Enterprise service accurate docking method and system based on artificial intelligence

Abstract

The invention relates to the technical field of customer interaction service, discloses an enterprise service accurate docking method and system based on artificial intelligence, and realizes accurate matching between enterprises and services by introducing a probability calculation and optimization mechanism of a candidate docking relationship on the basis of a knowledge graph structure. According to the method, the stability and the accuracy of the matching result can be improved by integrating the attribute characteristics, the historical behavior data and the multi-path information under the condition that the multi-source data is incomplete or noise exists, and the recommendation result can adapt to changes of enterprise requirements and service supply in a self-adaptive mode by combining dynamic probability updating and iterative convergence operation. Meanwhile, by introducing a negative sample rejection strategy in the recommendation process, the negative influence of historical low-success-rate docking and abnormal docking conditions on subsequent recommendation is effectively inhibited, so that the practicability and credibility of a recommendation list are improved, and finally more efficient, accurate and sustainable decision support is provided for enterprise service docking.

Description

Technical Field

[0001] This invention relates to the field of customer interaction service technology, and more specifically to a method and system for precise matching of enterprise services based on artificial intelligence. Background Technology

[0002] With the development of industrial digitalization and informatization, the demand for business cooperation between enterprises is becoming increasingly frequent, and online platforms and service matching systems have become important channels for cooperation. While existing enterprise service recommendation and matching systems can provide preliminary matching suggestions based on some attribute information of enterprises and services in practical applications, the overall accuracy and stability of the matching remain insufficient. When multi-source data is incomplete or contains noise, the reliability of the recommendation results decreases significantly; and when faced with changes in enterprise needs or dynamic adjustments in the business environment, the recommendation list lacks adaptability and struggles to reflect the latest cooperation potential in a timely manner. Furthermore, there is a lack of effective means to identify and utilize historical cooperation failures and abnormal matching patterns, leading to the potential recurrence of similar problems in subsequent recommendations, thus affecting the overall recommendation quality and the user experience for enterprises. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide an artificial intelligence-based method and system for precise matching of enterprise services, so as to solve the technical problems of poor overall matching accuracy, stability and matching quality in the current enterprise service matching process.

[0004] The first aspect of this invention discloses a method for precise matching of enterprise services based on artificial intelligence, the method comprising: A set of enterprise nodes is constructed based on enterprise attribute information, business operation records, and service call records from the enterprise side. A set of service nodes is constructed based on the service attribute information, service call records, and historical success records of the server. An enterprise service knowledge graph is constructed based on the enterprise node set and the service node set, and candidate docking relationships between enterprises and services are generated based on the node relationships in the knowledge graph. The matching probability of the candidate docking relationships is calculated using an uncertainty propagation algorithm to obtain the matching probability distribution between enterprises and services; A list of recommended enterprise service connections is generated based on the matching probability distribution and the preset connection decision rules.

[0005] Furthermore, the enterprise service knowledge graph also includes attribute nodes and relationship edges; wherein, The enterprise node is used to record the static attribute characteristics and behavioral characteristics of the enterprise; the behavioral characteristics include a sequence of behavioral operations generated based on business operation records and service call records. The service node is used to record the static attribute characteristics and dynamic evaluation characteristics of the service; the dynamic evaluation characteristics include performance evaluation and service call statistics generated based on service call records and historical success records. The attribute nodes are used to record shared characteristic information of enterprises and services.

[0006] Furthermore, the process of determining the weights of the relation edges includes: The dynamic feature matching degree is determined based on the similarity between enterprise behavioral characteristics and service dynamic evaluation characteristics; Static feature matching degree is determined based on the similarity between enterprise static attribute features and service static attribute features; Calculate the potential correlation degree based on the multi-hop path of attribute nodes from enterprise nodes to service nodes; The dynamic feature matching degree, static feature matching degree, and potential correlation degree are fused together according to preset weights to generate relation edge weights.

[0007] Furthermore, the step of generating candidate connection relationships between enterprises and services based on the node relationships in the knowledge graph specifically includes: Calculate the structural similarity based on the multi-hop path between enterprise nodes and service nodes, and select node pairs whose structural similarity meets the first preset threshold as a candidate docking edge set; Semantic path similarity is evaluated for the corresponding attribute node paths in the candidate docking edge set; the semantic path similarity is obtained by jointly calculating the path label matching degree and the path embedding similarity. The semantic path credibility is determined based on the semantic similarity evaluation results and the weight distribution of the attribute node path in the knowledge graph; the weight distribution of the attribute node path is obtained by weighted calculation of the edge weights of each relation edge in the path. The comprehensive matching score for each candidate docking edge is determined based on structural similarity and semantic path credibility.

