therefore, can be further expressed as:

[0049] For the semantic influence, taking the meta-path OPIPO as an example, the corresponding time-constrained paths have different influences on the edges formed between patents and domains. Specifically, the "application" relationship between an organization and a patent and the "include" relationship between a technical domain and a patent have different influences on the formation of the edges between the patent and the technical domain.

[0050] To optimize the influence of different edge types on the target edge, in the embodiments of the present invention, the influence weights of different edge types are randomly initialized to ensure that their sum is equal to 1. Then, an exponential softmax function is used to process these weights to obtain the weights of each meta-path. And the weight corresponding to the edge r i is denoted as SeIr i . Therefore, considering the semantic attention mechanism, the influence of the target node can be expressed as:

[0051] where SeIr i can be optimized through the semantic attention mechanism.

[0052] For the structural influence, in the time-constrained path, theoretically, an edge r iThe closer it is to the target edge r, the greater its impact on r. In the embodiments of the present invention, the edge hop count is used to represent this distance in this step. Similar to the semantic attention mechanism, let o ri,r be represented as the edge hop count from edge ri to edge i, and the corresponding weight is represented as StI(o ri,r ).

[0053] Under the combined action of the temporal, semantic, and structural attention mechanisms, and the similarity of the final nodes on the temporal constraint path can be expressed as:

[0054] To optimize the temporal, semantic, and structural attention mechanisms, as well as the vectorized representation of the nodes, in the embodiments of the present invention, a loss function is designed in this step, considering positive and negative samples in the form of negative sampling. The expression of the loss function is as follows:

[0055] where P represents the set of positive sample pairs, N represents the set of negative sample pairs, and represent the similarity calculation results of positive and negative samples respectively. The sigmoid function maps the similarity of the final nodes after the exponential function transformation to a probability, denoted by σ. Here, the number of positive and negative samples is the same. After calculating the loss function, in the embodiments of the present invention, adaptive moment estimation is combined with gradient descent to optimize the model.

[0056] In the embodiments of the present invention, for the enterprise technology partner recommendation step, specifically, after iterating the parameters and node vectors, the negative Euclidean distance between organizational nodes is calculated to represent the probability of their possible cooperation, and the similarities between the target organization and other organizations are sorted from large to small to generate a technical partner recommendation list for the target organization. And the Recall@k, Precision@k, nDCG@k, and HR@k parameters are used to evaluate the quality of the enterprise technology partner recommendation of the overall model.

[0057] In the embodiments of the present invention, to verify the recommendation performance of the proposed model, four metrics, namely Precision@k, Recall@k, nDCG@k, and Hits@k, are used to evaluate the FTP recommendation results. Among them, Precision@k is used to evaluate the accuracy of the recommendation results, reflecting the proportion of correctly predicted ones among the top k recommendations. Recall@k focuses on the coverage ability of the top k recommendation results for the prediction of actual enterprise technology partners. nDCG@k focuses on the rationality of the recommendation ranking, whether the recommendation results are prior to the actual FTPs. Hits@k tests whether there are valid results in the top k recommendation lists. The calculation methods of these metrics are as follows:

[0058] Among them, p represents the target company, and R k (p) represents the set of the top k recommended partners of p, and T(p) represents the set of the actual technical partners of p. represents an indicator function, which is 1 when the condition in (·) is satisfied, and 0 otherwise. ri represents the node at the ith position in the recommendation list.

[0059] The implementation basis of each embodiment of the present invention is achieved through programmed processing by a device with processor functions. Therefore, in engineering practice, the technical solutions and their functions of each embodiment of the present invention are encapsulated into various modules. Based on this actual situation, on the basis of the above embodiments, an embodiment of the present invention provides an enterprise technical partner recommendation system based on a temporal heterogeneous network, and this system is used to execute the enterprise technical partner recommendation method based on a temporal heterogeneous network in the above method embodiments.

[0060] See Figure 4 , this system includes: a data collection module, which is used to obtain the institutional data and patent data required to construct a temporal heterogeneous network, and through detailed node attribute extraction and transformation, provides rich and detailed representations for the entities in the network; a network construction module, which is used to describe the dynamic technical cooperation relationships between organizations, integrates the business information of institutions, and improves the accuracy of partner recommendation within a fine-grained time range; a prediction module, which is used to effectively capture the continuous evolution between the main bodies in the enterprise technical cooperation network by using the Hawkes method, mines various potential associations between enterprise technical cooperations through the meta-path and random walk methods, and optimizes the model parameters through various attention mechanisms to improve the model prediction accuracy; a recommendation module, which is used to optimize the attention mechanism and node embedding through positive and negative sample loss functions to generate potential technical partners of the target enterprise.

[0061] The enterprise technical partner recommendation system based on a temporal heterogeneous network provided by the embodiment of the present invention adopts Figure 4 several modules in, integrates a variety of technologies, and proposes a novel algorithm for embedding nodes in a temporal heterogeneous network. This model integrates the business information of institutions, retains the evolution process of the network, and improves the accuracy of partner recommendation within a future fine-grained time range.

