Customer service method and system based on machine learning and knowledge graph

Abstract

The invention discloses a customer service method and system based on machine learning and a knowledge graph. The method comprises the following steps: processing target information to obtain predicted business opportunity information output by a meta-learning model; determining a first risk assessment index between the to-be-assessed enterprise and a recommended enterprise in the predicted business opportunity information; determining a recommended enterprise topological structure corresponding to a recommended enterprise in the predicted business opportunity information in the enterprise knowledge graph; determining a plurality of second risk assessment indexes between the to-be-assessed enterprise and a plurality of candidate enterprises in the plurality of candidate enterprise topological structures, and if a third risk assessment index smaller than a second preset threshold exists in the plurality of second risk assessment indexes, determining a minimum target risk assessment index in the third risk assessment indexes, and determining a target candidate enterprise corresponding to the target risk assessment index. The technical problem that potential business opportunities hidden behind a small amount of data cannot be accurately mined by related technologies for a data scarcity scene is solved.

Description

Technical Field

[0001] This application relates to the field of data analysis, and in particular, to a customer service method and system based on machine learning and knowledge graph. Background Art

[0002] In the real business environment, it is difficult to obtain a large amount of data in many cases, such as emerging markets, rare events, niche product fields, etc. In these scenarios with scarce data, related technologies cannot accurately mine the potential business opportunities hidden behind a small amount of data.

[0003] For the above problems, no effective solutions have been proposed yet. Summary of the Invention

[0004] This application provides a customer service method and system based on machine learning and knowledge graph to at least solve the technical problem that related technologies cannot accurately mine the potential business opportunities hidden behind a small amount of data in scenarios with scarce data.

[0005] According to one aspect of this application, a customer service method based on machine learning and knowledge graph is provided, including: processing target information by using a pre-trained meta-learning model to obtain predicted business opportunity information output by the meta-learning model, where the target information at least includes: enterprise data of the enterprise to be evaluated and operation data of the market, where the enterprise data at least includes: enterprise attribute information and enterprise demand information, and the operation data at least includes: supply chain status information and industry specification information; receiving the causal path in the predicted business opportunity information, determining the confidence level of the predicted business opportunity information according to the causal path in the predicted business opportunity information, and determining the first risk assessment index between the enterprise to be evaluated and the recommended enterprise in the predicted business opportunity information when the confidence level is greater than a first preset threshold; determining the recommended enterprise topology structure corresponding to the recommended enterprise in the predicted business opportunity information in the enterprise knowledge graph and determining multiple candidate enterprise topology structures with a similarity greater than a third preset threshold to the recommended enterprise topology structure in the enterprise knowledge graph when the risk assessment index is greater than a second preset threshold; determining multiple second risk assessment indexes between the enterprise to be evaluated and multiple candidate enterprises in the multiple candidate enterprise topology structures, if there is a third risk assessment index less than the second preset threshold among the multiple second risk assessment indexes, determining the minimum target risk assessment index in the third risk assessment indexes and determining the target candidate enterprise corresponding to the target risk assessment index.

[0006] Optionally, the meta-learning model is trained by the following method: Obtain multiple training tasks, where each training task includes: a support set and a query set. The support set at least includes: enterprise data of a first enterprise with a preset causal path and operation data of the market, and first time-series data of the first enterprise. The first enterprise is an enterprise with a first label, and the first label is used to indicate that cooperation has been completed between enterprises. The first time-series data is time-series data of the return rate of the first enterprise before and after cooperation; the query set at least includes: enterprise data and operation data without a preset causal path, and dynamic data. The dynamic data at least includes: interest rate fluctuation data and natural disaster data; for each training task, use the support set to update the pre-configured meta-learning model. During the update process, use stochastic gradient descent to calculate the gradient of the loss function with respect to the initial model parameters, update the initial model parameters according to the gradient to obtain the target model parameters, and calculate the loss value of the query set under the target model parameters to obtain the loss value corresponding to each training task; Aggregate the loss values corresponding to each training task to obtain the target loss value, and based on the target loss value, update the initial model parameters to obtain the trained meta-learning model.

[0007] Optionally, determining the confidence level of the predicted business opportunity information according to the causal path in the predicted business opportunity information includes: encoding the causal path in the predicted business opportunity information to obtain a first path feature vector, and encoding multiple preset causal paths in the support set to obtain multiple second path feature vectors, where the length of the first path feature vector is the same as the length of the multiple second path feature vectors; determining a similarity matrix between the first path feature vector and each second path feature vector to obtain multiple similarity matrices; adding the elements at the corresponding positions of each similarity matrix to obtain an accumulation matrix, dividing each element in the accumulation matrix by the number of similarity matrices to obtain a mean matrix, and determining the global mean of the mean matrix as the causal path matching degree; determining the model prediction variance of the pre-trained meta-learning model, and obtaining a first weight corresponding to the causal path matching degree and a second weight corresponding to the model prediction variance; determining the confidence level of the predicted business opportunity information according to the causal path matching degree, the model prediction variance, the first weight, and the second weight.

[0008] Optionally, the first weight and the second weight are dynamically generated by a reinforcement learning model. The state in the reinforcement learning model includes: an enterprise risk preference index and a historical confidence accuracy rate. The actions in the reinforcement learning model include: adjusting the first weight and the second weight to maximize the long-term reward function, where the long-term reward function is determined according to the business opportunity conversion rate. The reward signal of the reinforcement learning model includes: whether the predicted business opportunity information is marked as adopted.

[0009] Optionally, determining multiple candidate enterprise topologies in the enterprise knowledge graph whose similarity to the recommended enterprise topology is greater than a third preset threshold includes: extracting a first neighborhood subgraph centered on the recommended enterprise in the enterprise knowledge graph, and using a graph neural network to encode the first neighborhood subgraph to obtain a first topology embedding vector; determining the business domain constraints corresponding to the recommended enterprise, and screening out an initial candidate enterprise set that meets the business domain constraints in the enterprise knowledge graph; in the initial candidate enterprise set, extracting the neighborhood subgraphs of each candidate enterprise to obtain multiple second neighborhood subgraphs, and using a graph neural network to encode each second neighborhood subgraph to obtain multiple second topology embedding vectors; calculating the similarity between the first topology embedding vector and each second topology embedding vector to obtain multiple similarities; determining multiple neighborhood subgraphs corresponding to the multiple second topology embedding vectors whose similarities are greater than the third preset threshold.