[0008] Further, the operation of calculating the matching probability of the candidate docking relationship through the uncertainty propagation algorithm is performed in the Bayesian network model; wherein, the Bayesian network model construction process specifically includes: Each candidate docking edge is used as a node to be inferred in the Bayesian network model, and the comprehensive matching score corresponding to the candidate docking edge is used as the initial probability value and input into the Bayesian network model. Based on the knowledge graph structure, a Bayesian network model structure is constructed with candidate docking edges as target nodes and attribute nodes involved in the semantic path of the candidate docking edges as intermediary nodes.

[0009] Furthermore, the process of calculating the matching probability of the candidate docking relationship using the uncertainty propagation algorithm specifically includes: The prior probability of the enterprise is calculated based on the static attribute features and behavioral features of the enterprise nodes corresponding to the candidate docking edges, and the prior probability is corrected according to the connection strength of the enterprise nodes in the knowledge graph; the connection strength of the enterprise nodes is obtained based on the number of edge weights and interaction frequency of the enterprise nodes with other nodes in the knowledge graph. The conditional probability of the server is calculated based on the dynamic evaluation feature information of the service node corresponding to the candidate docking edge. Based on the prior probability of the enterprise side and the conditional probability of the server side, the algorithm iteratively propagates within the Bayesian network model structure, updating the matching probability of each candidate connection edge until convergence.

[0010] Furthermore, the method also includes: Identify abnormal patterns of enterprises and services based on historical connection records; the abnormal patterns include abnormal connection patterns and matching failure patterns. In the process of matching probability calculation, abnormal patterns are introduced as negative samples. A penalty coefficient is introduced for candidate matching relationships that belong to negative samples, and their priority in the enterprise service matching recommendation list is reduced.

[0011] Furthermore, the penalty coefficient is dynamically adjusted based on the frequency of abnormal patterns and the matching failure rate; Before generating the enterprise service matching recommendation list, the final recommendation probability is obtained by multiplying the matching probability by a correction factor calculated based on the penalty coefficient; the final recommendation probability combines the positive matching score and the negative sample rejection effect.

[0012] Furthermore, the process of generating the enterprise service matching recommendation list includes: The candidate enterprise service connections are ranked according to the final recommendation probability; When the recommendation probabilities of multiple candidate docking relationships are similar, the candidate docking relationship that has not triggered the negative sample rejection mechanism shall be selected first. Output a list of recommended enterprise service partners, sorted by priority.

[0013] The second aspect of this invention discloses an artificial intelligence-based enterprise service precision matching system, the system comprising: The first construction module is used to build a set of enterprise nodes based on enterprise attribute information, business operation records and service call records on the enterprise side. The second construction module is used to build a set of service nodes based on the service attribute information of the server, the service call record, and the historical success record. The calculation module is used to calculate the matching probability of the candidate docking relationship through an uncertainty propagation algorithm, and obtain the matching probability distribution between enterprises and services; The list generation module is used to generate a recommended list of enterprise service connections based on the matching probability distribution and preset connection decision rules.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves precise matching between enterprises and services by introducing probabilistic calculation and optimization strategies for candidate matching relationships based on a knowledge graph structure. It not only improves the stability and accuracy of matching results by integrating attribute features, historical behavioral data, and multi-path information, even in the presence of incomplete or noisy multi-source data, but also, through dynamic probability updates and iterative convergence, enables the recommendation results to adapt to changes in enterprise demand and service supply, significantly enhancing responsiveness to dynamic adjustments in the business environment. Furthermore, by introducing a negative sample exclusion strategy during the recommendation process, it effectively suppresses the negative impact of historically low-success-rate matches and abnormal matches on subsequent recommendations, thereby improving the practicality and credibility of the recommendation list and ultimately providing more efficient, accurate, and sustainable decision support for enterprise service matching. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a precise matching method for enterprise services based on artificial intelligence, as disclosed in an embodiment of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example 1 The first aspect of this invention discloses a method for precise matching of enterprise services based on artificial intelligence. This method belongs to the field of customer interaction service technology. Please refer to [link / reference needed]. Figure 1 , Figure 1 This is a flowchart illustrating a method for precise matching of enterprise services based on artificial intelligence, as disclosed in an embodiment of the present invention. The method includes: A set of enterprise nodes is constructed based on enterprise attribute information, business operation records, and service call records from the enterprise side. A set of service nodes is constructed based on the service attribute information, service call records, and historical success records of the server. An enterprise service knowledge graph is constructed based on the enterprise node set and the service node set, and candidate docking relationships between enterprises and services are generated based on the node relationships in the knowledge graph. The matching probability of the candidate docking relationships is calculated using an uncertainty propagation algorithm to obtain the matching probability distribution between enterprises and services; A list of recommended enterprise service connections is generated based on the matching probability distribution and the preset connection decision rules.

