Enterprise technology partner recommendation method and system based on multilayer heterogeneous network

Abstract

The invention discloses an enterprise technology partner recommendation method based on a multi-layer heterogeneous network, and the method comprises the steps: obtaining enterprise industry and commerce information and patent information, constructing a target layer heterogeneous network containing enterprise nodes and patent nodes, and an auxiliary layer knowledge network, the auxiliary layer knowledge network is used for describing knowledge association between patents; designing a plurality of meta-paths based on the target layer heterogeneous network, and calculating the connection probability of enterprise nodes through different meta-paths; calculating a knowledge association probability between enterprise nodes based on connection edge weights of patent nodes in the auxiliary layer knowledge network; based on the connection probability and the knowledge association probability, generating an inter-enterprise cooperation probability through weighted fusion; and generating a potential technical partner recommendation list of the target enterprise according to the cooperation probability. According to the method, accurate and personalized recommendation is realized by integrating enterprise industry and commerce information, patent knowledge association and time sequence dynamic characteristics.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of matching recommendation of potential technology partners of enterprises, and particularly relates to a target enterprise potential technology partner recommendation method and system based on a multi-layer heterogeneous network. BACKGROUND

[0002] Under the background of global technology innovation complication and intensified industrial competition, enterprise technology partner recommendation has become a key means to break through the "technology shackles" and improve innovation efficiency. Existing technologies mainly focus on enterprise cooperation networks. Early methods are based on single-layer homogeneous networks and only focus on direct cooperation relationships between enterprises, such as predicting cooperation possibilities through similarity indicators such as common neighbors (CN) and Jaccard coefficients. With the development of heterogeneous network theory, some technologies construct heterogeneous networks containing multiple types of nodes such as enterprises, patents, and technology fields, and mine potential cooperation relationships by designing meta-paths (such as "enterprise-patent-enterprise").

[0003] The literature "Collaborator recommendation in heterogeneous bibliographic networks using random walks" retrieves partners in a weighted heterogeneous network through a random walk algorithm, improving recommendation diversity. The study constructs a heterogeneous network containing authors, papers, terms, and conferences, and uses the restart random walk (RWR) algorithm to calculate the correlation between nodes. On the DBLP and HEP-TH datasets, it achieved an average accuracy rate improvement of 15%. However, this method only considers single-layer network structure and does not distinguish the hierarchical differences between enterprise cooperation relationships and patent knowledge associations, resulting in insufficient identification of technology complementarity and cross-domain cooperation opportunities.

[0004] The literature "Identifying supply chain R&D partners via multilayer institutional cooperation network" constructs a three-layer network containing enterprises, institutions, and technology fields, and recommends upstream and downstream partners on the supply chain through node embedding and link prediction algorithms. The study verified the effectiveness of the multi-layer network in the case of an automobile parts supply chain, with a 22% improvement in recommendation accuracy compared to single-layer networks. However, the model still has the following limitations: first, the network structure is simplified to a one-way association between enterprises, institutions, and technology fields, and does not capture the intermediate bridge role of patents as knowledge carriers; second, the weight fusion mechanism uses fixed parameters and cannot adaptively adjust according to enterprise dynamic needs, resulting in insufficient flexibility of the recommendation strategy.

[0005] The technology A Novel Approach to Enterprise Technical Collaboration proposes a fusion recommendation model based on patent classification (IPC) and enterprise business information. It calculates the overlap of IPC and the matching degree of enterprise region and scale to generate cooperation probability. The study found that the recommendation accuracy rate of IPC classification alone is 68%, while the fusion of business information improves to 75% in an experiment of 200 manufacturing enterprises. However, this method still has significant defects: the technology matching degree calculation is based only on the static overlap of IPC classification, without considering the complementary nature of patents, leading to a bias towards technical homogeneity in recommended enterprises. In addition, the model lacks an effective identification mechanism for enterprises with less cooperation, such as start-ups with short establishment time and few patents, and the experimental results show that the recall rate of new cooperation relationships is less than 30%.

