CN-122019696-A - Intelligent customer service robot question answering method and system based on charging knowledge graph

CN122019696ACN 122019696 ACN122019696 ACN 122019696ACN-122019696-A

Abstract

The invention provides an intelligent customer service robot question-answering method and system based on a charging knowledge graph, and the method and system comprise the steps of carrying out intention recognition on a charging session query to obtain the query intention, carrying out entity link in the charging knowledge graph based on the query intention to obtain an entity linked with the query intention, carrying out multi-jump search in the charging knowledge graph based on the linked entity and the query intention to obtain nodes and associated rule terms consistent with the constraint of intention slots, generating an evidence chain based on the obtained nodes and associated rule terms, driving a large language model to generate natural language answers based on the evidence chain and predefined decision logic, and combining the natural language answers and rule term document information corresponding to the natural language answers to generate charging session answers. And according to the query intention, carrying out multi-hop search by combining the linked entities in the charging knowledge graph to generate answers, so that cross-topic interference can not occur, and the accuracy of charging session answers is improved.

Inventors

YAN ZHE
SU SHU
ZHAO YU
XUE RUI
ZHENG PIAOPIAO
YANG LI
Lin Yinxi
Shi shuanglong
Fan Zhenpeng
WANG JIE
DU BAOXING
LI YAO

Assignees

国网智慧车联网技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251209

Claims (10)

1. An intelligent customer service robot question-answering method based on a charging knowledge graph is characterized by comprising the following steps: Receiving a charging session query input by a user by adopting an intelligent customer service robot, and carrying out intention recognition on the charging session query to obtain a query intention; Based on the query intention, entity linking is carried out in a pre-constructed charging knowledge graph, and an entity to which the query intention is linked is obtained; performing a multi-hop search in the charging knowledge graph based on the linked entity and the query intent to obtain nodes and associated rule terms consistent with the intent slot constraints; and combining the natural language answer with rule clause document information corresponding to the natural language answer to generate a charging session answer, and feeding back the charging session answer to the user through the intelligent customer service robot.
2. The method according to claim 1, wherein the construction process of the charging knowledge graph comprises: obtaining entities, relations, attributes of the entities and rule clauses from a multi-source data document of FAQ, service rules, charging rules, site information and order information of charging service through an information extraction technology; Defining the entity as a node of a knowledge graph, defining the relationship as an edge of the knowledge graph, defining the attribute as a characteristic of the node, and performing association binding on the rule clause and the corresponding node; and determining the version of the rule clause, and adding a corresponding version identifier for the rule clause to form a charging knowledge graph.
3. The method of claim 2, wherein the entity comprises a site, a charging post, a connector, a charge, an order, a payment, an invoice, a member, a service package, a channel and a rule, wherein the relationship comprises at least one of a site-owned-device relationship formed according to an owned relationship between the site and the charging post, a charging-applicable-site/order relationship formed according to an applicable relationship between the charge rule and the site or the order, a member-constraint-rule relationship formed according to a constraint relationship between the member and the rule, an invoice-for-order relationship formed according to a use relationship between the invoice and the order, a FAQ-answer-rule or flow relationship formed according to an answer relationship between the FAQ and the rule or flow, and a channel-support-payment relationship formed according to a support relationship between the channel and the payment.
4. The method of claim 1, wherein the performing a multi-hop search in the charging knowledge graph based on the linked entity and the query intent obtains nodes and associated rule terms consistent with the intent slot constraint, generating an evidence chain based on the obtained nodes and associated rule terms, comprises performing a multi-hop search in a corresponding subgraph of the charging knowledge graph with the linked entity as a starting point by using a graph traversal algorithm to obtain nodes consistent with the intent slot constraint, extracting associated rule terms from the obtained nodes, vectorizing text content of the associated rule terms, performing semantic search, screening candidate rule term fragments similar to the query intent semantics, integrating rule term version information associated with the corresponding multi-source data document based on the obtained nodes, the candidate rule term fragments, and generating an evidence chain.
5. The method of claim 4, wherein the performing a multi-hop search in the corresponding subgraph of the charging knowledge-graph using a graph traversal algorithm with the linked entity as a starting point, obtaining nodes consistent with the intended slot constraints, comprises: And in the searching process of each jump, only traversing the edge consistent with the intended slot constraint according to the intended slot constraint in the query intention, and selecting the node consistent with the intended slot constraint.
6. The method of claim 4, wherein the integrating based on the obtained nodes, the candidate rule term segments, associated rule term version information, and corresponding multi-source data documents generates a chain of evidence, comprising: The method comprises the steps of obtaining a candidate rule clause fragment, obtaining rule clause version information of the candidate rule clause fragment, obtaining a corresponding multi-source data document, packaging the obtained node, the candidate rule clause fragment, the associated rule clause version information and the corresponding multi-source data document to form an initial evidence set, carrying out consistency check on the initial evidence set based on the charging session query, eliminating evidence fragments which conflict with session constraint or are out of date, and reordering the evidence fragments reserved after the check according to the correlation with the query intention to form a final evidence chain.
7. The method of any of claims 1-6, wherein the generating of the charging session answer comprises at least one of: When the ambiguity of the charging session query due to the lack of the key intention slot constraint is identified based on the evidence chain, generating a charging session answer containing common situations based on the natural language answer and the set default rule terms corresponding to the natural language answer, and embedding a clear question for the key intention slot constraint in the charging session answer, and simultaneously embedding a multi-source data document corresponding to the charging session answer; when the evidence link shows that a plurality of rule clauses with effective versions exist, based on the natural language answer, the latest effective rule clause version corresponding to the natural language answer is preferentially selected to generate a charging session answer, the version of the rule clause according to the charging session answer is marked, and meanwhile, a multi-source data document corresponding to the charging session answer is embedded; The intention slot constraint and the key intention slot constraint comprise at least one of a channel used by a user to initiate a charging session query, a region where the user is located or time information related to the query.
8. The method according to claim 1, further comprising, after the charge knowledge-graph construction: Updating and conflict detection are carried out on rule clauses in the charging knowledge graph through an increment updating pipeline; In response to a successful update, incrementing a version number of the rule term document and synchronously updating a search index associated with the charging knowledge graph; Constructing a synonymous dictionary covering common term variants in the entity and rule term document information in the charging knowledge graph; Applying the synonymous dictionary to the entity link, the charging knowledge graph retrieval and the charging session answer generation process; and when detecting that the charging knowledge graph is updated in error, executing version rollback operation, and recovering the charging knowledge graph to a state before updating.
9. The method of claim 1, wherein the step of determining the position of the substrate comprises, The node comprises at least one of a site, a charging pile, an order, a payment tool, an invoice, a member and a service package; the rule clauses comprise at least one of a charging rule, a membership benefit rule, a refund rule and an invoicing rule; The multi-source data file comprises a common problem solving file of charging service, a charging service operation procedure and flow file, a charging rule and price policy file, equipment information and state files of a charging station and a charging pile, and a user order and transaction record file.
10. An intelligent customer service robot question-answering system based on a charging knowledge graph is characterized by comprising: The system application layer is used for receiving a charging session query input by a user by adopting the intelligent customer service robot, and carrying out intention recognition on the charging session query to obtain a query intention; The functional component layer is used for carrying out entity link in a pre-constructed charging knowledge graph based on the query intention to obtain an entity to which the query intention is linked; performing a multi-hop search in the charging knowledge graph based on the linked entity and the query intent to obtain nodes and associated rule terms consistent with the intent slot constraints; The basic support layer is used for driving the large language model to generate a natural language answer based on the evidence chain and the predefined decision logic, combining the natural language answer with rule clause document information corresponding to the natural language answer to generate a charging session answer, and feeding back the charging session answer to the user through the intelligent customer service robot.