[0018] Furthermore, the enterprise service knowledge graph also includes attribute nodes and relationship edges; wherein, The enterprise node is used to record the static attribute characteristics and behavioral characteristics of the enterprise; the behavioral characteristics include a sequence of behavioral operations generated based on business operation records and service call records. The service node is used to record the static attribute characteristics and dynamic evaluation characteristics of the service; the dynamic evaluation characteristics include performance evaluation and service call statistics generated based on service call records and historical success records. The attribute nodes are used to record shared characteristic information of enterprises and services.

[0019] Specifically, enterprise attribute information refers to a set of static information that reflects the basic characteristics of an enterprise, including but not limited to its business type, industry category, size, registered location, years of establishment, and main business focus. Enterprise business operation records include sequences of business operations collected from business systems, management systems, or transaction systems. These sequences reflect the enterprise's business behavior patterns and operational frequencies at different times, such as time-series records of key business events like order generation, contract signing, and after-sales processing. By structurally encoding the above static and behavioral features, enterprise nodes are constructed within the enterprise node set. Each node contains its unique identifier, static feature vector, and behavioral feature vector, thus providing foundational data for subsequent graph structure construction and matching calculations.

[0020] Service attribute information refers to the set of static information for each service provided by the server, including but not limited to the service category, functional scope, target user group, industry sector, and service provider qualification level. For service call records and historical success records, the former reflects the number of times the service is called in actual business operations, the distribution of call time, and the source of the call; the latter reflects the proportion of successful task completion in past calls and the corresponding conditions. Based on the above information, by vectorizing the static features of the service and encoding the call records and success records as dynamic evaluation features, each service node in the service node set is constructed. Each node also contains a unique identifier, a static feature vector, and a dynamic evaluation feature vector to describe the service's stability and dynamic performance.

[0021] Furthermore, in this embodiment of the invention, attribute nodes are used to record shared characteristic information between enterprises and services. These are attribute features that exist and are related in both enterprise and service nodes, such as being in the same industry sector, using the same technical standards, having services adapted to the enterprise's main business category, or meeting the same security and compliance requirements. Attribute nodes can include industry tag nodes, technical standard nodes, business process nodes, compliance nodes, etc., and reflect the implicit relationships between enterprises and services by establishing connection edges with enterprise and service nodes. The introduction of such attribute nodes helps to achieve indirect associations using shared features when direct relationships between nodes are insufficient, thereby improving the comprehensiveness and accuracy of subsequent matching calculations.

[0022] Furthermore, the process of determining the weights of relation edges includes: The dynamic feature matching degree is determined based on the similarity between enterprise behavioral characteristics and service dynamic evaluation characteristics; Static feature matching degree is determined based on the similarity between enterprise static attribute features and service static attribute features; Calculate the potential correlation degree based on the multi-hop path of attribute nodes from enterprise nodes to service nodes; The dynamic feature matching degree, static feature matching degree, and potential correlation degree are fused together according to preset weights to generate relation edge weights.

[0023] In this embodiment of the invention, the weight of the relation edge is determined by comprehensively considering dynamic feature matching degree, static feature matching degree, and potential correlation degree. Specifically, enterprise behavioral features reflect the enterprise's business operation sequence, service call habits, and call result characteristics within a certain time window, while service dynamic evaluation features reflect the dynamic performance of the service within the same time window, such as performance fluctuations, call stability, and response speed. The degree of matching between the two is quantified through similarity calculation models (such as cosine similarity, Pearson correlation coefficient, etc.). Secondly, enterprise static attribute features include, but are not limited to, the enterprise's business type, industry category, and scale level, while service static attribute features include, but are not limited to, the functional module to which the service belongs, the scope of applicable industries, and the qualification level of the provider. The similarity between the two can be obtained through label matching, encoding distance, or multi-dimensional attribute similarity calculation methods.