[0006] Patent application CN 118797164 A, published on October 18, 2024, proposes a knowledge graph-based technology cooperation recommendation method. This method first collects data, structures the technology intelligence through named entity recognition and relationship extraction, and constructs a technology intelligence knowledge graph. Then, it performs the same structural processing on the technical demand documents provided by the demand side. Next, it uses the DSSM deep semantic similarity model to match the demand features with the semantic features in the knowledge graph, obtaining a similarity score. According to the score, if it is greater than or equal to the threshold, the next demand is matched, otherwise the data is re-collected and modeled. Then, the path embedding recommendation algorithm based on the knowledge graph and the PGPR model are used to mine the optimal path, combined with historical cooperation team data, to generate recommendations and sort them through matrix decomposition. Finally, the opinions of the demand side personnel are collected, and the historical feedback data is used to optimize the recommendations. This method enables enterprises to quickly and accurately find cooperation teams in universities, saving time, but it may have limitations in handling dynamic changes and time dimension information, and its reliance on feedback data may limit the initial recommendation effectiveness.

[0007] In summary, the existing technology mainly has three defects: first, the data source is single, relying on single-dimensional data such as IPC classification, ignoring the deep association between enterprise business information and patents, making it difficult to tap the potential of low-cooperation enterprises; second, the recommendation object is limited, mostly based on historical cooperation, with insufficient identification of new cooperation relationships, and simple network structure, ignoring cross-layer technology association; third, the weight mechanism is static and lacks explanation, fixed weights cannot adapt to dynamic demands, and black-box calculations are difficult to explain the recommendation logic. SUMMARY

[0008] In order to overcome the above-mentioned deficiencies of the prior art, the present application provides a kind of enterprise technology partner recommendation method based on multi-layer heterogeneous network, by integrating enterprise business information, patent knowledge association and time sequence dynamic characteristics, realize precision, personalized recommendation.

[0009] According to an aspect of the present application, a kind of enterprise technology partner recommendation method based on multi-layer heterogeneous network is provided, comprising: obtain enterprise business information, patent information, build the target layer heterogeneous network and auxiliary layer knowledge network containing enterprise node, patent node, the target layer heterogeneous network is used to describe the cooperation between enterprises, the auxiliary layer knowledge network is used to describe the knowledge association between patent; design multiple meta-paths based on target layer heterogeneous network, calculate the connection probability between enterprise nodes through different meta-paths; based on the edge weight of patent node in auxiliary layer knowledge network, calculate the knowledge association probability between enterprise nodes; based on the connection probability and knowledge association probability, generate the cooperation probability between enterprises by weighted fusion; generate the potential technology cooperation partner recommendation list of target enterprise according to cooperation probability.

[0010] As a further technical solution, the construction of the target layer heterogeneous network includes: with enterprise as node, with historical cooperation relationship between enterprises as edge, build enterprise cooperation network; expand enterprise node attribute, including enterprise size, technical field preference and regional information; generate target layer heterogeneous network containing enterprise node and attribute.

[0011] As a further technical solution, the construction of the auxiliary layer knowledge network includes: extract patent abstract text and IPC classification number, and the potential knowledge association between patents is mined by natural language processing technology; with patent as node, with the technical similarity between patents, complementarity as edge weight, build auxiliary layer knowledge network.

[0012] As a further technical solution, the meta-path designed includes at least one of the following: enterprise-patent-enterprise, for capturing the direct cooperation relationship formed by enterprises through common patent; enterprise-patent-technical field-patent-enterprise, for capturing the indirect cooperation relationship formed by enterprises through shared technical field; enterprise-patent-patent-enterprise, for capturing the technical association formed by enterprises through patent citation relationship.

[0013] As a further technical solution, when generating the probability of cooperation between enterprises through weighted fusion, it also includes: The weight parameters θ in the weighted fusion are determined by an adaptive optimization algorithm, which includes gradient descent or particle swarm optimization, with the goal of maximizing the recommendation accuracy index.

[0014] As a further technical solution, the enterprise business information includes the country of origin of the enterprise, the BvD classification of the industry to which the enterprise belongs, and the size of the enterprise; the patent information includes the patent abstract, IPC classification number, and applicant.

[0015] According to one aspect of the present invention, an enterprise technology partner recommendation system based on a multi-layer heterogeneous network is provided, comprising: The data collection module is used to acquire enterprise business registration information and patent information, and to extract enterprise node attributes and patent node attributes; A network construction module is used to construct a target-layer heterogeneous network and an auxiliary-layer knowledge network. The target-layer heterogeneous network integrates enterprise cooperation relationships and business registration information, while the auxiliary-layer knowledge network integrates knowledge associations between patents. The prediction module is used to calculate the connection probability between enterprises based on the target layer meta-path, calculate the knowledge association probability based on the auxiliary layer knowledge network, and generate the cooperation probability through weighted fusion. The recommendation module is used to generate a list of potential technology partners based on the probability of cooperation.