[0010] Optionally, obtaining the expected payback period of the enterprise to be evaluated; in the case where the expected payback period is less than the preset threshold, extracting a first neighborhood subgraph with n hops centered on the recommended enterprise in the enterprise knowledge graph, where n is a positive integer; in the case where the expected payback period is greater than the preset threshold, extracting a first neighborhood subgraph with m hops centered on the recommended enterprise in the enterprise knowledge graph, where m is a positive integer greater than n.

[0011] Optionally, determine the first risk assessment indicator between the enterprise to be evaluated and the recommended enterprises in the predicted business opportunity information, including: extracting multi-source data of the first enterprise and the second enterprise from the enterprise knowledge graph, where the multi-source data includes: financial data, market performance data, operation data, industry data, and enterprise association data. The financial data includes: target indicators in the balance sheet, income statement, and cash flow statement. The market performance data includes: market share, brand awareness, and product market price fluctuations. The operation data includes: production efficiency, inventory turnover rate, supply chain stability, and customer satisfaction. The industry data includes: industry growth rate and industry competition situation. The enterprise association data includes: the association relationship between enterprises; preprocess the multi-source data, and extract features from the preprocessed data to obtain the target feature vector, where the preprocessing at least includes: removing outliers, filling missing values, unifying data formats, and data standardization; input the target feature vector into the risk assessment model to obtain the first risk assessment indicator between the enterprise to be evaluated and the recommended enterprises, where the risk assessment model is trained by the following method: collect historical data of enterprises from multiple data sources, clean the historical data, fill missing values, correct errors, and unify the format, and based on domain knowledge, extract features related to risk assessment from the processed historical data to obtain the target data set; divide the target data set into a training set and a validation set, and train the risk assessment model based on the training set and the validation set until the preset training stop condition is met, where the risk assessment model is a neural network model.

[0012] According to another aspect of the present application, there is also provided a customer service system based on machine learning and knowledge graph, including: a generation module, configured to process target information by using a pre-trained meta-learning model to obtain predicted business opportunity information output by the meta-learning model, where the target information at least includes: enterprise data of an enterprise to be evaluated and operation data of the market, where the enterprise data at least includes: enterprise attribute information and enterprise demand information, and the operation data at least includes: supply chain status information and industry specification information; a first determination module, configured to receive a causal path in the predicted business opportunity information, determine a confidence level of the predicted business opportunity information according to the causal path in the predicted business opportunity information, and determine a first risk assessment index between the enterprise to be evaluated and a recommended enterprise in the predicted business opportunity information when the confidence level is greater than a first preset threshold; a second determination module, configured to, when the risk assessment index is greater than a second preset threshold, determine a topology structure of the enterprise to be evaluated corresponding to the enterprise to be evaluated in the enterprise knowledge graph, and determine a plurality of candidate enterprise topology structures in the enterprise knowledge graph whose similarity to the topology structure of the enterprise to be evaluated is greater than a third preset threshold; a third determination module, configured to determine a plurality of second risk assessment indexes between the enterprise to be evaluated and a plurality of candidate enterprises in the plurality of candidate enterprise topology structures, if there is a third risk assessment index less than the second preset threshold among the plurality of second risk assessment indexes, determine the smallest target risk assessment index in the third risk assessment indexes, and determine a target candidate enterprise corresponding to the target risk assessment index.

[0013] According to another aspect of the present application, there is also provided a non-volatile storage medium, where the storage medium includes a stored program, and when the program runs, it controls a device where the storage medium is located to execute the above-mentioned customer service method based on machine learning and knowledge graph.

[0014] According to another aspect of the present application, there is also provided an electronic device, including: a memory and a processor, where the processor is configured to run a program stored in the memory, and when the program runs, it executes the above-mentioned customer service method based on machine learning and knowledge graph.

[0015] According to another aspect of the present application, there is also provided a computer program, where when the computer program is executed by a processor, it implements the above-mentioned customer service method based on machine learning and knowledge graph.

[0016] According to another aspect of the present application, there is also provided a computer program product, where the computer program product includes a non-volatile computer-readable storage medium, where the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned customer service method based on machine learning and knowledge graph.

[0017] In this application, a pre-trained meta-learning model is used to process target information to obtain predicted business opportunity information output by the meta-learning model. The target information at least includes: enterprise data of the enterprise to be evaluated and operation data of the market. The enterprise data at least includes: enterprise attribute information and enterprise demand information. The operation data at least includes: supply chain status information and industry specification information. Receive the causal path in the predicted business opportunity information. According to the causal path in the predicted business opportunity information, determine the confidence level of the predicted business opportunity information. And when the confidence level is greater than the first preset threshold, determine the first risk assessment index between the enterprise to be evaluated and the recommended enterprise in the predicted business opportunity information. When the risk assessment index is greater than the second preset threshold, determine the recommended enterprise topology structure corresponding to the recommended enterprise in the predicted business opportunity information in the enterprise knowledge graph, and determine multiple candidate enterprise topology structures in the enterprise knowledge graph whose similarity to the recommended enterprise topology structure is greater than the third preset threshold. Determine multiple second risk assessment indexes between the enterprise to be evaluated and multiple candidate enterprises in the multiple candidate enterprise topology structures. If there is a third risk assessment index less than the second preset threshold among the multiple second risk assessment indexes, determine the smallest target risk assessment index among the third risk assessment indexes, and determine the target candidate enterprise corresponding to the target risk assessment index. In view of the data-scarce scenario, the potential business opportunities hidden behind a small amount of data are accurately mined through the meta-model, and the risk assessment indexes of the potential business opportunities are judged. When the risk assessment indexes do not meet the requirements, determine similar enterprises in the enterprise knowledge graph whose similarity to the enterprise indicated by the potential business opportunities meets the requirements. If the risk assessment indexes of the similar enterprises meet the requirements, determine the similar enterprises as candidate enterprises corresponding to the potential business opportunities, achieving the purpose of accurately mining the potential business opportunities hidden behind a small amount of data, thereby realizing the technical effect of assisting risk prediction and customer relationship management and enhancing the competitiveness of enterprises, and further solving the technical problem that in the data-scarce scenario, the related technologies cannot accurately mine the potential business opportunities hidden behind a small amount of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0019] Figure 1 is a flowchart of a customer service method based on machine learning and knowledge graph according to an embodiment of the present application;

[0020] Figure 2 is a structural diagram of a customer service system based on machine learning and knowledge graph according to an embodiment of the present application;

[0021] Figure 3It is a hardware structure block diagram of a computer terminal for a customer service method based on machine learning and knowledge graph according to an embodiment of the present application. Detailed implementation manners

[0022] In order to enable those skilled in the art of the present technology to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order different from those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0024] According to an embodiment of the present application, a method embodiment of a customer service method based on machine learning and knowledge graph is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0025] Figure 1 It is a flowchart of a customer service method based on machine learning and knowledge graph according to an embodiment of the present application. As Figure 1 shown, the method includes the following steps:

[0026] Step S102, process the target information by using a pre-trained meta-learning model to obtain predicted business opportunity information output by the meta-learning model, where the target information at least includes: enterprise data of the enterprise to be evaluated and operation data of the market, where the enterprise data at least includes: enterprise attribute information and enterprise demand information, and the operation data at least includes: supply chain status information and industry specification information.

