CN-121996752-A - Knowledge graph-based heat supply service customer service answering method, equipment and medium

CN121996752ACN 121996752 ACN121996752 ACN 121996752ACN-121996752-A

Abstract

The application discloses a heating service customer service question-answering method, equipment and medium based on a knowledge graph, and relates to the technical field of artificial intelligent customer service; analyzing a heat supply service problem proposed by a user, generating a semantic vector of the enhanced problem, calculating semantic matching degree with each entity node in a heat supply question-answer knowledge graph, determining a search starting point, performing multi-hop traversal in the heat supply question-answer knowledge graph, collecting end nodes reached by a traversal path to form a candidate node set, calculating an expected benefit value through a reinforcement learning network model, screening target response nodes to generate a heat supply service response text, collecting interactive feedback data of the user, calculating a reward signal value, and optimizing the reinforcement learning network model. By combining the deep semantic reasoning of the knowledge graph with the continuous strategy optimization of reinforcement learning, the core performance of the intelligent customer service system is remarkably improved.

Inventors

ZHANG JUNMING
MIN WANLI
DING XIN
TIAN DIAN

Assignees

神思电子技术股份有限公司

Dates

Publication Date: 20260508
Application Date: 20251226

Claims (10)

1. The heat supply service customer service question answering method based on the knowledge graph is characterized by comprising the following steps of: Based on a historical heat supply service record, constructing a heat supply question-answer knowledge graph, and performing semantic embedding on the heat supply question-answer knowledge graph; analyzing a heat supply service problem proposed by a user, extracting a key problem entity, identifying a question intention category of the user, generating an enhanced problem semantic vector based on the key problem entity and the question intention category, and calculating the semantic matching degree of the enhanced problem semantic vector and each entity node in the heat supply question-answer knowledge graph; Determining an entity node corresponding to the highest semantic matching degree as a retrieval starting point, taking the retrieval starting point as an initial node, performing multi-hop traversal along entity relation edges in the heat supply question-answer knowledge graph, and collecting end nodes reached by a traversal path as candidate nodes to form a candidate node set; Calculating expected benefit values of candidate nodes in the candidate node set according to the questioning intention category through a reinforcement learning network model, screening target response nodes, generating a heat supply service response text based on questioning and answering semantic information corresponding to the target response nodes, and returning to a user terminal; and collecting interactive feedback data generated by the user aiming at the heat supply service response text, calculating a reward signal value based on the interactive feedback data through a reward function, and optimizing network parameters of the reinforcement learning network model according to the reward signal value.
2. The heat supply service customer service answering method based on the knowledge graph according to claim 1, wherein the history heat supply service record is based on, a heat supply questioning and answering knowledge graph is constructed, and semantic embedding is carried out on the heat supply questioning and answering knowledge graph, and the method specifically comprises the following steps: the method comprises the steps of obtaining a history heat supply service record of a heat supply service, wherein the history service record comprises a history work order text, a history customer service dialogue log and a heat supply policy document; analyzing the history heat supply service record, identifying and extracting a heat supply service object entity and a corresponding service logic relationship between the entities, and constructing a heat supply question-answer knowledge graph by taking the heat supply service object entity as an entity node and the service logic relationship as an entity relationship edge; and respectively vectorizing each entity node and each entity relation edge in the heat supply question-answering knowledge graph to generate a corresponding entity semantic embedding vector and a corresponding relation semantic embedding vector, and associating with the corresponding entity node and entity relation edge in the heat supply question-answering knowledge graph.
3. The method for providing heat supply service customer service answering according to claim 2, wherein the analyzing the heat supply service questions provided by the user, extracting key question entities, identifying the question intention category of the user, and generating enhanced question semantic vectors based on the key question entities and the question intention category, comprises: Collecting a heat supply service problem proposed by a user through a user terminal, carrying out semantic coding on the heat supply service problem, and generating a problem semantic context vector; identifying the question intention category of the user through an intention classification model based on the question semantic context vector, and outputting a corresponding question intention label; Extracting key problem entities representing heat supply service elements in the heat supply service problems through a named entity recognition algorithm; generating an enhanced question semantic vector based on the intent semantic representation of the question intent tag and the question semantic representation of the key question entity.
4. A heat supply service customer service answering method based on a knowledge graph according to claim 3, wherein the calculating the semantic matching degree between the enhanced problem semantic vector and each entity node in the heat supply answering knowledge graph specifically comprises: aiming at each entity node in the heat supply question-answering knowledge graph, acquiring a corresponding entity semantic embedding vector; and calculating the semantic similarity between the enhanced problem semantic vector and the entity semantic embedded vector to obtain the semantic matching degree corresponding to each entity node.
5. A method for providing a service for providing a heat supply service and answering a customer service based on a knowledge graph according to claim 3, wherein the calculating the expected benefit value of each candidate node in the candidate node set according to the question intention category by means of a reinforcement learning network model, and screening target answering nodes specifically comprise: generating an intention semantic vector based on the intention semantic representation of the question intention label, and acquiring candidate entity semantic embedded vectors corresponding to candidate nodes in the candidate node set; Fusing the intention semantic vector and the candidate entity semantic embedded vector for each candidate node to obtain a corresponding candidate feature vector; inputting the candidate feature vectors into a pre-trained reinforcement learning network model, and carrying out nonlinear transformation on the candidate feature vectors through a full connection layer of a strategy network in the reinforcement learning network model to obtain advanced feature representation of the candidate nodes; And calculating expected benefit values corresponding to the candidate nodes through an output layer of the strategy network based on the advanced feature representation, and taking the candidate node corresponding to the highest value in the expected benefit values as a target response node.
6. The method for providing heat supply service customer service answering according to claim 1, wherein the searching start point is used as an initial node, multi-hop traversal is performed along the entity relationship side in the heat supply customer service answering knowledge graph, and the end node reached by the traversal path is collected as a candidate node to form a candidate node set, which specifically comprises: Taking the searching starting point as an initial node, and accessing a next-hop node along an entity relationship side connected with the initial node in the heat supply question-answer knowledge graph; recording the current path depth of the access traversal path in real time, judging whether the current path depth reaches a preset maximum traversal depth threshold value or not, and starting traversal circulation; If the maximum traversal depth threshold is not reached, accessing a new next-hop node by taking the next-hop node as a current node along an entity relation edge connected with the current node, updating the current path depth of the accessed traversal path, and judging whether the updated current path depth reaches the maximum traversal depth threshold again; If the maximum traversal depth threshold is reached, determining that traversal of the currently accessed branch path is terminated, and collecting the currently accessed node as a candidate terminal node; and after all the reachable nodes are traversed, stopping the traversing cycle, and counting all candidate end nodes to form a candidate node set, wherein the reachable nodes are entity nodes which take the searching starting point as initial nodes and the current path depth does not reach the maximum traversing depth threshold in the heat supply question-answer knowledge graph.
7. The knowledge-graph-based heat supply service customer service answering method according to claim 1, wherein the mutual feedback data comprises user satisfaction score, number of questions and dialogue interruption mark; the calculating, by the reward function, a reward signal value based on the interactive feedback data specifically includes: Determining a penalty value corresponding to the session interrupt identifier according to a preset interrupt level mapping relation; And weighting the user satisfaction score, the challenge times and the penalty value according to preset reward function weights to generate a reward signal value.
8. The knowledge-graph-based heat supply service customer service answering method according to claim 7, wherein the optimizing the network parameters of the reinforcement learning network model according to the reward signal value specifically comprises: Generating original selection probability of the corresponding candidate node through a normalization function according to expected benefit values of candidate nodes in the candidate node set, and obtaining original selection probability distribution of the candidate node set; Calculating a strategy gradient of a strategy network in the reinforcement learning network model based on the reward signal values and the original selection probability distribution; And according to the strategy gradient, adjusting network parameters of the strategy network through a strategy optimization algorithm to increase expected benefit values of candidate nodes in similar states.
9. A knowledge graph-based heat supply service customer service answering apparatus, comprising: At least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform a knowledge-based heat service customer service answering method according to any one of claims 1 to 8.
10. A non-transitory computer storage medium storing computer executable instructions configured to perform a knowledge-based heat service customer service answering method according to any one of claims 1 to 8.