[0016] As a further technical solution, the data collection module processes the patent abstract using natural language processing technology, including using the BERT model to extract the technical field features of the patent and using Word2Vec to generate a vector representation of the enterprise's geographical information.

[0017] According to one aspect of the present invention, an enterprise technology partner recommendation device based on a multi-layer heterogeneous network is provided, comprising: at least one processor; at least one memory storing program instructions executable by the processor; the processor invoking the program instructions to execute the enterprise technology partner recommendation method based on a multi-layer heterogeneous network.

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

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a two-layer architecture comprising a heterogeneous target layer network and an auxiliary layer knowledge network, integrating enterprise business information, patent technology attributes, and time-series dynamic features, thus overcoming the limitations of existing technologies based on a single data source. By characterizing direct cooperative relationships between enterprises at the target layer and mining the technological similarities and complementarities between patents at the auxiliary layer, it achieves deep integration of multi-dimensional information, enabling a more comprehensive representation of enterprise technology needs and cooperation potential, and significantly improving the ability to identify potential partners from enterprises with low cooperation frequency.

[0020] 2. This invention designs a multi-path mining mechanism, extracting meta-path instances such as enterprise-patent-enterprise and enterprise-patent-technology field-patent-enterprise through a random walk algorithm, effectively capturing multiple types of associations such as direct cooperation, co-occurrence in technology fields, and patent citations. Compared to the traditional single-layer network's "1→n" recommendation model that relies solely on historical cooperation data, this invention can discover new "0→1" cooperation relationships across fields and organizational types, expanding the recommendation scope and improving the efficiency of mining innovative cooperation opportunities.

[0021] 3. By introducing an adaptive weight fusion formula and combining it with optimization algorithms such as gradient descent to dynamically adjust weight parameters, this invention solves the problem of insufficient adaptability of static weights in existing technologies. This mechanism can adjust the influence ratio of the target layer and the auxiliary layer in real time according to the dynamic needs of enterprises, significantly improving the flexibility and accuracy of recommendation strategies in scenarios of rapid technological iteration or cooperation in emerging fields.

[0022] 4. At the implementation level of the recommender system, this invention constructs a complete closed loop from data processing to result generation through multi-dimensional attribute extraction in the data collection module, weighted calculation using the attention mechanism in the prediction module, and interpretable output in the recommender module. In particular, through structured meta-path analysis and visualization of weight contributions, it provides clear technical matching criteria for the recommendation results (such as regional collaboration, technological complementarity, and organizational type suitability), solving the interpretability problem of traditional black-box models and enhancing the credibility of enterprise decision-making.

[0023] In summary, this invention effectively overcomes the shortcomings of existing technologies in terms of data fusion, recommendation scope, and adaptability through multi-layer network architecture innovation, multi-source association mining, and dynamic optimization mechanisms. It achieves an improvement in enterprise technology partner recommendation from "limited matching based on historical data" to "intelligent discovery oriented towards potential needs," providing an efficient solution for promoting inter-enterprise technology collaboration and driving industrial innovation. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic diagram of data processing for the enterprise technology partner recommendation method based on multilayer heterogeneous networks provided in this embodiment of the invention; Figure 2 This is a schematic diagram of a multi-layered heterogeneous network for enterprise technology cooperation provided in an embodiment of the present invention. Figure 3 A schematic diagram of an enterprise technology partner recommendation system based on a multi-layer heterogeneous network provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0026] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be decomposed, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. 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 scope of protection claimed by the present invention.

[0029] This invention discloses a method for recommending enterprise technology partners based on multi-layer heterogeneous networks, used to achieve accurate matching and recommendation of potential technology partners for target enterprises. The method specifically includes: S1: Data Collection and Node Attribute Extraction. Through multi-source data acquisition and attribute expansion, this provides fundamental data support for constructing temporal heterogeneous networks, specifically including two parts: data collection and attribute processing.

[0030] Acquire enterprise and patent data, expand enterprise business attributes, including enterprise size, region, organization type, business scope and patent technology attributes, including IPC classification number, abstract text and citation relationship, and construct a target layer heterogeneous network and an auxiliary layer knowledge network containing enterprise nodes and patent nodes.