[0024] Furthermore, to measure the potential indirect business relationships between enterprises and services, this invention also calculates the potential correlation degree based on the multi-hop path of attribute nodes from the enterprise node to the service node. Here, the multi-hop path of attribute nodes refers to the path starting from the enterprise node, passing through one or more attribute nodes in sequence, such as shared industry tags, commonly used functional module tags, or historical interaction tags, and finally reaching the target service node. The number of hops in each path and the correlation strength of the nodes on the path together determine the value of the potential correlation degree.

[0025] Dynamic feature matching degree, static feature matching degree, and potential correlation degree are fused according to preset weights. The preset weights are set based on historical matching effect evaluation results, feature importance analysis, and business scenario priority, in order to balance the influence of different features in the calculation of relationship edge weights, thereby generating the final relationship edge weights that reflect the overall matching degree between the enterprise and the service.

[0026] In this invention, dynamic feature matching degree, static feature matching degree, and potential correlation degree play different roles in the calculation of relationship edge weights. Dynamic feature matching degree reflects the degree of fit between a company's recent business behavior and the dynamic performance of its services, used to identify combinations of enterprise services that are more likely to generate stable interactions in actual operation. Static feature matching degree reflects the compatibility between a company's basic attributes and the inherent characteristics of its services, helping to quickly screen objects with long-term cooperation potential in the early stages of business cooperation. Potential correlation degree, through multi-hop path mining of attribute nodes, uncovers indirect relationships and can discover enterprise service pairs that have not had direct interaction but possess potential collaborative value. The combination of these three factors can enhance the ability to discover potential cooperation opportunities while ensuring the accuracy of recommendation results, thereby strengthening the comprehensiveness of enterprise-service matching.

[0027] Furthermore, generating candidate connection relationships between enterprises and services based on node relationships in the knowledge graph specifically includes: Calculate the structural similarity based on the multi-hop path between enterprise nodes and service nodes, and select node pairs whose structural similarity meets the first preset threshold as a candidate docking edge set; Semantic path similarity is evaluated for the corresponding attribute node paths in the candidate docking edge set; the semantic path similarity is obtained by jointly calculating the path label matching degree and the path embedding similarity. The semantic path credibility is determined based on the semantic similarity evaluation results and the weight distribution of the attribute node path in the knowledge graph; the weight distribution of the attribute node path is obtained by weighted calculation of the edge weights of each relation edge in the path. The comprehensive matching score for each candidate docking edge is determined based on structural similarity and semantic path credibility.

[0028] In this embodiment of the invention, a multi-hop path between an enterprise node and a service node refers to a sequence of paths in a knowledge graph that starts from an enterprise node, passes through several intermediate nodes and relational edges, and finally reaches a service node. Unlike the attribute node multi-hop paths from enterprise nodes to service nodes described earlier, this multi-hop path is not limited to passing through attribute nodes but may also pass through other types of business-related nodes. It focuses on reflecting the connection patterns and structural characteristics of enterprises and services within the overall graph structure, rather than simply relying on transmission paths that depend on shared attribute characteristics.

[0029] Based on this, structural similarity calculation is performed on multi-hop paths to measure the similarity of connection patterns between different enterprise service node pairs at the graph structure level. Node pairs with structural similarity reaching a first preset threshold are then selected to form a candidate docking edge set. In this invention, a candidate docking edge refers to an enterprise service node pair that has high structural similarity, may have business matching potential, but has not yet established a formal docking relationship. The corresponding attribute node paths in the candidate docking edge set refer to all paths through attribute nodes in the knowledge graph for these candidate docking edges, used to further evaluate their potential cooperative relationships from a semantic level.

[0030] Subsequently, this invention performs semantic path similarity evaluation on each attribute node path, a process that integrates two types of indicators: path label matching degree and path embedding similarity. Path label matching degree is used to determine the degree of direct semantic correspondence between the labels of each node and edge in the path, such as attribute category and relationship type; path embedding similarity calculates the similarity of paths in high-dimensional space based on vectorized representation methods, thereby capturing cases where the labels are different on the surface but semantically similar.