[0027] In the above step S102, the enterprise attribute information includes but is not limited to: structured data such as enterprise scale (number of employees, registered capital), years of establishment, industry classification (such as NAICS code), technical qualifications (number of patents, proportion of R & D investment), etc. The enterprise demand information is, for example, the technical cooperation demand in the enterprise bidding documents extracted through natural language processing, and the gap description in the supply chain procurement plan (such as "urgently need a precision bearing supplier to replace imports").

[0028] The supply chain status information includes but is not limited to: on-time delivery rates of upstream and downstream enterprises (calculated through ERP system logs), inventory turnover rates, abnormal events at logistics nodes (such as satellite remote sensing monitoring data of port congestion). The industry specification information includes but is not limited to: specification texts (such as updates to environmental protection emission standards), requirements in industry technologies, and production capacity adjustment guidelines issued by industry associations.

[0029] Step S104: Receive the causal path in the predicted business opportunity information. According to the causal path in the predicted business opportunity information, determine the confidence level of the predicted business opportunity information, and when the confidence level is greater than the first preset threshold, determine the first risk assessment indicator between the enterprise to be evaluated and the recommended enterprise in the predicted business opportunity information.

[0030] Step S106: When the risk assessment indicator is greater than the second preset threshold, determine the recommended enterprise topology structure corresponding to the recommended enterprise in the enterprise knowledge graph, and determine multiple candidate enterprise topology structures in the enterprise knowledge graph whose similarity to the recommended enterprise topology structure is greater than the third preset threshold.

[0031] In the above step S106, the nodes of the enterprise knowledge graph represent various entity concepts in the enterprise knowledge system, including but not limited to employees within the enterprise (such as information about the employees' names, positions, departments they belong to, etc.), departments (department names, functions, levels, etc.), business processes (such as steps and links of processes like procurement processes, production processes, etc.), products (product models, specifications, categories they belong to, production batches, etc.), partners (names of enterprises such as suppliers, distributors, etc., cooperation areas, cooperation durations, etc.), customers (basic information of customers, purchase histories, customer grades, etc.), projects (names of projects, goals, progress, participants, etc.), technologies (names of technologies mastered or applied by the enterprise, technical parameters, technical sources, etc.), equipment (equipment numbers, models, uses, maintenance records, etc.), document materials (such as names, content summaries, release dates of internal enterprise rules and regulations documents, business documents, etc.).

[0032] The edges of an enterprise knowledge graph are used to represent the relationships between nodes, reflecting the association methods between different entities. Common edges include organizational relationships within an enterprise, such as the "subordination relationship" between employees and departments (indicating that an employee belongs to a certain department) and the "superior-subordinate relationship" between departments (clarifying the hierarchical management structure between departments); the "sequence relationship in business processes" (indicating the execution order of different process steps) and the "participant relationship" (associating the personnel or departments participating in a certain process); the "supply relationship" between products and partners (indicating that the supplier provides raw materials, etc. for the product) and the "purchase relationship" between products and customers (reflecting the purchase behavior of customers for products); the "participation relationship" between projects and employees (reflecting that employees participate in a certain project) and the "association relationship" between projects and departments (explaining the supporting role of departments for projects); the "application relationship" between technologies and products (the application situation of technologies in products) and the "association relationship" between technologies and equipment (certain equipment requires specific technologies for operation or maintenance); the "maintenance relationship" between equipment and maintenance personnel (who is responsible for the maintenance of equipment); the "association relationship" between document materials and departments or employees (which departments or employees need to follow or refer to this document), etc.

[0033] Step S108: Determine multiple second risk assessment indicators between the enterprise to be evaluated and multiple candidate enterprises in the topological structures of multiple candidate enterprises. If there is a third risk assessment indicator less than the second preset threshold among the multiple second risk assessment indicators, determine the smallest target risk assessment indicator among the third risk assessment indicators, and determine the target candidate enterprise corresponding to the target risk assessment indicator.

[0034] The above steps S102 to S108 accurately mine potential business opportunities hidden behind a small amount of data through the meta-model, and judge the risk assessment indicators of potential business opportunities. In the case where the risk assessment indicators do not meet the requirements, determine similar enterprises in the enterprise knowledge graph whose similarity to the enterprise indicated by the potential business opportunity meets the requirements. If the risk assessment indicators of the similar enterprises meet the requirements, determine the similar enterprises as the candidate enterprises corresponding to the potential business opportunity, achieving the purpose of accurately mining potential business opportunities hidden behind a small amount of data, thereby realizing the technical effects of assisting risk prediction and customer relationship management and enhancing the competitiveness of enterprises.

[0035] The following Figure 1 illustrates and explains the steps shown.

[0036] According to some alternative embodiments of the present application, the meta-learning model is trained by the following method: obtaining a plurality of training tasks, where each training task includes: a support set and a query set. The support set at least includes: enterprise data of a first enterprise with a preset causal path and operation data of the market, and first time-series data of the first enterprise, where the first enterprise is an enterprise carrying a first label, and the first label is used to indicate that cooperation has been completed between enterprises, and the first time-series data is time-series data of the return rate of the first enterprise before and after cooperation; the query set at least includes: enterprise data and operation data without a preset causal path, and dynamic data, where the dynamic data at least includes: interest rate fluctuation data and natural disaster data; for each training task, use the support set to update a pre-configured meta-learning model. During the update process, use stochastic gradient descent to calculate the gradient of the loss function with respect to the initial model parameters, update the initial model parameters according to the gradient to obtain target model parameters, and calculate the loss value of the query set under the target model parameters to obtain the loss value corresponding to each training task; perform an aggregation process on the loss values corresponding to each training task to obtain a target loss value, and based on the target loss value, update the initial model parameters to obtain a meta-learning model that has completed training.