[0062] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The difference is only in setting corresponding functional modules, and its principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art, on the basis of the above system embodiments, refer to the specific technical solutions in other method embodiments, obtain corresponding technical means by combining technical features, and the technical solutions constituted by these technical means, and on the premise of ensuring the practicability of the technical solutions, improve the modules in the above system embodiments to obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments. For example: Based on the content of the above system embodiments, as a preferred embodiment, in the enterprise technology partner recommendation system based on a temporal heterogeneous network provided by the embodiments of the present invention, the data collection module is further configured to execute the following instructions: For the collected institution and patent attributes, perform expansion, that is, supplement the industrial and commercial information or advantageous discipline information of the institution, and use BERT to classify the coarse-grained technical field to which the patent belongs; Convert the supplemented institution and patent attributes into vectors using Word2Vec.

[0063] Based on the content of the above system embodiments, as a preferred embodiment, in the enterprise technology partner recommendation system based on a temporal heterogeneous network provided by the embodiments of the present invention, the network construction module is further configured to execute the following instructions: Arbitrarily combine any three non-repeating IPCs in the classification number set of each patent, and only retain the combinations that appear more than once in all patents as valid IPC classification nodes; In the temporal heterogeneous network, nodes represent organizations (represented by names), patents (represented by patent application numbers), and technical fields (represented by IPC classification numbers); The edges between organizations and patents reflect patent applications, the edges between patents and technical fields represent subordination relationships, and the edges between patents represent citation relationships; Each edge is accompanied by a timestamp t, marking the establishment time of this relationship, thus providing a basis for future analysis of the dynamic evolution of the temporal heterogeneous network.

[0064] Based on the content of the above system embodiments, as a preferred embodiment, in the enterprise technology partner recommendation system based on a temporal heterogeneous network provided by the embodiments of the present invention, the prediction module is further configured to execute the following instructions: On the basis of the research on inter-organizational technical cooperation, 4 meta-paths are designed, and multiple technical cooperation paths between organizations in the temporal heterogeneous network are mined using the meta-paths; While considering the semantic and structural impacts of paths, the influence of time-proximal events based on the Hawkes process is combined; By designing three attention mechanisms and a loss function, the vector representation of nodes in the temporal heterogeneous network is optimized in the low-dimensional space.

[0065] Based on the content of the above system embodiments, as a preferred embodiment, in the enterprise technology partner recommendation system based on the temporal heterogeneous network provided in the embodiments of the present invention, the prediction module is further configured to execute the following instructions: In order to extract specific semantic and temporal information in the temporal heterogeneous network, a random walk method is adopted to generate meta-path instances following four different meta-paths.

[0066] To explain the influence of historical events on the formation of edges between two nodes in the network, time-constrained paths are selected as distance-constrained edges based on the Hawkes process; Based on the extracted distance-constrained edges to analyze this influence, the influence of adjacent nodes is refined into three aspects: semantics, time, and structure. An attention mechanism is assigned to each aspect to assist in calculating node similarity; The Euclidean distance is used to calculate the node embedding similarity, and then the influence of neighbor nodes is calculated based on the semantic attention mechanism, the time attention mechanism, and the structure attention mechanism; A loss function is designed to optimize the algorithm, considering positive and negative samples in the form of negative sampling; After calculating the loss function, adaptive moment estimation is combined with gradient descent to optimize the model.

[0067] Based on the content of the above system embodiments, as a preferred embodiment, in the enterprise technology partner recommendation system based on the temporal heterogeneous network provided in the embodiments of the present invention, the recommendation module is further configured to execute the following instructions: After iterating the parameters and node vectors, calculate the negative Euclidean distance between organizational nodes to represent the probability of their possible cooperation, and sort the similarity between the target organization and other organizations from large to small to generate a technical partner recommendation list for the target organization.

[0068] The method of the embodiments of the present invention is implemented relying on an electronic device. Therefore, it is necessary to introduce the relevant electronic device. For this purpose, an embodiment of the present invention provides an electronic device, such as Figure 5As shown, the electronic device includes: at least one processor, a communications interface, at least one memory, and a communication bus. Among them, the at least one processor, the communications interface, and the at least one memory communicate with each other through the communication bus. The at least one processor invokes the logical instructions in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0069] In addition, when the logical instructions in the foregoing at least one memory are implemented in the form of software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention essentially, or the part that contributes to the prior art, or part of the technical solution is embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (a personal computer, a server, or a network device) to execute all or part of the steps of the methods described in the method embodiments of the present invention. The foregoing storage medium includes: various media for storing program codes such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs.

[0070] Based on the same inventive concept as the foregoing embodiments, the embodiments of the present invention further provide a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method for recommending enterprise technology partners based on a temporal heterogeneous network as follows: Obtain institutional data and patent data, and construct a temporal heterogeneous network that integrates the semantic and structural relationships among institutions, patents, and technical fields; Randomly walk in the temporal heterogeneous network to generate meta-path instances; Use the Hawkes process theory to screen time-constrained paths, and calculate the path influence between node pairs according to the time, semantic, and structural associations of the time-constrained paths; The path influence is weighted by combining embedding similarity, structural attention mechanism, and time attention mechanism, and the semantic attention mechanism further adjusts the contributions of different edge types; Optimize the attention mechanism and node embedding through positive and negative sample loss functions to generate a recommended list of potential technology partners for target enterprises.