[0031] After obtaining the semantic similarity assessment results, this invention further combines the weight distribution of attribute node paths in the knowledge graph to determine the credibility of semantic paths. Path weight distribution is a graph term that weights the edge weights of each relation edge in a path according to their order in the path, reflecting the importance and confidence level of the path in the overall spectrum. By combining semantic similarity with weight distribution, it is possible to effectively suppress false paths that are semantically similar but based on low-confidence edges.

[0032] Furthermore, the comprehensive matching score of each candidate connection edge is calculated by combining structural similarity and semantic path credibility. This score is used to reflect the overall matching degree of the candidate connection edge in both structural and semantic dimensions, thereby providing a quantitative basis for subsequent accurate recommendation and connection.

[0033] In this invention, by combining structural similarity with semantic path credibility, it is possible to capture the implicit structural connection patterns between enterprises and services, and to ensure that such structural relationships have business rationality and data support at the semantic level. This significantly improves the accuracy and reliability of candidate connection edge screening and reduces the generation of invalid or low-value connection suggestions.

[0034] Further, the operation of calculating the matching probability of the candidate docking relationship through the uncertainty propagation algorithm is performed in the Bayesian network model; wherein, the Bayesian network model construction process specifically includes: Each candidate docking edge is used as a node to be inferred in the Bayesian network model, and the comprehensive matching score corresponding to the candidate docking edge is used as the initial probability value and input into the Bayesian network model. Based on the knowledge graph structure, a Bayesian network model structure is constructed with candidate docking edges as target nodes and attribute nodes involved in the semantic path of the candidate docking edges as intermediary nodes.

[0035] Furthermore, the process of calculating the matching probability of the candidate docking relationship using the uncertainty propagation algorithm specifically includes: The prior probability of the enterprise is calculated based on the static attribute features and behavioral features of the enterprise nodes corresponding to the candidate docking edges, and the prior probability is corrected according to the connection strength of the enterprise nodes in the knowledge graph; the connection strength of the enterprise nodes is obtained based on the number of edge weights and interaction frequency of the enterprise nodes with other nodes in the knowledge graph. The conditional probability of the server is calculated based on the dynamic evaluation feature information of the service node corresponding to the candidate docking edge. Based on the prior probability of the enterprise side and the conditional probability of the server side, the algorithm iteratively propagates within the Bayesian network model structure, updating the matching probability of each candidate connection edge until convergence.

[0036] In this embodiment of the invention, each candidate connection edge selected in the knowledge graph is considered as a node to be inferred in the Bayesian network model because the model needs to update and infer the probability states of these nodes. Prior to this, each candidate connection edge has already obtained a comprehensive matching score. By using this score as the initial probability value of the node and inputting it into the Bayesian network model, it means that the model already possesses prior judgment criteria based on the fusion of structural similarity and semantic credibility before inference begins.

[0037] In terms of model structure construction, the overall structure of the knowledge graph is used as a reference. Candidate connection edges are treated as target nodes, while attribute nodes traversed by the semantic paths involved in these edges are set as intermediary nodes. This is because attribute nodes themselves carry shared feature information between enterprises and services, and in Bayesian networks, they can serve as intermediate links for information transmission and probability correction. Through this combination of target nodes and intermediary nodes, Bayesian networks can capture the conditional dependencies between enterprise nodes, service nodes, and their shared attributes, thereby more accurately modeling the matching probability of candidate connection edges.

[0038] In the specific uncertainty propagation algorithm calculation process, this invention first calculates the prior probability of the enterprise node corresponding to the candidate connection edge based on its static attribute characteristics and behavioral characteristics. For example, an enterprise with many successful call records in the same service domain will have a correspondingly higher prior probability. Subsequently, the prior probability is corrected according to the connection strength of the enterprise node in the knowledge graph. The connection strength can be calculated by statistically analyzing the number of edge weights and interaction frequency between the enterprise node and other nodes. For example, if an enterprise frequently interacts with multiple high-weight service nodes, its connection strength index is higher, and the credibility of the prior probability is thus improved.

[0039] The conditional probability calculation on the server side is based on the dynamic evaluation features of the service nodes corresponding to the candidate connection edges. For example, the success rate of recent calls from different enterprises, average response time, and service quality score can all be used as input factors for the conditional probability calculation. Through dynamic evaluation features, the model can reflect the actual performance of the service nodes in the current time period, thereby avoiding matching bias caused by discrepancies between historical performance and the current situation.