[0031] S2: Construction of a Multi-Layer Heterogeneous Network. Based on the expanded node attributes, a two-layer architecture is constructed, comprising a target-layer heterogeneous network and an auxiliary-layer knowledge network. The construction of the multi-layer heterogeneous network includes: generating regional and business scope vectors from the organization's business attributes using Word2Vec; extracting technical feature vectors from the patent text using the BERT model; storing node attributes and edge relationships through a graph database; and supporting dynamic updates to the network structure and time-series data.

[0032] S3: Meta-path Instance Generation and Path Influence Calculation. Based on the motivations for enterprise technology cooperation, four core meta-paths are designed, and the network is traversed using a random walk algorithm. The meta-path design covers: direct cooperation paths, i.e., enterprise-patent-enterprise, capturing direct collaborative relationships formed by enterprises through joint patent applications; technology field association paths, i.e., enterprise-patent-technology field-patent-enterprise, exploring the possibility of indirect cooperation formed by enterprises through shared technology fields; patent citation paths, i.e., enterprise-patent-patent-enterprise, identifying technology dependencies formed by patent technology citations; and organization type association paths, i.e., enterprise-patent-applicant type-patent-enterprise, reflecting enterprises' cooperation preferences for specific organization types (such as universities and research institutions).

[0033] S4: Partner Recommendation and Evaluation. Based on historical cooperation frequency and business attribute similarity, the connection probability between enterprise nodes on the target layer heterogeneous network is calculated. Based on patent text semantic matching, IPC classification number cross-degree, and technical phrase co-occurrence frequency, the knowledge association probability between enterprise nodes on the auxiliary layer knowledge network is calculated. The target layer connection probability and the auxiliary layer knowledge association probability are weighted and fused to obtain the cooperation probability between enterprises. The weight parameter θ in the weighted fusion formula is determined by an adaptive optimization algorithm, including gradient descent or particle swarm optimization. Using historical cooperation data as the training set, the weight parameter is iteratively optimized by minimizing the cross-entropy loss function to achieve adaptive matching to the dynamic needs of enterprises.

[0034] As a preferred embodiment, such as Figure 1 As shown, the present invention describes four stages: data collection and node attribute extraction, multi-layer heterogeneous network construction, meta-path instance generation, and potential enterprise partner recommendation.

[0035] In this invention, the data collection and node attribute extraction process includes collecting enterprise data and patent data, performing preliminary data processing, and then extracting node attributes. Specifically, a multi-source data acquisition platform is used to obtain basic data such as enterprise size, enterprise type, business scope, patent abstracts, and IPC classification numbers from heterogeneous data sources such as enterprise business databases and patent databases. After data cleaning and deduplication preprocessing, natural language processing technology is used to extract features from text attributes, and numerical and category attributes are standardized and encoded to form structured node attribute data. This provides standardized and feature-rich basic data support for constructing a temporal heterogeneous network, ensuring the accuracy and completeness of subsequent network construction.

[0036] In this embodiment of the invention, the data collection step involves collecting and analyzing patent data within a specific industry. Patent searches are conducted according to the emerging strategic industries classification, specifically the new energy vehicle industry. Data is sourced from the IncoPat database, and patent types are limited to "application" and "patent." Legal status is limited to "grant" and "publication," with the grantor being the United States. Patent data includes publication number, application number, title, abstract, current assignee, application date, IPC classification number, and patents cited within the last three years. The geographical location and business scope of the patent assignee are obtained from the Orbis database.

[0037] In the node attribute extraction step of this invention, various natural language processing techniques are employed. For enterprise information attributes, business registration information is collected, including the province, location, business scope, industry classification, and number of employees. For the geographical location of the enterprise, Word2Vec is directly used to convert it into a vector. For other text-based business registration information, keywords are first extracted from the text using TextRank, and then Word2Vec is used to generate vectors. For patent attributes, the abstract text of each patent is collected, and BERT is used to classify the patent into a coarse-grained technical field. Finally, the classification result of each patent is represented as a vector. Through this extraction and transformation of node attributes, rich and detailed representations of entities in the network are provided, laying a solid foundation for subsequent stages of the model.

[0038] For enterprise size, the number of employees is used as a numerical indicator and directly as a characteristic value.

[0039] For different business types, such as manufacturing and service industries, one-hot encoding can be used to convert them into vectors. Assuming there are n business types, the business type feature vector is an n-dimensional vector, with the position corresponding to the type set to 1 and the remaining positions set to 0.

[0040] The patent abstract text is encoded using a pre-trained BERT model and converted into a fixed-length vector h. p abstract The BERT model encodes semantic information from text into a vector space.