[0037] In the above embodiment, for example, there is historical case 1: after the release of specification P1, enterprise A cooperates with supplier B, and the cost is reduced by 15%. For the above historical case 1, the support set is, for example: the original text of specification P1, the industry label of enterprise A, the time-series data of the return rate before and after cooperation, and the causal path: "specification P1 → industry subsidy → increased demand of enterprise A → cooperation with B". For example, there is historical case 2: specification P2 causes enterprise C to replace logistics provider D, and the efficiency is increased by 20%. For the above historical case 2, the support set is, for example: the abstract of specification P2, the scale attribute of enterprise C, the time-series data of the return rate before and after cooperation, and the causal path: "specification P2 → logistics subsidy → enterprise C selects D". The query set is, for example: a new specification P3 is released, the interest rate fluctuation data in the past month, and it is necessary to predict the potential cooperation business opportunities of enterprise E.

[0038] Through the preset causal path and time-series data in the support set, the model can learn causal chains such as "specification → technological investment → cooperation benefits". Compared with traditional meta-learning (which only relies on statistical correlation), it can improve the logical interpretability of the prediction results. In addition, after injecting interest rate fluctuations and disaster data into the query set, the business opportunity recognition accuracy of the model in emergency scenarios can be improved.

[0039] According to some other alternative embodiments of the present application, to determine the confidence level of the predicted business opportunity information based on the causal path in the predicted business opportunity information, it can be achieved through the following method: Encode the causal path in the predicted business opportunity information to obtain a first path feature vector, and encode multiple preset causal paths in the support set to obtain multiple second path feature vectors, where the length of the first path feature vector is the same as the lengths of the multiple second path feature vectors; Determine the similarity matrix between the first path feature vector and each second path feature vector to obtain multiple similarity matrices; Add the elements at the corresponding positions of each similarity matrix to obtain an accumulation matrix, divide each element in the accumulation matrix by the number of similarity matrices to obtain a mean matrix, and determine the global mean of the mean matrix as the causal path matching degree; Determine the model prediction variance of the pre-trained meta-learning model, and obtain the first weight corresponding to the causal path matching degree and the second weight corresponding to the model prediction variance; Determine the confidence level of the predicted business opportunity information based on the causal path matching degree, the model prediction variance, the first weight, and the second weight.

[0040] Preferably, the first weight and the second weight are dynamically generated by a reinforcement learning model. Among them, the state in the reinforcement learning model includes: the enterprise risk preference index and the historical confidence accuracy rate. The actions in the reinforcement learning model include: adjusting the first weight and the second weight to maximize the long-term reward function, where the long-term reward function is determined according to the business opportunity conversion rate. The reward signal of the reinforcement learning model includes: whether the predicted business opportunity information is marked as adopted.

[0041] Specifically, the enterprise risk preference index in the state space is used to quantify the enterprise's risk tolerance level (for example, when the enterprise selects a cost-sensitive cooperation B, it reflects a low risk preference); The historical confidence accuracy rate can be determined by the matching degree between the prediction results of the statistical model for similar specifications and the final adoption of the business opportunity. The rewards include: Immediate reward: When the predicted probability of supplier cooperation > threshold (such as 75%) and is marked as "under evaluation" by the enterprise, the reward is +0.5; Long-term reward: If the supplier finally signs a contract (business opportunity conversion), the reward is superimposed exponentially with the conversion delay (for example, the conversion reward within 30 days is +2, and within 60 days is +1). The reinforcement learning model forms a closed-loop feedback through the reward signal, so that the adjustment strategy of the first weight / second weight evolves over time.

[0042] For example, the causal path can be encoded through a graph attention network, controlling the dimensions of the first path feature vector and the multiple second path feature vectors to be d (such as 256 dimensions), padding with zeros when insufficient, and truncating when too long. Further, the cosine similarity is used to measure the similarity between the first path feature vector and each second path feature vector.

[0043] The first path feature vector is, for example, d = 256 - dimensional and describes the path of "lithium ore export restrictions → battery cost increase → demand for alternative materials". The second path feature vector is also d = 256 - dimensional and describes, for example, "cobalt price increase → ternary material substitution → lithium iron phosphate cooperation".

[0044] After calculating the similarity matrix, the average value of all elements in the mean matrix is 0.72, and the model prediction variance = 0.15, the dynamic adjustment after dynamic adjustment = 0.65, = 0.35, and the final confidence is Confidence = + * (1 - ) = 0.7025.

[0045] It can be understood that different enterprises (such as risk - averse and risk - preferring types) have significantly different tolerances for the causal path matching degree and the model prediction variance. Through reinforcement learning, the and can be dynamically adjusted according to the real - time input enterprise risk preference index. Using the historical confidence accuracy (such as the actual conversion rate of the business opportunities adopted in the past 100 predictions), the weight allocation strategy is corrected to avoid the disconnection between the static weight and the business reality. The long - term reward function is defined as the moving - window average of the business opportunity conversion rate (such as the conversion rate in the past 30 days), forcing the model to seek an optimal balance between exploration (trying new weight combinations) and exploitation (relying on historical effective weights), rather than only optimizing short - term indicators.

[0046] In the above steps, the weights are adjusted in real - time according to the enterprise risk profile and market feedback, and the conversion rate is taken as the core optimization goal to ensure a strong correlation between the technical solution and the business benefits.

[0047] As some optional embodiments of the present application, determining multiple candidate enterprise topological structures in the enterprise knowledge graph whose similarity with the recommended enterprise topological structure is greater than a third preset threshold can be achieved by the following method: extracting a first neighborhood subgraph centered on the recommended enterprise in the enterprise knowledge graph, and encoding the first neighborhood subgraph using a graph neural network to obtain a first topological structure embedding vector; determining the business domain constraints corresponding to the recommended enterprise, and screening out an initial set of candidate enterprises that meet the business domain constraints in the enterprise knowledge graph; in the initial set of candidate enterprises, extracting the neighborhood subgraph of each candidate enterprise to obtain multiple second neighborhood subgraphs, and encoding each second neighborhood subgraph using a graph neural network to obtain multiple second topological structure embedding vectors; calculating the similarity between the first topological structure embedding vector and each second topological structure embedding vector to obtain multiple similarities; determining multiple neighborhood subgraphs corresponding to multiple second topological structure embedding vectors whose similarity is greater than a third preset threshold. Wherein, node types include but are not limited to: enterprise entities (including attributes such as registered capital, industry classification), products, technologies, specifications, etc. Edge types include but are not limited to: supply chain relationships (upstream / downstream), equity investment, technical cooperation, and compliance constraints.