[0040] Finally, the prior probabilities of the enterprise and the conditional probabilities of the server are used as inputs for iterative propagation within the constructed Bayesian network structure. The propagation process uses mediator attribute nodes to influence and correct the probability information of different candidate connection edges, continuously updating the matching probabilities of each candidate connection edge in multiple iterations until convergence to a stable value. In this process, the uncertainty propagation algorithm balances the weights of directly observed information and structured association information, thus outputting more globally consistent matching probability results. For example, when a candidate connection edge has a high direct matching score, but the probabilities of multiple attribute nodes along its associated path are low, the final convergence result will appropriately reduce the matching probability of that edge to reflect the uncertainty of the overall network environment.

[0041] In this invention, a Bayesian network model combined with an uncertainty propagation algorithm is used to extend the process of calculating the matching probability between enterprises and services from a single score judgment to a structured probabilistic reasoning process. By introducing the initial probability of candidate connection edges, mediator attribute nodes, and knowledge graph connection strength correction strategies, the model can comprehensively utilize structural information, attribute information, and dynamic performance to achieve a more accurate and robust probability assessment of candidate connection relationships, thereby improving the credibility of the matching results.

[0042] Furthermore, as a preferred embodiment, this embodiment is a more detailed description based on the above technical solution. The method for calculating the granular input in the uncertainty propagation is not unique and does not affect the implementation of the above technical solution. Candidate docking edges... Each related attribute node path ( The support and inhibition factors of the set of attribute node paths associated with edge e are compressed into a logarithmic ratio. As a granular input in the propagation of uncertainty, then:

[0043]

[0044]

[0045] in, This is the uncertainty attenuation factor, with a value range of (0,1]. It is set according to the sparsity of the knowledge graph, the distribution of relation edge weights, and the requirements for convergence speed. As the supporting evidence kernel for the path, the path label matching degree is... Similarity with path embedding Perform geometric equilibrium, then superimpose the cumulative decay of the number of relation edge weights; this forms the path suppression evidence kernel, using... , and the right of rebuttal Construct opposing channels to form a counterbalancing structure of logarithmic difference; Let t be the set of messages related to the neighborhood of path p in round t; Variance (a measure of uncertainty); This is the hop count decay coefficient (the longer the path and the farther the edge, the smaller the contribution). This is a balance coefficient used for control. and The balance; , These are the support / disproven weights for the r-th edge on the path, and all values ​​are > 0. This is the hop number of the edge in the path.

[0046] The above formula applies the logarithmic difference of the support kernel and the suppression kernel to the messages of path p, and adds uncertainty attenuation related to the variance of the path's neighborhood message set. When the neighborhood divergence is large (high variance), the contribution of the path is automatically reduced. This not only transforms the semantic evidence of multi-hop paths of attribute nodes into propagable quantifiable messages, but also explicitly controls uncertainty at the propagation level, achieving adaptive optimization of the matching probability of candidate docking relationships. Thus, it effectively improves the accuracy of matching results in complex knowledge graph structures.

[0047] Furthermore, the method also includes: Identify abnormal patterns of enterprises and services based on historical connection records; the abnormal patterns include abnormal connection patterns and matching failure patterns. In the process of matching probability calculation, abnormal patterns are introduced as negative samples. A penalty coefficient is introduced for candidate matching relationships that belong to negative samples, and their priority in the enterprise service matching recommendation list is reduced.

[0048] Furthermore, the penalty coefficient is dynamically adjusted based on the frequency of abnormal patterns and the matching failure rate; Before generating the enterprise service matching recommendation list, the final recommendation probability is obtained by multiplying the matching probability by a correction factor calculated based on the penalty coefficient; the final recommendation probability combines the positive matching score and the negative sample rejection effect.

[0049] Furthermore, the process of generating the enterprise service matching recommendation list includes: The candidate enterprise service connections are ranked according to the final recommendation probability; When the recommendation probabilities of multiple candidate docking relationships are similar, the candidate docking relationship that has not triggered the negative sample rejection mechanism shall be selected first. Output a list of recommended enterprise service partners, sorted by priority.

[0050] Specifically, in this embodiment of the invention, an abnormal pattern refers to a connection behavior pattern that deviates significantly from normal business matching patterns, identified by analyzing historical enterprise-service connection records. This pattern includes two categories: first, an abnormal connection pattern, where an enterprise and a service have been connected multiple times in historical data, but ultimately fail to form an effective cooperation or clearly violate conventional business logic; for example, an enterprise frequently requests to connect with services unrelated to its business. Second, a matching failure pattern, where a certain type of candidate connection relationship is repeatedly recommended to the enterprise in historical matching, but ultimately fails to establish an effective cooperation. These abnormal patterns are considered negative signals in this invention, used to identify potentially low-value or even invalid connection relationships.