[0041] One-hot encoding of the IPC classification number yields a vector h. p IPC Assume there are a total of m different IPC classification numbers, which, after encoding, result in an m-dimensional vector; The final patent feature vector h p h can be obtained by concatenation or weighted summation. p abstract and h pIPC The combination yields: h p =[h p abstract ;h p IPC ].

[0042] In this embodiment of the invention, during the construction of the multi-layer heterogeneous network, a two-layer architecture network is constructed using graph data modeling technology based on the extracted node attributes. The target layer heterogeneous network integrates multiple types of nodes, such as enterprises, patents, and enterprise attributes, and defines multiple types of edges, such as enterprise-patent and enterprise-enterprise, to depict the direct association and attribute mapping relationship of enterprise cooperation. The auxiliary layer knowledge network uses patent knowledge elements as nodes and constructs edges through semantic similarity and co-occurrence relationships between knowledge elements to form a knowledge association network. The two layers of networks are both independent and interconnected, providing a structural foundation for multi-dimensional relationship modeling.

[0043] In this embodiment of the invention, when constructing the two-layer heterogeneous network, this step builds patent-patent co-occurrence edges based on patent knowledge elements. For example, the semantic similarity of patent abstracts is calculated using cosine similarity; if the similarity exceeds a certain threshold θ... sim Then, an edge is added between the two patent nodes. The auxiliary layer network is composed of graph G. auxiliary =(V patent E auxiliary ) indicates that V patent It is a set of patent nodes, E auxiliary It is a set of patent-patent co-occurrence edges. Different types of edges are clearly defined, such as enterprise-patent and enterprise-enterprise attribute edges. An enterprise-patent edge indicates that an enterprise owns or participates in a patent; an enterprise-enterprise attribute edge can represent the similarity between enterprises in attributes such as size and type. The target layer network can be represented as a heterogeneous graph G. target =(V enterprise ∪V patent ∪V attribute E target ), where V enterprise It is a collection of enterprise nodes, V attribute It is a collection of enterprise attribute nodes, E target It is a set containing edges of different types.

[0044] For patent node v in the auxiliary layer patent co-occurrence network p Its neighbor node set is .

[0045] h p It is a patent node v p The feature vectors are denoted by W, which is a shared weight matrix used to perform a linear transformation on the node features. This transformation is achieved through Wh... p Map node features to a new feature space.

[0046] 'a' is an attention vector used to calculate the attention weights between nodes.

[0047] It is node v p Its neighbor node v p′ The attention weight is calculated using the following formula:

[0048] Among them, [Wh p ||Wh p′ ] indicates that Wh p and Wh p′ Perform the splicing operation. LeakyReLU is an activation function with the formula LeakyReLU(x)={x,x≥0;αx,x<0}, where α is a small positive number (usually 0.01) used to address the "neuron death" problem of the ReLU function; The patent node v is updated by weighting and summing the features of neighboring nodes using attention weights. p The representation of h p′ :

[0049] Where σ is the activation function, here the Sigmoid function; For the enterprise node v in the target layer heterogeneous network c Its patent neighbor node set is N p (v c The set of attribute neighbor nodes is N. a (v c ); E c-p E represents the set of types of firm-patent edges. c-a Represents the set of types for enterprise-attribute edges; For each edge type r, there is a corresponding transformation matrix Wr, which is used to perform different linear transformations on the node features connected by different edge types; It is an enterprise node v c Its patent neighbor node v p Attention weights under edge type r It is an enterprise node v c Its attribute neighbor node v a Attention weights under edge type r; Enterprise node v c The update indicates The calculation formula is:

[0050] Where h a It is an attribute node v a eigenvectors, These represent the calculation results under edge type r.

[0051] In this embodiment of the invention, the meta-path instance generation and path influence calculation stage includes at least the following meta-paths: Enterprise-Patent-Enterprise (EPE), used to capture direct cooperative relationships formed by enterprises through shared patents; Enterprise-Patent-Technology Field-Patent-Enterprise (EPTPE), used to capture indirect cooperative relationships formed by enterprises through shared technology fields; and Enterprise-Patent-Patent-Enterprise (EPPPE), used to capture technological associations formed by enterprises through patent citation relationships.

[0052] For each metapath P, let its length be |P|, and at the t-th position on the metapath, the enterprise node v c and The representations are hc (t) and .

