CN-122001842-A - Intelligent customer service real-time auxiliary system and method based on contact type enterprise communication

CN122001842ACN 122001842 ACN122001842 ACN 122001842ACN-122001842-A

Abstract

The invention relates to an intelligent customer service real-time auxiliary system and method based on contact type enterprise communication, wherein the system comprises a message acquisition and desensitization module, a platform callback chat message generation module and a platform callback chat message generation module, wherein the message acquisition and desensitization module is used for receiving a platform callback chat message and generating a desensitization standard message event; the system comprises a conversation construction module, an intention recognition module, a knowledge retrieval module, a real-time pushing module, a work order generation and data analysis module and a customer service key index statistics module, wherein the conversation construction module is used for constructing a virtual conversation unit, the intention recognition module is used for outputting business intention, sub intention and entity information of a current conversation, the knowledge retrieval module is used for acquiring matching content from a knowledge base, the emotion analysis module is used for monitoring emotion types, intensity values and emotion trends of customer messages in real time, the real-time pushing module is used for pushing intention recognition results, the matching content and emotion early warning results to a side bar interface, and the work order generation and data analysis module is used for automatically generating a prefilled work order and counting the customer service key index. By constructing the virtual session unit, the method and the device can realize real-time structural processing, semantic understanding and auxiliary pushing of the continuous chat stream of the contact.

Inventors

DIAO KE
ZHAN RUI
LIU JIAXIN
XIANG LIGANG
SUN YANG
SI XIAOLEI
YU HUA
XIE ZICHAO
WANG GANYU
ZHAO FEIXIANG
ZHANG KEQIANG
CHEN JIAN

Assignees

柒贰零（北京）健康科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (10)

1. The utility model provides an intelligent customer service real-time auxiliary system based on contact type enterprise communication which characterized in that, the system is disposed in high in the clouds server to imbed customer service terminal through the side bar interface of enterprise instant messaging platform, include: the message acquisition and desensitization module is used for receiving chat messages recalled by the enterprise instant communication platform and decrypting the chat messages, and generating desensitization standard message events with anonymous identifiers; the conversation construction module is used for grouping the desensitization standard message events in the continuous chat stream of the contact person according to a preset rule to construct a virtual conversation unit, wherein the preset rule comprises time interval judgment of adjacent messages, intention transition judgment based on semantic deviation and artificial instruction judgment; the intention recognition module is used for carrying out semantic analysis on the multi-round dialogue context in the virtual session unit and outputting the business intention, the sub intention and the entity information of the current session; the knowledge retrieval module is used for acquiring matching content from a knowledge base through a retrieval enhancement generation technology based on the information output by the intention recognition module; The emotion analysis module is used for monitoring emotion types, intensity values and emotion trends of the client messages in real time to obtain emotion early warning results; The real-time pushing module is used for pushing the intention recognition result, the matching content and the emotion early warning result to the side bar interface in a long connection mode; and the work order generation and data analysis module is used for automatically generating a pre-filling work order according to the intention recognition result and counting customer service key indexes based on the virtual session unit and the customer service behavior log.
2. The intelligent customer service real-time assistance system based on contact-type business communication of claim 1, wherein the message collection and desensitization module comprises an automatic desensitization unit configured to identify private data in messages using named entity identification techniques; the intent recognition module is configured to construct a contextual window containing the last N rounds of conversations, perform an reference resolution process using a self-attention mechanism, and output top-level business domain intents and specific business sub-intents using a hierarchical classification architecture.
3. The intelligent customer service real-time assistance system based on contact-type enterprise communication according to claim 1, wherein the session construction module determines whether the business phase is shifted by calculating the cosine similarity S of the current message vector V t and the semantic gravity vector V context of the current virtual session when performing the intention transition determination, and the calculation formula is: ; When S is smaller than a preset semantic offset threshold When the system determines that an intended transition has occurred and opens a new virtual session element.
4. The intelligent customer service real-time assistance system based on contact-type enterprise communication according to claim 1, wherein the knowledge retrieval module comprises a reordering unit and adopts a two-way concurrent retrieval architecture, the two-way concurrent retrieval architecture comprises: The accurate matching circuit is used for carrying out hard matching on the inverted index of the knowledge base based on the keywords; a semantic vector way for retrieving semantic neighbor segments in a vector database based on vector similarity; And the reordering unit is used for carrying out secondary scoring on the two-way retrieval result by combining the context background of the current virtual session, and selecting the knowledge content with the highest matching degree score to generate the recommendation card.
5. The intelligent customer service real-time assistance system based on contact-type enterprise communication of claim 4, wherein the real-time pushing module is further configured to record a reference, edit or ignore a behavior of a customer service person to a recommended card, and asynchronously send the behavior data as a feedback sample to a model training unit for fine tuning a model of the intention recognition module and the reordering unit.
6. The intelligent customer service real-time assistance system based on contact type enterprise communication according to claim 1, wherein the emotion analysis module continuously outputs emotion intensity trend, and when the emotion weighted score in the sliding window exceeds a preset risk threshold, the real-time pushing module automatically triggers risk early warning at a side bar and places a top pushing pacifying or upgrading processing instruction.
7. The intelligent customer service real-time assistance system based on contact-type enterprise communication of claim 1, further comprising: And the degradation processing module is used for automatically switching the sidebar interface to a degradation mode when the long connection interruption with the cloud server is detected, starting the keyword matching logic and the manual search entrance of the local cache, and automatically synchronizing the message record during the interruption after the connection is recovered so as to supplement the virtual session unit.
8. An intelligent customer service real-time assistance method based on contact enterprise communication, which is characterized by being applied to the system of any one of claims 1 to 7, and comprising the following steps: The method comprises the steps of S1, message desensitization acquisition, namely receiving chat messages and carrying out mask processing on privacy data by using a named entity recognition technology to generate standard message events; S2, constructing a virtual session, namely grouping standard message events according to adjacent message time intervals, semantic offset and manual statement instructions, and constructing a virtual session unit; s3, multi-round semantic analysis, namely carrying out reference resolution by combining session context, and identifying current business intention, sub-intention and entity information; s4, performing double-path retrieval of keywords and vectors based on intention and entity information, and reordering results to obtain matching knowledge; s5, emotion analysis, namely monitoring emotion types, intensity values and emotion trends of client messages in real time to obtain emotion early warning results; s6, pushing auxiliary content, namely pushing service intention, matching knowledge and emotion early warning results to a customer service terminal in real time through a sidebar; and S7, service linkage and feedback, namely automatically generating a pre-filling work order when a specific intention is identified, and collecting the quotation feedback of customer service on push content in real time to drive model evolution.
9. The intelligent customer service real-time assistance method based on contact type enterprise communication according to claim 8, wherein the step S2 comprises: S201, receiving a current standard message event m t ; S202, calculating adjacent message time interval If (if) If T max is the preset threshold, the step S206 is skipped; S203, extracting the message vector V t , calculating the cosine similarity S of the message vector V context and the conversation center of gravity vector if , If the threshold value is the semantic offset threshold value, determining that the intended transition occurs; s204, monitoring whether the work instruction or the service triggering action of the junction man is included; S205, if the judgment is no, associating m t with the current active session, and dynamically and smoothly updating V context ; if any one of the determinations is yes, initializing a new virtual session ID and a statistics timer.
10. The intelligent customer service real-time auxiliary method based on contact type enterprise communication according to claim 8, wherein the auxiliary content pushed in the step S6 is presented in the form of a structured card, and the auxiliary content is supported to be edited and sent by a customer service person by referring to a primary input box in a one-key manner.