Description

Knowledge graph-based heat supply service customer service answering method, equipment and medium Technical Field The application relates to the technical field of artificial intelligence customer service, in particular to a heat supply service customer service question-answering method, equipment and medium based on a knowledge graph. Background In urban civil security public utility systems, the scale of heat supply service coverage users is enormous. With the deepening of digital transformation in the field of public utilities, the automation and specialized interaction of a heat supply service process are realized by using an artificial intelligence technology, and the heat supply service process gradually becomes the core direction of industry optimization service efficiency, so that an intelligent customer service system for heat supply service is induced. The intelligent customer service system for heat supply service is an important application of artificial intelligence technology in the field of public utilities, and aims to process various service requests of users about heat supply policy, fault repair, cost inquiry, business handling and the like through automatic response. However, the existing heat supply service customer service system is generally constructed based on a rule engine or a search model, and gradually introduces a knowledge graph technology to perform structural modeling on entities and relations thereof in the heat supply field, so that obvious defects exist when the complex, dynamic and proprietary service requirements are faced. On the one hand, the heat supply knowledge relates to a large amount of expertise, dynamic policies and complex equipment association relations, and the traditional knowledge graph is difficult to realize deep semantic reasoning and implicit demand understanding across multiple rounds of conversations, such as fault positioning with unclear user description or compound query related to historical cost and ladder policies. On the other hand, the existing system lacks the capability of self-optimizing according to continuous interactive feedback of users, and knowledge representation and response strategies cannot be adjusted in a self-adaptive mode, so that response accuracy is reduced, knowledge updating is delayed, and service flexibility is insufficient when seasonal report peaks, policy updating or novel faults are faced. Disclosure of Invention In order to solve the problems, the application provides a heat supply service customer service question answering method based on a knowledge graph, which comprises the following steps: Based on a historical heat supply service record, constructing a heat supply question-answer knowledge graph, and performing semantic embedding on the heat supply question-answer knowledge graph; analyzing a heat supply service problem proposed by a user, extracting a key problem entity, identifying a question intention category of the user, generating an enhanced problem semantic vector based on the key problem entity and the question intention category, and calculating the semantic matching degree of the enhanced problem semantic vector and each entity node in the heat supply question-answer knowledge graph; Determining an entity node corresponding to the highest semantic matching degree as a retrieval starting point, taking the retrieval starting point as an initial node, performing multi-hop traversal along entity relation edges in the heat supply question-answer knowledge graph, and collecting end nodes reached by a traversal path as candidate nodes to form a candidate node set; Calculating expected benefit values of candidate nodes in the candidate node set according to the questioning intention category through a reinforcement learning network model, screening target response nodes, generating a heat supply service response text based on questioning and answering semantic information corresponding to the target response nodes, and returning to a user terminal; and collecting interactive feedback data generated by the user aiming at the heat supply service response text, calculating a reward signal value based on the interactive feedback data through a reward function, and optimizing network parameters of the reinforcement learning network model according to the reward signal value. On the other hand, the application also provides a heat supply service customer service answering device based on the knowledge graph, which comprises: At least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform a knowledge-graph-based heat service customer service answering method as described in the above examples. In another aspect, the application also provides a non-volatile computer storage medium storing computer executable instructions configured for a heat supply service