[0048] Furthermore, the expected payback period of the enterprise to be evaluated is obtained; when the expected payback period is less than a preset threshold, the first neighborhood subgraph of n hops centered on the recommended enterprise is extracted from the enterprise knowledge graph, where n is a positive integer; when the expected payback period is greater than the preset threshold, the first neighborhood subgraph of m hops centered on the recommended enterprise is extracted from the enterprise knowledge graph, where m is a positive integer greater than n.

[0049] In the above embodiment, the expected investment payback period T of the enterprise to be evaluated, for example, when T is less than 2, it is a short-term investment, and the enterprise focuses on direct suppliers / customers (1 hop) by default; when T is greater than 2, it is a long-term investment, and the enterprise focuses on indirect related parties by default.

[0050] It is worth noting that short-term investment (short payback period): requires quickly locking in directly related companies (such as tier 1 suppliers / customers), focusing on high-time cooperation relationships, and avoiding the introduction of noise data due to over-expansion of the neighborhood. Long-term investment (long payback period) requires exploring potential indirect connections (such as tier 3 technical partners), capturing cross-level strategic synergy opportunities, and breaking through local optimal limitations. For short-term investment scenarios, the number of subgraph nodes is controlled to 1 by limiting the number of hops (n=1). The time consumption of graph encoding is reduced from minutes to seconds. For long-term investment scenarios, a larger number of hops is allowed. Although the amount of calculation increases, scalability can be achieved through the distributed graph computing framework.

[0051] In summary, the above steps avoid missing opportunities caused by "using short hop counts for long cycles" or introducing noise due to "using long hop counts for short cycles". Moreover, resources are allocated on demand, short-term tasks are lightweighted, and long-term tasks are deepened, meeting the high-efficiency requirements of industrial-level systems.

[0052] As some alternative embodiments of the present application, determining the first risk assessment index between the enterprise to be evaluated and the recommended enterprise in the predicted business opportunity information can be achieved through the following method: Extract multi-source data of the first enterprise and the second enterprise from the enterprise knowledge graph, where the multi-source data includes: financial data, market performance data, operation data, industry data, and enterprise association data. The financial data includes: target indicators in the balance sheet, income statement, and cash flow statement. The market performance data includes: market share, brand awareness, and product market price fluctuations. The operation data includes: production efficiency, inventory turnover rate, supply chain stability, and customer satisfaction. The industry data includes: industry growth rate and industry competition situation. The enterprise association data includes: the association relationship between enterprises; preprocess the multi-source data, and extract features from the preprocessed data to obtain the target feature vector, where the preprocessing at least includes: removing outliers, filling in missing values, unifying the data format, and data standardization; input the target feature vector into the risk assessment model to obtain the first risk assessment index between the enterprise to be evaluated and the recommended enterprise, where the risk assessment model is trained through the following method: Collect historical data of enterprises from multiple data sources, clean the historical data, fill in missing values, correct errors, and unify the format, and extract features related to risk assessment from the processed historical data based on domain knowledge to obtain the target data set; divide the target data set into a training set and a validation set, and train the risk assessment model based on the training set and the validation set until the preset training stop condition is met, where the risk assessment model is a neural network model.

[0053] Through multi-source fusion, dynamic training, and deep association mining, the following breakthroughs are achieved: 1. Comprehensiveness: Cover internal and external risk factors of enterprises and solve the "blind spot" problem of traditional models. 2. Accuracy: Combine neural networks with domain knowledge to improve the ability to identify long-tail risks. 3. Interpretability: Visualize artificial features and association paths to meet the review requirements of financial institutions.

[0054] Figure 2 is a structural diagram of a customer service system based on machine learning and knowledge graph according to an embodiment of the present application, as Figure 2 shown, the system includes:

[0055] A generation module 22, configured to process target information by using a pre-trained meta-learning model to obtain predicted business opportunity information output by the meta-learning model, where the target information at least includes: enterprise data of an enterprise to be evaluated and operation data of the market, and the enterprise data at least includes: enterprise attribute information and enterprise demand information, and the operation data at least includes: supply chain status information and industry specification information.

[0056] A first determination module 24, configured to receive a causal path in the predicted business opportunity information, determine a confidence level of the predicted business opportunity information according to the causal path in the predicted business opportunity information, and determine a first risk assessment index between the enterprise to be evaluated and a recommended enterprise in the predicted business opportunity information when the confidence level is greater than a first preset threshold.

[0057] A second determination module 26, configured to, when the risk assessment index is greater than a second preset threshold, determine a topology structure of the enterprise to be evaluated corresponding to the enterprise to be evaluated in the enterprise knowledge graph, and determine a plurality of candidate enterprise topology structures in the enterprise knowledge graph whose similarity to the topology structure of the enterprise to be evaluated is greater than a third preset threshold.

[0058] A third determination module 28, configured to determine a plurality of second risk assessment indexes between the enterprise to be evaluated and a plurality of candidate enterprises in the plurality of candidate enterprise topology structures, and if there is a third risk assessment index less than the second preset threshold among the plurality of second risk assessment indexes, determine the smallest target risk assessment index among the third risk assessment indexes, and determine a target candidate enterprise corresponding to the target risk assessment index.

[0059] Optionally, the meta-learning model is trained through the following steps: obtaining a plurality of training tasks, where each training task includes: a support set and a query set, and the support set at least includes: enterprise data of a first enterprise with a preset causal path and operation data of the market, and first time-series data of the first enterprise, where the first enterprise is an enterprise with a first label, the first label is used to indicate that cooperation has been completed between enterprises, and the first time-series data is time-series data of the rate of return of the first enterprise before and after cooperation; the query set at least includes: enterprise data and operation data without a preset causal path, and dynamic data, and the dynamic data at least includes: interest rate fluctuation data and natural disaster data; for each training task, updating a pre-configured meta-learning model by using the support set, and in the updating process, calculating a gradient of a loss function with respect to initial model parameters by using stochastic gradient descent, updating the initial model parameters according to the gradient to obtain target model parameters, and calculating a loss value of the query set under the target model parameters to obtain a loss value corresponding to each training task; performing an aggregation process on the loss values corresponding to each training task to obtain a target loss value, and updating the initial model parameters based on the target loss value to obtain a trained meta-learning model.