Description

Intelligent customer service real-time auxiliary system and method based on contact type enterprise communication Technical Field The invention belongs to the technical field of enterprise-level intelligent customer service, and particularly relates to an intelligent customer service real-time auxiliary system and method based on contact type enterprise communication. Background Along with the popularization of enterprise instant messaging platforms, a customer service mode is gradually shifted to a contact communication mode represented by enterprise WeChat, a long-term friend relationship is established between customer service and clients, and the two parties communicate to form a continuous chat stream across time and scenes. The existing system constructed based on the conversational model has obvious technical defects when applied to the scene: 1. The data has no boundary, the system can not automatically identify the service stage, and key indexes such as first response time length and the like are difficult to count. 2. The operation path is long, customer service needs to frequently switch between the chat window and the knowledge base and work order system, and the efficiency is low. 3. And the perception is lacking, namely the semantics cannot be understood in real time, the intention is recognized, the emotion change is perceived, and an early warning mechanism is lacked. 4. Knowledge precipitation is difficult, retrieval is highly dependent on manual work, and recovery quality is limited by personal ability. Disclosure of Invention The invention aims to provide an intelligent customer service real-time auxiliary system and method for carrying out real-time acquisition, semantic analysis, intention recognition, knowledge retrieval, emotion analysis and service linkage on chat contents based on a contact continuous dialogue scene of an enterprise instant messaging platform. The invention provides an intelligent customer service real-time auxiliary system based on contact type enterprise communication, which is deployed on a cloud server and is embedded into a customer service terminal through a side bar interface of an enterprise instant communication platform, and comprises the following components: the message acquisition and desensitization module is used for receiving chat messages recalled by the enterprise instant communication platform and decrypting the chat messages, and generating desensitization standard message events with anonymous identifiers; the conversation construction module is used for grouping the desensitization standard message events in the continuous chat stream of the contact person according to a preset rule to construct a virtual conversation unit, wherein the preset rule comprises time interval judgment of adjacent messages, intention transition judgment based on semantic deviation and artificial instruction judgment; the intention recognition module is used for carrying out semantic analysis on the multi-round dialogue context in the virtual session unit and outputting the business intention, the sub intention and the entity information of the current session; the knowledge retrieval module is used for acquiring matching content from a knowledge base through a retrieval enhancement generation technology based on the information output by the intention recognition module; The emotion analysis module is used for monitoring emotion types, intensity values and emotion trends of the client messages in real time to obtain emotion early warning results; The real-time pushing module is used for pushing the intention recognition result, the matching content and the emotion early warning result to the side bar interface in a long connection mode; and the work order generation and data analysis module is used for automatically generating a pre-filling work order according to the intention recognition result and counting customer service key indexes based on the virtual session unit and the customer service behavior log. Further, the message collection and desensitization module comprises an automatic desensitization unit configured to identify private data in a message using named entity identification techniques; the intent recognition module is configured to construct a contextual window containing the last N rounds of conversations, perform an reference resolution process using a self-attention mechanism, and output top-level business domain intents and specific business sub-intents using a hierarchical classification architecture. Further, when the session construction module performs the intention transition determination, it determines whether the business stage is shifted by calculating the cosine similarity S between the current message vector V t and the semantic gravity vector V context of the current virtual session, where the calculation formula is as follows: ; When S is smaller than a preset semantic offset threshold When the system determines that an intended tr