Description

Intelligent customer service robot question answering method and system based on charging knowledge graph Technical Field The invention belongs to the technical field of charging knowledge graph construction, and particularly relates to an intelligent customer service robot question-answering method and system based on a charging knowledge graph. Background With the popularization of new energy automobiles and public charging services, the scale of platform users exceeds five tens of millions, and online customer service demands continue to increase, so that a new energy charging question-answering system is particularly important in practical application. When the charging question-answering system in the prior art lacks the constraint of the domain structure, the retrieval stage is guided by the similarity, cross-subject interference and recall drift are easy to occur, and the generation stage can be rationalized to make a conclusion which looks smooth but is not based under the condition of insufficient evidence. Disclosure of Invention In order to overcome the defects in the prior art, in a first aspect, the present application provides an intelligent customer service robot question-answering method based on a charging knowledge graph, which includes: Receiving a charging session query input by a user by adopting an intelligent customer service robot, and carrying out intention recognition on the charging session query to obtain a query intention; Based on the query intention, entity linking is carried out in a pre-constructed charging knowledge graph, and an entity to which the query intention is linked is obtained; performing a multi-hop search in the charging knowledge graph based on the linked entity and the query intent to obtain nodes and associated rule terms consistent with the intent slot constraints; and combining the natural language answer with rule clause document information corresponding to the natural language answer to generate a charging session answer, and feeding back the charging session answer to the user through the intelligent customer service robot. Preferably, the construction process of the charging knowledge graph includes: obtaining entities, relations, attributes of the entities and rule clauses from a multi-source data document of FAQ, service rules, charging rules, site information and order information of charging service through an information extraction technology; Defining the entity as a node of a knowledge graph, defining the relationship as an edge of the knowledge graph, defining the attribute as a characteristic of the node, and performing association binding on the rule clause and the corresponding node; and determining the version of the rule clause, and adding a corresponding version identifier for the rule clause to form a charging knowledge graph. Preferably, the entity comprises a site, a charging pile, a connector, charging, an order, payment, an invoice, a member, a service package, a channel and a rule, and the relation comprises at least one of the following: forming a site-owned-equipment relationship according to the owned relationship between the site and the charging pile; according to the applicable relation between the charging rule and the site or the order, a formed relation of charging-applicable-site/order is formed; forming a relationship of 'member-constraint-rule' according to the constraint relationship between the member and the rule; Forming an invoice-order-used relationship according to the use relationship between the invoice and the order; forming a relationship of 'FAQ-answer-rule or process' according to the answer relationship between FAQ and rule or process; And forming a channel-support-payment relation according to the support relation between the channel and the payment. Preferably, the method comprises the steps of executing multi-hop search in the charging knowledge graph based on the linked entity and the query intention to obtain nodes and associated rule terms consistent with the query intention slot constraint, generating an evidence chain based on the obtained nodes and associated rule terms, executing multi-hop search in the corresponding subgraph of the charging knowledge graph by adopting a graph traversal algorithm with the linked entity as a starting point to obtain the nodes consistent with the query intention slot constraint, extracting the associated rule terms from the obtained nodes, vectorizing text content of the associated rule terms, executing semantic search, screening out candidate rule term fragments similar to the query intention semantics, and integrating based on the obtained nodes, the candidate rule term fragments, associated rule term version information and corresponding multi-source data documents to generate the evidence chain. Preferably, the performing multi-hop search in the corresponding subgraph of the charging knowledge graph by using the linked entity as a starting point and adopting a g