[0051] In the matching probability calculation process, this invention introduces abnormal patterns as negative samples into the probability correction calculation. Negative samples refer to candidate matchmaking relationships identified as abnormal patterns in historical data. For these candidate matchmaking relationships that are considered negative samples, a penalty coefficient is applied to reduce their ranking priority in the final recommendation list. The penalty coefficient is dynamically adjusted based on the frequency of abnormal patterns and the matching failure rate. For example, when the frequency of a certain type of abnormal pattern increases significantly in the past three months, and the corresponding matching failure rate exceeds a preset threshold, the penalty coefficient is automatically increased, thereby further reducing its recommendation priority and avoiding the repeated pushing of invalid matchmaking relationships.

[0052] Before generating the enterprise service matching recommendation list, the aforementioned matching probability is multiplied by a correction factor calculated based on a penalty coefficient to obtain the final recommendation probability. The final recommendation probability not only integrates positive matching scores but also incorporates the exclusion effect of negative samples, ensuring that high-quality matching relationships have an advantage in the ranking, while low-value or invalid relationships are automatically downweighted.

[0053] During the generation of the recommendation list, all candidate enterprise service connections are first sorted according to the final recommendation probability. When multiple candidate connections have similar recommendation probabilities, the candidate connection that has not triggered the negative sample exclusion mechanism will be prioritized to ensure the diversity and effectiveness of the recommendation results. Finally, the output enterprise service connection recommendation list is an ordered list arranged by priority, which can be directly provided to enterprises for connection reference, thereby improving the overall connection success rate and platform resource utilization efficiency.

[0054] Example 2 The second aspect of this invention discloses an artificial intelligence-based enterprise service precision matching system, the system comprising: The first construction module is used to build a set of enterprise nodes based on enterprise attribute information, business operation records and service call records on the enterprise side. The second construction module is used to build a set of service nodes based on the service attribute information of the server, the service call record, and the historical success record. The calculation module is used to calculate the matching probability of the candidate docking relationship through an uncertainty propagation algorithm, and obtain the matching probability distribution between enterprises and services; The list generation module is used to generate a recommended list of enterprise service connections based on the matching probability distribution and preset connection decision rules.

[0055] It should be noted that the specific implementation process of Example 2 is similar to that of Example 1, and will not be repeated in Example 2.

[0056] Finally, it should be noted that the above-described embodiments include multiple parallel implementations of the present invention. Deleting or otherwise adjusting one or more implementations will not affect the implementation of the solution. Furthermore, the enterprise service precision matching method and system based on artificial intelligence disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. 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 of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for precise matching of enterprise services based on artificial intelligence, characterized in that, The method includes: A set of enterprise nodes is constructed based on enterprise attribute information, business operation records, and service call records from the enterprise side. A set of service nodes is constructed based on the service attribute information, service call records, and historical success records of the server. An enterprise service knowledge graph is constructed based on the enterprise node set and the service node set, and candidate docking relationships between enterprises and services are generated based on the node relationships in the knowledge graph. The matching probability of the candidate docking relationships is calculated using an uncertainty propagation algorithm to obtain the matching probability distribution between enterprises and services; A list of recommended enterprise service connections is generated based on the matching probability distribution and the preset connection decision rules.

2. The method for precise matching of enterprise services based on artificial intelligence according to claim 1, characterized in that, The enterprise service knowledge graph also includes attribute nodes and relationship edges; among them... The enterprise node is used to record the static attribute characteristics and behavioral characteristics of the enterprise; the behavioral characteristics include a sequence of behavioral operations generated based on business operation records and service call records. The service node is used to record the static attribute characteristics and dynamic evaluation characteristics of the service; the dynamic evaluation characteristics include performance evaluation and service call statistics generated based on service call records and historical success records. The attribute nodes are used to record shared characteristic information of enterprises and services.

3. The method for precise matching of enterprise services based on artificial intelligence according to claim 2, characterized in that, The process of determining the weights of the relation edges includes: The dynamic feature matching degree is determined based on the similarity between enterprise behavioral characteristics and service dynamic evaluation characteristics; Static feature matching degree is determined based on the similarity between enterprise static attribute features and service static attribute features; Calculate the potential correlation degree based on the multi-hop path of attribute nodes from enterprise nodes to service nodes; The dynamic feature matching degree, static feature matching degree, and potential correlation degree are fused together according to preset weights to generate relation edge weights.