[0060] Optionally, according to the causal path in the predicted business opportunity information, determine the confidence of the predicted business opportunity information, including the following steps: Encode the causal path in the predicted business opportunity information to obtain a first path feature vector, and encode multiple preset causal paths in the support set to obtain multiple second path feature vectors, where the length of the first path feature vector is the same as the length of the multiple second path feature vectors; Determine the similarity matrix between the first path feature vector and each second path feature vector to obtain multiple similarity matrices; Add the elements at the corresponding positions of each similarity matrix to obtain an accumulation matrix, divide each element in the accumulation matrix by the number of similarity matrices to obtain a mean matrix, and determine the global mean of the mean matrix as the causal path matching degree; Determine the model prediction variance of the pre-trained meta-learning model, and obtain the first weight corresponding to the causal path matching degree and the second weight corresponding to the model prediction variance; Determine the confidence of the predicted business opportunity information according to the causal path matching degree, the model prediction variance, the first weight, and the second weight.

[0061] Optionally, the first weight and the second weight are dynamically generated by a reinforcement learning model, where the state in the reinforcement learning model includes: enterprise risk preference indicators and historical confidence accuracy, and the actions in the reinforcement learning model include: adjusting the first weight and the second weight to maximize the long-term reward function, where the long-term reward function is determined according to the business opportunity conversion rate, and the reward signal of the reinforcement learning model includes: whether the predicted business opportunity information is marked as adopted.

[0062] Optionally, determine multiple candidate enterprise topological structures in the enterprise knowledge graph whose similarity to the recommended enterprise topological structure is greater than a third preset threshold, including the following steps: Extract a first neighborhood subgraph centered on the recommended enterprise in the enterprise knowledge graph, and use a graph neural network to encode the first neighborhood subgraph to obtain a first topological structure embedding vector; Determine the business domain constraints corresponding to the recommended enterprise, and screen out an initial candidate enterprise set in the enterprise knowledge graph that satisfies the business domain constraints; In the initial candidate enterprise set, extract the neighborhood subgraph of each candidate enterprise to obtain multiple second neighborhood subgraphs, and use a graph neural network to encode each second neighborhood subgraph to obtain multiple second topological structure embedding vectors; Calculate the similarity between the first topological structure embedding vector and each second topological structure embedding vector to obtain multiple similarities; Determine multiple neighborhood subgraphs corresponding to multiple second topological structure embedding vectors whose similarities are greater than the third preset threshold.

[0063] Optionally, the customer service system based on machine learning and knowledge graph is also used to perform the following steps: obtaining the expected payback period of the enterprise to be evaluated; in the case that the expected payback period is less than the preset threshold, extracting the first neighborhood subgraph of n hops centered on the recommended enterprise in the enterprise knowledge graph, where n is a positive integer; in the case that the expected payback period is greater than the preset threshold, extracting the first neighborhood subgraph of m hops centered on the recommended enterprise in the enterprise knowledge graph, where m is a positive integer greater than n.

[0064] Optionally, determining the first risk assessment index between the enterprise to be evaluated and the recommended enterprise in the predicted business opportunity information includes the following steps: extracting multi-source data of the first enterprise and the second enterprise in the enterprise knowledge graph, where the multi-source data includes: financial data, market performance data, operation data, industry data, and enterprise association data, the financial data includes: target indicators in the balance sheet, income statement, and cash flow statement, the market performance data includes: market share, brand awareness, product market price fluctuations, the operation data includes: production efficiency, inventory turnover rate, supply chain stability, customer satisfaction, the industry data includes: industry growth rate, industry competition situation, the enterprise association data includes: association relationships between enterprises; preprocessing the multi-source data, and extracting features from the preprocessed data to obtain the target feature vector, where the preprocessing at least includes: removing outliers, filling missing values, unifying data formats, and data standardization; inputting the target feature vector into the risk assessment model to obtain the first risk assessment index between the enterprise to be evaluated and the recommended enterprise, where the risk assessment model is trained by the following method: collecting historical data of enterprises from multiple data sources, cleaning, filling missing values, correcting errors, and unifying the format of the historical data, and extracting features related to risk assessment from the processed historical data based on domain knowledge to obtain the target data set; dividing the target data set into a training set and a validation set, and training the risk assessment model based on the training set and the validation set until the preset training stop condition is met, where the risk assessment model is a neural network model.

[0065] It should be noted that the above Figure 2 Each of the modules can be a program module (for example, a set of program instructions that implement a specific function), or a hardware module. For the latter, it can be presented in the following forms, but not limited to this: The manifestation form of each of the above modules is a processor, or the functions of each of the above modules are implemented by a processor.

[0066] It should be noted that Figure 2 The preferred implementation manners of the embodiments shown can be referred to Figure 1 the relevant descriptions of the embodiments shown, which will not be elaborated here.

[0067] Figure 3The hardware structure block diagram of a computer terminal for implementing a customer service method based on machine learning and knowledge graph is shown. As Figure 3 shown, the computer terminal 30 may include one or more (shown as 302a, 302b,..., 302n in the figure) processors 302 (the processor 302 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 304 for storing data, and a transmission module 306 for communication functions. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 3 the structure shown is only schematic and does not limit the structure of the above electronic device. For example, the computer terminal 30 may further include more or fewer components than Figure 3 shown, or have a different configuration from Figure 3 shown.

[0068] It should be noted that the above one or more processors 302 and / or other data processing circuits are generally referred to as "data processing circuits" herein. The data processing circuit may be embodied in software, hardware, firmware, or any combination thereof, in whole or in part. In addition, the data processing circuit may be a single independent processing module, or be incorporated in whole or in part into any one of the other elements in the computer terminal 30. As involved in the embodiments of the present application, the data processing circuit is a processor control.

[0069] The memory 304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the customer service method based on machine learning and knowledge graph in the embodiments of the present application. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, that is, implements the above customer service method based on machine learning and knowledge graph. The memory 304 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 304 may further include a memory remotely disposed relative to the processor 302, and these remote memories may be connected to the computer terminal 30 through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0070] The transmission module 306 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the computer terminal 30. In one example, the transmission module 306 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 306 can be a Radio Frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0071] The display can be, for example, a touch-screen liquid crystal display (LCD), which enables a user to interact with the user interface of the computer terminal 30.

[0072] It should be noted here that in some alternative embodiments, the above Figure 3 illustrated computer terminal may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware elements and software elements. It should be pointed out that Figure 3 is only an example of a specific specific instance and is intended to show the types of components that may exist in the above computer terminal.