[0053] It is a node pair (v c ,v c′ The semantic similarity under metapath P is calculated using the following formula:

[0054] Among them, cosine (hc) (t) , ) is hc (t) and The cosine similarity is calculated using the following formula: .

[0055] Let P set It is the set of all defined metapaths, β P The weight of the metapath P is determined by a weighted average. Node pairs final semantic similarity for:

[0056] In this embodiment of the invention, the partner recommendation and evaluation process first predicts the possible links of the nodes in the manner described above, then selects the top k results as the recommendation list, and selects indicators to evaluate the model's prediction effect.

[0057] For node pair (v c ,v p ), embed it into vector h c′′ and h p By concatenating the '' vectors, we obtain a new vector x=[h] c ′′;h p ′]; x is input into a multilayer perceptron (MLP), which consists of multiple fully connected layers, each of which can be represented as y. l+1 =σ(W l y l +b l ), where y l W is the input vector of the l-th layer. l b is the weight matrix of the l-th layer. l σ is the bias vector of the l-th layer, and σ is the activation function; The output of the last layer is processed by the Sigmoid function to obtain the probability P(v) of the existence of the link. c ,v p ); Using the binary cross-entropy loss function, for all node pairs (v i ,v j For any ∈ V, the formula for calculating the loss function L is:

[0058] Where y i,j It is a node pair (v i ,v j The true label of y, if a link exists. i,j =1, otherwise yi,j =0; Use the Adam optimizer to update the model's parameters (including W, a, W0). r W l b l (etc.). The Adam optimizer combines the advantages of Adagrad and RMSProp, adaptively adjusting the learning rate for each parameter. In each iteration, the parameters are updated based on the gradient of the loss function to minimize the loss function; In this embodiment of the invention, to verify the recommendation performance of the proposed model, three metrics—Precision@k, Recall@k, and nDCG@k—are used to evaluate the FTP recommendation results. Precision@k evaluates the accuracy of the recommendation results, reflecting the proportion of correctly predicted results among the top k recommendations. Recall@k focuses on the coverage of the top k recommendations with respect to actual enterprise technology partners. nDCG@k focuses on the rationality of the recommendation ranking, whether the recommended results take precedence over the actual FTP. These metrics are calculated as follows:

[0059] The implementation of the various embodiments of the present invention is based on programmed processing through a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide an enterprise technology partner recommendation system based on a multi-layer heterogeneous network. This system is used to execute the enterprise technology partner recommendation method based on a multi-layer heterogeneous network in the above method embodiments.

[0060] See Figure 3 The system includes: a data collection and node attribute extraction module, which transforms enterprise business information and patent data into structured node attributes through multi-source acquisition and processing; a multi-layer heterogeneous network construction module, forming a two-layer architecture of a target-layer heterogeneous network and an auxiliary-layer knowledge network; a meta-path instance generation and influence calculation module, which designs core meta-paths and uses random walks to traverse the network to determine the weight of each path on the node relationship; and an enterprise partner recommendation module, which calculates connection probabilities at the target layer based on historical cooperation frequency and business attribute similarity, and calculates knowledge association probabilities at the auxiliary layer through patent text semantic matching and IPC classification number cross-degree, and weights and fuses the two to obtain the enterprise cooperation probability, completing partner recommendation and evaluation, and achieving accurate modeling of the entire process from data processing to cooperation prediction.

[0061] The enterprise technology partner recommendation system based on multi-layer heterogeneous networks provided in this invention adopts... Figure 4 This paper proposes a novel multi-layer heterogeneous network node embedding algorithm by integrating several modules and multiple technologies. The model incorporates enterprise business information, improving the accuracy of predicting enterprise cooperation when existing cooperation is insufficient, and contributing to the improvement of the industrial chain in emerging industries.