4. The method for precise matching of enterprise services based on artificial intelligence according to claim 3, characterized in that, The process of generating candidate connection relationships between enterprises and services based on the node relationships in the knowledge graph specifically includes: Calculate the structural similarity based on the multi-hop path between enterprise nodes and service nodes, and select node pairs whose structural similarity meets the first preset threshold as a candidate docking edge set; Semantic path similarity is evaluated for the corresponding attribute node paths in the candidate docking edge set; the semantic path similarity is obtained by jointly calculating the path label matching degree and the path embedding similarity. The semantic path credibility is determined based on the semantic similarity evaluation results and the weight distribution of the attribute node path in the knowledge graph; the weight distribution of the attribute node path is obtained by weighted calculation of the edge weights of each relation edge in the path. The comprehensive matching score for each candidate docking edge is determined based on structural similarity and semantic path credibility.

5. The method for precise matching of enterprise services based on artificial intelligence according to claim 4, characterized in that, The operation of calculating the matching probability of the candidate docking relationship using the uncertainty propagation algorithm is performed in the Bayesian network model; wherein, the Bayesian network model construction process specifically includes: Each candidate docking edge is used as a node to be inferred in the Bayesian network model, and the comprehensive matching score corresponding to the candidate docking edge is used as the initial probability value and input into the Bayesian network model. Based on the knowledge graph structure, a Bayesian network model structure is constructed with candidate docking edges as target nodes and attribute nodes involved in the semantic path of the candidate docking edges as intermediary nodes.

6. The method for precise matching of enterprise services based on artificial intelligence according to claim 5, characterized in that, The process of calculating the matching probability of the candidate docking relationship using the uncertainty propagation algorithm specifically includes: The prior probability of the enterprise is calculated based on the static attribute features and behavioral features of the enterprise nodes corresponding to the candidate docking edges, and the prior probability is corrected according to the connection strength of the enterprise nodes in the knowledge graph; the connection strength of the enterprise nodes is obtained based on the number of edge weights and interaction frequency of the enterprise nodes with other nodes in the knowledge graph. The conditional probability of the server is calculated based on the dynamic evaluation feature information of the service node corresponding to the candidate docking edge. Based on the prior probability of the enterprise side and the conditional probability of the server side, the algorithm iteratively propagates within the Bayesian network model structure, updating the matching probability of each candidate connection edge until convergence.

7. The method for precise matching of enterprise services based on artificial intelligence according to claims 1-6, characterized in that, The method further includes: Identify abnormal patterns of enterprises and services based on historical connection records; the abnormal patterns include abnormal connection patterns and matching failure patterns. In the process of matching probability calculation, abnormal patterns are introduced as negative samples. A penalty coefficient is introduced for candidate matching relationships that belong to negative samples, and their priority in the enterprise service matching recommendation list is reduced.

8. The method for precise matching of enterprise services based on artificial intelligence according to claim 7, characterized in that, The penalty coefficient is dynamically adjusted based on the frequency of abnormal patterns and the matching failure rate. Before generating the enterprise service matching recommendation list, the final recommendation probability is obtained by multiplying the matching probability by a correction factor calculated based on the penalty coefficient; the final recommendation probability combines the positive matching score and the negative sample rejection effect.

9. The method for precise matching of enterprise services based on artificial intelligence according to claim 8, characterized in that, The process of generating the enterprise service integration recommendation list includes: The candidate enterprise service connections are ranked according to the final recommendation probability; When the recommendation probabilities of multiple candidate docking relationships are similar, the candidate docking relationship that has not triggered the negative sample rejection mechanism shall be selected first. Output a list of recommended enterprise service partners, sorted by priority.

10. A precise matching system for enterprise services based on artificial intelligence, characterized in that, The system includes: The first construction module is used to build a set of enterprise nodes based on enterprise attribute information, business operation records and service call records on the enterprise side. The second construction module is used to build a set of service nodes based on the service attribute information of the server, the service call record, and the historical success record. The calculation module is used to calculate the matching probability of the candidate docking relationship through an uncertainty propagation algorithm, and obtain the matching probability distribution between enterprises and services; The list generation module is used to generate a recommended list of enterprise service connections based on the matching probability distribution and preset connection decision rules.

Images