[0071] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0072] Persons skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0073] These computer-usable program codes can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 specified in one block or multiple blocks.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to produce a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 specified in one block or multiple blocks.

[0075] In summary of the above embodiments, the present invention integrates multiple technologies. According to the Hawkes process theory, a method is proposed that can effectively capture the continuous evolution among entities in an enterprise technology cooperation network, and can avoid information loss caused by slice discretization while considering the influence of adjacent events. The present invention innovatively proposes a deep semantic extraction method applicable to the temporal heterogeneous network of enterprise technology cooperation, enhances the expression ability of the network, more precisely mines potential cooperation opportunities between enterprises, improves the efficiency of enterprises in finding partners, promotes cooperation and communication between enterprises, and drives innovation and development in the industry.

[0076] Based on the Hawkes process theory, the present invention proposes a method that can effectively capture the continuous evolution among entities in an enterprise technology cooperation network. It can avoid information loss caused by slice discretization while considering the influence of adjacent event occurrences.

[0077] The present invention innovatively proposes a deep semantic extraction method applicable to the temporal heterogeneous network of enterprise technology cooperation. Based on four types of technical cooperation association meta-paths and combined with the random walk method, it mines various potential associations between enterprise technology cooperations.

[0078] When embedding nodes in the temporal heterogeneous network, the present invention designs attention mechanism optimization models for the proposed time-constrained paths and distance-constrained edges from three perspectives: structure, semantics, and time sequence, respectively, to improve the recommendation accuracy of the model.

[0079] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. An enterprise technology partner recommendation method based on a temporal heterogeneous network, characterized in that Including: Obtain institutional data and patent data, and construct a temporal heterogeneous network that integrates the semantic and structural relationships among institutions, patents, and technical fields; Generate meta-path instances by random walk from the temporal heterogeneous network; Screen time-constrained paths with the help of the Hawkes process theory, and calculate the path influence between node pairs according to the time, semantic, and structural associations of the time-constrained paths; The path influence is weighted by combining embedding similarity, structural attention mechanism, and time attention mechanism, and the semantic attention mechanism further adjusts the contributions of different edge types; Optimize the attention mechanism and node embeddings through positive and negative sample loss functions to generate a recommended list of potential technical partners for the target enterprise.

2. The enterprise technology partner recommendation method based on a temporal heterogeneous network according to claim 1, wherein The construction of the temporal heterogeneous network further includes: Expand the relevant attributes of institutions and patents, and construct nodes in the temporal heterogeneous network based on the expanded data.

3. The enterprise technology partner recommendation method based on a temporal heterogeneous network according to claim 1, wherein The generation of the meta-path instances includes: Design four meta-paths according to the essential motivation of institutional technical cooperation; Use the random walk method to extract meta-path instances respectively following the four meta-paths.

4. The enterprise technology partner recommendation method based on a temporal heterogeneous network according to claim 3, wherein The method further includes: Based on the analysis of the extracted distance-constrained edges, refine the influence of neighbor nodes into three aspects: semantics, time, and structure, and assign attention mechanisms to them respectively to assist in the calculation of node similarity.

5. The enterprise technology partner recommendation method based on a temporal heterogeneous network according to claim 1, wherein The method further includes: Use the Euclidean distance to measure the attribute similarity between nodes, and process it through the exponential softmax function to become the weights of each meta-path.

6. The enterprise technology partner recommendation method based on a temporal heterogeneous network according to claim 1, wherein The method further includes: Adopt the form of negative sampling to consider positive and negative samples in designing the loss function, combine the adaptive moment estimation in the gradient descent method to optimize the model, and analyze the recommendation quality of the enterprise technical partners of the overall model.

7. An enterprise technology partner recommendation system based on a temporal heterogeneous network, characterized in that Including: A data collection module, used to obtain institutional data and patent data required for constructing the temporal heterogeneous network, and provide rich and detailed representations for entities in the network through careful node attribute extraction and transformation; A network construction module, used to describe the dynamic technical cooperation relationships between organizations, integrate the business information of institutions, and improve the accuracy of partner recommendation within a fine-grained time range; A prediction module, used to capture the continuous evolution among entities in the enterprise technical cooperation network by the Hawkes method, mine various potential associations between enterprise technical cooperations through meta-path and random walk methods, and optimize model parameters through various attention mechanisms; A recommendation module, used to optimize the attention mechanism and node embeddings through positive and negative sample loss functions to generate potential technical partners for the target enterprise.

8. An enterprise technology partner recommendation device based on a temporal heterogeneous network, characterized in that, Including: At least one processor, at least one memory, and a communication interface; wherein, the processor, memory, and communication interface communicate with each other; The memory stores program instructions executed by the processor, and the processor calls the program instructions to execute the method for recommending enterprise technical partners based on a temporal heterogeneous network according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the method for recommending enterprise technical partners based on a temporal heterogeneous network according to any one of claims 1 to 6.

Images