[0073] It should be noted that Figure 3 the illustrated computer terminal is used to execute Figure 1 the customer service method based on machine learning and knowledge graph shown above. Therefore, the relevant explanations in the above method for executing commands also apply to this electronic device, which will not be elaborated here.

[0074] The embodiment of the present application also provides a non-volatile storage medium. The non-volatile storage medium includes a stored program, wherein when the program runs, it controls the device where the storage medium is located to execute the above-mentioned customer service method based on machine learning and knowledge graph.

[0075] A program for a non-volatile storage medium to perform the following functions: processing target information using a pre-trained meta-learning model to obtain predicted business opportunity information output by the meta-learning model, where the target information at least includes: enterprise data of the enterprise to be evaluated and operation data of the market, where the enterprise data at least includes: enterprise attribute information and enterprise demand information, and the operation data at least includes: supply chain status information and industry specification information; receiving a causal path in the predicted business opportunity information, determining the confidence level of the predicted business opportunity information according to the causal path in the predicted business opportunity information, and determining a first risk assessment index between the enterprise to be evaluated and the recommended enterprise in the predicted business opportunity information when the confidence level is greater than a first preset threshold; when the risk assessment index is greater than a second preset threshold, determining a recommended enterprise topology structure corresponding to the recommended enterprise in the predicted business opportunity information in the enterprise knowledge graph, and determining a plurality of candidate enterprise topology structures in the enterprise knowledge graph whose similarity to the recommended enterprise topology structure is greater than a third preset threshold; determining a plurality of second risk assessment indexes between the enterprise to be evaluated and a plurality of candidate enterprises in the plurality of candidate enterprise topology structures, if there is a third risk assessment index less than the second preset threshold among the plurality of second risk assessment indexes, determining the smallest target risk assessment index among the third risk assessment indexes, and determining a target candidate enterprise corresponding to the target risk assessment index.

[0076] An embodiment of the present application further provides an electronic device, including: a memory and a processor, where the processor is used to run a program stored in the memory, and when the program runs, it executes the above customer service method based on machine learning and knowledge graph.

[0077] The processor is used to run a program that performs the following functions: process target information using a pre-trained meta-learning model to obtain predicted business opportunity information output by the meta-learning model, where the target information at least includes: enterprise data of the enterprise to be evaluated and operation data of the market, and the enterprise data at least includes: enterprise attribute information and enterprise demand information, and the operation data at least includes: supply chain status information and industry specification information; receive the causal path in the predicted business opportunity information, determine the confidence level of the predicted business opportunity information according to the causal path in the predicted business opportunity information, and determine the first risk assessment index between the enterprise to be evaluated and the recommended enterprise in the predicted business opportunity information when the confidence level is greater than the first preset threshold; when the risk assessment index is greater than the second preset threshold, determine the recommended enterprise topology structure corresponding to the recommended enterprise in the predicted business opportunity information in the enterprise knowledge graph, and determine multiple candidate enterprise topology structures in the enterprise knowledge graph whose similarity to the recommended enterprise topology structure is greater than the third preset threshold; determine multiple second risk assessment indexes between the enterprise to be evaluated and multiple candidate enterprises in the multiple candidate enterprise topology structures, if there is a third risk assessment index less than the second preset threshold among the multiple second risk assessment indexes, determine the minimum target risk assessment index in the third risk assessment indexes, and determine the target candidate enterprise corresponding to the target risk assessment index.

[0078] The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0079] In the above embodiments of the present application, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0080] In the above embodiments of the present application, the information collected is information and data authorized by the user or fully authorized by all parties, and the processing of relevant data such as collection, storage, use, processing, transmission, provision, disclosure and application complies with relevant laws, regulations and standards, takes necessary protection measures, does not violate public order and good customs, and provides a corresponding operation entry for the user to choose to authorize or refuse.

[0081] In several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of the units or modules can be in an electrical or other form.

[0082] The unit described as a separation component may or may not be physically separated. The component displayed as a unit may or may not be a physical unit, that is, it may be located in one place or distributed across multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0083] In addition, each functional unit in various embodiments of the present application may be integrated into a processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit.

[0084] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the related technology, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs and other various media that can store program codes.

[0085] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A customer service method based on machine learning and knowledge graph, characterized in that, Including: Processing target information using a pre-trained meta-learning model to obtain predicted business opportunity information output by the meta-learning model, where the target information at least includes: enterprise data of the enterprise to be evaluated and operation data of the market, where the enterprise data at least includes: enterprise attribute information and enterprise demand information, and the operation data at least includes: supply chain status information and industry specification information; Receiving a causal path in the predicted business opportunity information, determining a confidence level of the predicted business opportunity information according to the causal path in the predicted business opportunity information, and determining a first risk assessment index between the enterprise to be evaluated and a recommended enterprise in the predicted business opportunity information when the confidence level is greater than a first preset threshold; When the risk assessment index is greater than a second preset threshold, determining a recommended enterprise topology structure corresponding to the recommended enterprise in the predicted business opportunity information in an enterprise knowledge graph, and determining a plurality of candidate enterprise topology structures with a similarity greater than a third preset threshold to the recommended enterprise topology structure in the enterprise knowledge graph; Determining a plurality of second risk assessment indexes between the enterprise to be evaluated and a plurality of candidate enterprises in the plurality of candidate enterprise topology structures, if there is a third risk assessment index less than the second preset threshold among the plurality of second risk assessment indexes, determining the smallest target risk assessment index among the third risk assessment indexes, and determining a target candidate enterprise corresponding to the target risk assessment index.

2. The method according to claim 1, wherein The meta-learning model is trained by the following method: Obtaining a plurality of training tasks, where each training task includes: a support set and a query set, and the support set at least includes: enterprise data of a first enterprise with a preset causal path and operation data of the market, and first time-series data of the first enterprise, where the first enterprise is an enterprise with a first label, and the first label is used to indicate that cooperation has been completed between enterprises, and the first time-series data is time-series data of the return rate of the first enterprise before and after cooperation; the query set at least includes: enterprise data and operation data without the preset causal path, and dynamic data, where the dynamic data at least includes: interest rate fluctuation data, natural disaster data; For each training task, updating a pre-configured meta-learning model using the support set, and during the updating process, calculating the gradient of the loss function for the initial model parameters using stochastic gradient descent, updating the initial model parameters according to the gradient to obtain target model parameters, and calculating the loss value of the query set under the target model parameters to obtain the loss value corresponding to each training task; Performing an aggregation process on the loss values corresponding to each training task to obtain a target loss value, and updating the initial model parameters based on the target loss value to obtain a trained meta-learning model.