[0062] It should be noted that the system embodiments provided by this invention, in addition to implementing the methods in the above method embodiments, are also used to implement the methods in other method embodiments provided by this invention. The difference lies only in setting corresponding functional modules, and their principles are basically the same as those of the above system embodiments provided by this invention. As long as those skilled in the art, based on the above system embodiments and referring to the specific technical solutions in other method embodiments, obtain corresponding technical means and technical solutions composed of these technical means by combining technical features, and improve the modules in the above system embodiments while ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments. For example: Based on the above system embodiments, as a preferred embodiment, the enterprise technology partner recommendation system based on multi-layer heterogeneous networks provided in this invention further includes a data collection module configured to execute the following instructions: By utilizing professional data acquisition tools and interfaces, we can obtain raw data such as enterprise size, type, business scope, patent abstracts, and IPC classification numbers from multiple sources, including enterprise business databases and patent databases. The collected data is cleaned to remove duplicate, erroneous, and incomplete data, and data correction is performed to ensure the accuracy and integrity of the data, laying the foundation for subsequent processing. By using natural language processing technologies, such as word segmentation, part-of-speech tagging, and named entity recognition, key features are extracted from textual information such as patent abstracts and business scopes, and the text is transformed into a computable feature vector. For numerical and categorical attributes such as enterprise size and type, a standardization method is used to scale the numerical attributes to a uniform range, and one-hot encoding is used to convert the categorical attributes into binary vectors for easier model processing.

[0063] As a preferred embodiment, the enterprise technology partner recommendation system based on a multi-layer heterogeneous network provided in this embodiment of the invention further includes a multi-layer heterogeneous network construction module configured to execute the following instructions: Based on the expanded node attributes, a heterogeneous network for the target layer is constructed, which includes nodes of different types such as enterprises, patents, and attributes, comprehensively covering all kinds of information related to enterprise cooperation.

[0064] Define the different types of edges in the target layer, such as enterprise-patent and enterprise-attribute, and use these edges to characterize the direct relationships between enterprises, such as the patents owned by the enterprise and the attribute characteristics of the enterprise; Using patent knowledge elements, such as semantic fragments and technical phrases, as nodes, an auxiliary layer knowledge network is constructed to deeply explore the knowledge connections behind the patents; By calculating semantic similarity and statistical co-occurrence relationships, the edges between auxiliary layer nodes are determined, forming a network structure that reflects the inherent connections of patent knowledge.

[0065] As a preferred embodiment, the enterprise technology partner recommendation system based on multi-layer heterogeneous networks provided in this embodiment of the invention further includes a meta-path instance generation and path influence calculation module used to execute the following instructions: By deeply analyzing the motivations for enterprise technology cooperation and combining industry characteristics and business needs, four representative core meta-paths, such as "enterprise-patentee-enterprise", are designed to capture potential cooperative relationships between enterprises. Using a random walk algorithm, a large number of meta-path instances are generated by traversing a two-layer network according to a preset length and rules to simulate the interaction process of enterprises in the network. The frequency of occurrence of meta-paths is statistically analyzed to quantify the strength of associations between nodes, and the contribution of each meta-path to the node relationships is evaluated by analyzing this data. Taking into account indicators such as the frequency and correlation strength of meta-paths, a mathematical model is used to calculate the influence weight of each meta-path in characterizing node relationships, providing a basis for subsequent probability calculations.

[0066] As a preferred embodiment, the enterprise technology partner recommendation system based on a multi-layer heterogeneous network provided in this embodiment of the invention further includes a partner recommendation and evaluation module configured to execute the following instructions: In the target layer heterogeneous network, considering the historical cooperation frequency and business attribute similarity of enterprises, machine learning algorithms or similarity calculation methods are used to calculate the connection probability between enterprise nodes, reflecting the possibility of direct cooperation between enterprises. In the auxiliary layer knowledge network, a knowledge association model is constructed based on information such as the semantic matching degree of patent texts and the crossover degree of IPC classification numbers. The probability of knowledge association between enterprise nodes is calculated to reflect the degree of fit between enterprises at the level of technical knowledge. The weights of the connection probability of the target layer and the knowledge association probability of the auxiliary layer are determined by using the analytic hierarchy process or machine learning algorithm. The probabilities of the two layers are then weighted and fused to obtain the comprehensive cooperation probability between enterprises. Potential partners are ranked based on the probability of cooperation after integration, and the companies with the greatest cooperation potential are selected to complete accurate partner recommendations and evaluations.

[0067] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, embodiments of the present invention provide an electronic device, such as... Figure 4 As shown, the electronic device includes: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor invokes logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0068] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, is embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0069] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0070] These computer-usable program codes may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] In summary, this invention constructs a two-layer heterogeneous network. The target layer integrates enterprise business information and cooperative relationships, while the auxiliary layer mines patent semantics and classification number associations, achieving multi-dimensional data fusion. It designs multiple paths to capture different associations and employs an adaptive algorithm to dynamically adjust inter-layer weights, enhancing recommendation flexibility. By leveraging the complementary nature of techniques in the auxiliary layer and combining meta-path analysis, it provides interpretable recommendation criteria, filling the gap in identifying new cooperative relationships. These improvements effectively enhance the comprehensiveness, accuracy, and interpretability of the recommendations.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions 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. A method for recommending enterprise technology partners based on multilayer heterogeneous networks, characterized in that, include: Obtain enterprise business registration information and patent information, and construct a target layer heterogeneous network and an auxiliary layer knowledge network containing enterprise nodes and patent nodes. The target layer heterogeneous network is used to describe the cooperative relationship between enterprises, and the auxiliary layer knowledge network is used to describe the knowledge association between patents. Based on the design of a heterogeneous network at the target layer, multiple meta-paths are used to calculate the connection probability between enterprise nodes through different meta-paths. Based on the edge weights of patent nodes in the auxiliary layer knowledge network, the probability of knowledge association between enterprise nodes is calculated. Based on the connection probability and knowledge association probability, the cooperation probability between enterprises is generated by weighted fusion. Generate a list of potential technology partners for the target company based on the probability of cooperation.

2. The enterprise technology partner recommendation method based on multilayer heterogeneous networks according to claim 1, characterized in that, The construction of the target layer heterogeneous network includes: A corporate cooperation network is constructed, with enterprises as nodes and historical cooperation relationships between enterprises as edges. Expand enterprise node attributes, including enterprise size, technology preferences, and geographic information; Generate a target layer heterogeneous network containing enterprise nodes and attributes.

3. The enterprise technology partner recommendation method based on multilayer heterogeneous networks according to claim 1, characterized in that, The construction of the auxiliary layer knowledge network includes: Extract patent abstract text and IPC classification numbers, and use natural language processing technology to uncover potential knowledge connections between patents; Using patents as nodes and the technological similarity and complementarity between patents as edge weights, an auxiliary layer knowledge network is constructed.

4. The enterprise technology partner recommendation method based on multilayer heterogeneous networks according to claim 1, characterized in that, The meta-path of the design includes at least one of the following: Enterprise-Patent-Enterprise is used to capture direct collaborative relationships between enterprises through shared patents. The "Company-Patent-Technology Field-Patent-Company" structure is used to capture indirect collaborative relationships formed by companies through shared technology fields. The Enterprise-Patent-Patent-Enterprise model is used to capture the technological connections formed by enterprises through patent citation relationships.

5. The enterprise technology partner recommendation method based on multilayer heterogeneous networks according to claim 1, characterized in that, When generating the probability of cooperation between firms through weighted fusion, it also includes: The weight parameters θ in the weighted fusion are determined by an adaptive optimization algorithm, which includes gradient descent or particle swarm optimization, with the goal of maximizing the recommendation accuracy index.

6. The enterprise technology partner recommendation method based on multilayer heterogeneous networks according to claim 1, characterized in that, The enterprise business registration information includes the country of origin of the enterprise, the BvD classification of the industry to which the enterprise belongs, and the size of the enterprise. The patent information includes the patent abstract, IPC classification number, and applicant.

7. An enterprise technology partner recommendation system based on multi-layer heterogeneous networks, characterized in that, include: The data collection module is used to acquire enterprise business registration information and patent information, and to extract enterprise node attributes and patent node attributes; A network construction module is used to construct a target-layer heterogeneous network and an auxiliary-layer knowledge network. The target-layer heterogeneous network integrates enterprise cooperation relationships and business registration information, while the auxiliary-layer knowledge network integrates knowledge associations between patents. The prediction module is used to calculate the connection probability between enterprises based on the target layer meta-path, calculate the knowledge association probability based on the auxiliary layer knowledge network, and generate the cooperation probability through weighted fusion. The recommendation module is used to generate a list of potential technology partners based on the probability of cooperation.

8. The enterprise technology partner recommendation system based on a multi-layer heterogeneous network according to claim 7, characterized in that, The data collection module processes patent abstracts using natural language processing technology, including using the BERT model to extract technical features of the patents and using Word2Vec to generate vector representations of enterprise geographic information.

9. A device for recommending enterprise technology partners based on multilayer heterogeneous networks, characterized in that, include: At least one processor; At least one memory stores program instructions that can be executed by the processor; The processor invokes the program instructions to execute the enterprise technology partner recommendation method based on a multilayer heterogeneous network as described in any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the enterprise technology partner recommendation method based on a multilayer heterogeneous network as described in any one of claims 1 to 6.

Images