3. The method according to claim 2, characterized in that, Determining the confidence level of the predicted business opportunity information according to the causal path in the predicted business opportunity information, including: Encode the causal paths in the predicted business opportunity information to obtain a first path feature vector, and encode multiple preset causal paths in the support set to obtain multiple second path feature vectors, where the length of the first path feature vector is the same as the lengths of the multiple second path feature vectors; Determine a similarity matrix between the first path feature vector and each of the second path feature vectors to obtain multiple similarity matrices; Add the elements at the corresponding positions of each similarity matrix to obtain an accumulation matrix, divide each element in the accumulation matrix by the number of similarity matrices to obtain an average matrix, and determine the global average of the average matrix as the causal path matching degree; Determine the model prediction variance of a pre-trained meta-learning model, and obtain a first weight corresponding to the causal path matching degree and a second weight corresponding to the model prediction variance; Determine the confidence level of the predicted business opportunity information according to the causal path matching degree, the model prediction variance, the first weight, and the second weight.

4. The method according to claim 3, wherein The first weight and the second weight are dynamically generated by a reinforcement learning model, where the state in the reinforcement learning model includes: an enterprise risk preference index and a historical confidence accuracy rate, and the actions in the reinforcement learning model include: adjusting the first weight and the second weight to maximize a long-term reward function, where the long-term reward function is determined according to the business opportunity conversion rate, and the reward signal of the reinforcement learning model includes: whether the predicted business opportunity information is marked as having been adopted.

5. The method according to claim 1, characterized in that, Determine multiple candidate enterprise topologies in the enterprise knowledge graph whose similarity to the recommended enterprise topology is greater than a third preset threshold, including: Extract a first neighborhood subgraph centered on the recommended enterprise in the enterprise knowledge graph, and use a graph neural network to encode the first neighborhood subgraph to obtain a first topology embedding vector; Determine the business domain constraint corresponding to the recommended enterprise, and screen out an initial candidate enterprise set in the enterprise knowledge graph that satisfies the business domain constraint; In the initial candidate enterprise set, extract the neighborhood subgraph of each candidate enterprise to obtain multiple second neighborhood subgraphs, and use a graph neural network to encode each of the second neighborhood subgraphs to obtain multiple second topology embedding vectors; Calculate the similarity between the first topology embedding vector and each of the second topology embedding vectors to obtain multiple similarities; Determine multiple neighborhood subgraphs corresponding to the multiple second topology embedding vectors whose similarities are greater than the third preset threshold.

6. The method according to claim 5, characterized in that, The method further includes: Obtain the expected payback period of the enterprise to be evaluated; In the case where the expected payback period is less than a preset threshold, extract an n-hop first neighborhood subgraph centered on the recommended enterprise in the enterprise knowledge graph, where n is a positive integer; In the case where the expected payback period is greater than the preset threshold, extract an m-hop first neighborhood subgraph centered on the recommended enterprise in the enterprise knowledge graph, where m is a positive integer greater than n.

7. The method according to claim 1, characterized in that Determine the first risk assessment indicator between the enterprise to be evaluated and the recommended enterprises in the predicted business opportunity information, including: Extract multi-source data of the first enterprise and the second enterprise from the enterprise knowledge graph. Among them, the multi-source data includes: financial data, market performance data, operation data, industry data, and enterprise association data. The financial data includes target indicators in the balance sheet, income statement, and cash flow statement. The market performance data includes market share, brand awareness, and product market price fluctuations. The operation data includes production efficiency, inventory turnover rate, supply chain stability, and customer satisfaction. The industry data includes industry growth rate and industry competition situation. The enterprise association data includes the association relationship between enterprises; Preprocess the multi-source data, and extract features from the preprocessed data to obtain the target feature vector. Among them, the preprocessing at least includes: removing outliers, filling missing values, unifying data formats, and data standardization; Input the target feature vector into the risk assessment model to obtain the first risk assessment indicator between the enterprise to be evaluated and the recommended enterprise. Among them, the risk assessment model is trained by the following method: collect historical data of enterprises from multiple data sources, clean the historical data, fill in missing values, correct errors, and unify the format, and extract features related to risk assessment from the processed historical data based on domain knowledge to obtain the target data set; divide the target data set into a training set and a validation set, and train the risk assessment model based on the training set and the validation set until the preset training stop condition is met. Among them, the risk assessment model is a neural network model.

8. A customer service system based on machine learning and knowledge graph, characterized in that, Including: A generation module for processing target information using a pre-trained meta-learning model to obtain the predicted business opportunity information output by the meta-learning model. Among them, the target information at least includes: enterprise data of the enterprise to be evaluated and operation data of the market. Among them, the enterprise data at least includes: enterprise attribute information and enterprise demand information. The operation data at least includes: supply chain status information and industry specification information; A first determination module for receiving the causal path in the predicted business opportunity information, determining the confidence of the predicted business opportunity information according to the causal path in the predicted business opportunity information, and determining the first risk assessment indicator between the enterprise to be evaluated and the recommended enterprise in the predicted business opportunity information when the confidence is greater than the first preset threshold; A second determination module for determining the topology structure of the enterprise to be evaluated corresponding to the enterprise to be evaluated in the enterprise knowledge graph and determining multiple candidate enterprise topology structures with a similarity greater than the third preset threshold to the topology structure of the enterprise to be evaluated in the enterprise knowledge graph when the risk assessment indicator is greater than the second preset threshold; A third determination module, configured to determine a plurality of second risk assessment indicators between the enterprise to be evaluated and a plurality of candidate enterprises in the topological structures of the plurality of candidate enterprises. If there is a third risk assessment indicator less than the second preset threshold among the plurality of second risk assessment indicators, determine the smallest target risk assessment indicator among the third risk assessment indicators, and determine the target candidate enterprise corresponding to the target risk assessment indicator.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored program, wherein when the program runs, it controls the device where the non-volatile storage medium is located to execute the customer service method based on machine learning and knowledge graph according to any one of claims 1 to 7.

10. An electronic device, characterized in that, Comprising: A memory and a processor, the processor is configured to run the program stored in the memory, wherein when the program runs, it executes the customer service method based on machine learning and knowledge graph according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the customer service method based on machine learning and knowledge graph according to any one of claims 1 to 7.

Images