CN-116860934-B - Multi-scene multi-round dialogue method and system for virtual customer service

CN116860934BCN 116860934 BCN116860934 BCN 116860934BCN-116860934-B

Abstract

The invention relates to a multi-scene multi-round dialogue method and a system for virtual customer service. The dialogue scheduling component is used for managing the operation sequence and the calling relation of each scene dialogue component, the business handling component is responsible for processing dialogues related to business handling, the knowledge base query component is responsible for processing general knowledge retrieval dialogues of non-business handling types, and the chat component is responsible for processing general dialogues of non-business handling and non-knowledge base query types. The multi-scene multi-round dialogue method for virtual customer service provided by the invention can enable the virtual customer service robot to flexibly switch between different dialogue scenes, and the switching is automatically perceived by dialogue contents during each round of natural dialogue without external intervention, thereby solving the problems of dialogue splitting and the like and further promoting the development of virtual digital people in personification, interaction capability and the like.

Inventors

LUO MINGYU
HUANG LIN
LIN JIAN

Assignees

东云睿连(武汉)计算技术有限公司

Dates

Publication Date: 20260512
Application Date: 20230703

Claims (9)

1. A multi-scenario, multi-round dialogue system for virtual customer service, wherein the multi-scenario comprises a business handling scenario, a knowledge base query scenario and a chatting scenario, and the multi-scenario is characterized by comprising a dialogue scheduling component, a business handling component, a knowledge base query component and a chatting component; The dialogue scheduling component is used for managing the operation sequence and the calling relation of each scene dialogue component and providing an interface for interaction with the outside, and in each round of dialogue, receiving the text or the voice input from the outside, returning the response text or the voice after processing and waiting for the next input; the business handling component is used for processing the dialogue related to business handling, maintaining all business rule handling lines and recording the current business handling progress; The knowledge base query component is used for processing general knowledge retrieval type dialogue of non-business handling type, maintaining a knowledge base and managing a context recorder for tracking query progress; The chat component is used for processing chat conversations which do not belong to business handling and do not belong to content retrieval at the same time, and maintaining the history conversation content of the current multi-round chat; the business handling component comprises a business handling line, a progress recorder, a business handling inquiry interface and a progress resetting interface; The business handling line comprises a linear data structure formed by nodes in sequence in the business handling steps, supports single or multiple business handling lines, wherein each node at least comprises a preset corpus entering the node, information pointing to the next node and a processing method for user reply; The progress recorder is a variable for recording the current node position, comprises a current line and the node position of the line, points to an initial node under the initial condition, does not comprise preset corpus and next node information, takes the first node of each service handling line as a skip node, and takes the service description of each service handling line as a trigger text; the service handling inquiry interface is used for executing single-round service handling dialogue, transmitting user input to the node pointed by the current progress recorder, returning a response text by the node and pointing the progress recorder to a new node according to the node rule; and the progress resetting interface is used for resetting the progress recorder and pointing the progress recorder to the initial node.
2. The multi-scenario, multi-round dialog system for virtual customer service of claim 1, wherein the dialog scheduling component comprises: the information receiving module is used for receiving query text or query voice input by a user; the conversion module is used for analyzing the input query voice into a query text by adopting a voice-to-text algorithm; The dispatching response module is used for taking the query text as input, sequentially calling the service handling component, the knowledge base query component and the chat component to match the query text, obtaining a corresponding response text as an output text, and confirming the dialogue intention; and the output module is used for returning the output text to the user.
3. The multi-scenario, multi-round dialog system for virtual customer service according to claim 1, further comprising a special jump node in the business transaction line, wherein the special jump node contains a special trigger text and a jump node, a response text or a log-out information after the trigger.
4. A multi-scenario, multi-round dialog system for virtual customer service according to claim 3, wherein the node checks the reply information of the user and decides the node jump information, comprising: after receiving the reply message of the user, the node judges whether the node contains a special trigger text, if so, the reply message of the user is matched with the special trigger text of the node, if the matching is unsuccessful or the node does not contain the special trigger text, the node enters the next node, and if the matching is successful, the node jumps to the triggered jump node contained in the node.
5. The multi-scenario, multi-round dialog system for virtual customer service of claim 1, wherein the knowledge base query component comprises a knowledge base module, a search context module, a knowledge base query interface, and a context reset interface; The knowledge base module comprises one or more directed acyclic tree structures, a root node, a leaf node and a knowledge content module, wherein the directed acyclic tree structures comprise knowledge labels and knowledge content; The search context module comprises a sub-tag list of the current search node and position indexes of all sub-tags, and is empty when the search node is in a leaf node at the beginning or when the search node is in a leaf node; the knowledge base query interface is used for executing a single round of knowledge base query dialogue, and directly searching any knowledge label of the non-leaf node during query; the context reset interface is configured to reset the search context module to null.
6. The multi-scenario, multi-round dialog system for virtual customer service according to claim 5, wherein the knowledge base query interface, when executing a single round of knowledge base query dialog, performs the following operations: Acquiring a query text of a user, and judging whether a search context module is empty or not; If the highest similarity is larger than a first preset threshold value, returning the lower label or knowledge content of the sub label of the highest similarity to update the search context module, otherwise, returning a 'matching failure' signal; If the search context module is empty, calculating the similarity between the query text and all knowledge labels, if the highest similarity is larger than a second preset threshold, returning the lower label or knowledge content of the knowledge label corresponding to the highest similarity to update the search context module, otherwise, returning a 'search failure' signal.
7. The multi-scenario, multi-round dialog system for virtual customer service of claim 5, wherein the chat component comprises a multi-round dialog chat model, a history chat log module, a chat log query interface, and a chat log reset interface; the multi-turn dialogue chatting model is a generated Seq2Seq neural network model, takes a history chatting record and a query text of a current turn as inputs, and outputs a chatting response text of the current turn; The history chatting recording module is used for recording a query text and a response text in a section of history round number, the history round number is limited or unlimited, and the history chatting recording module is empty in the initial condition; The chat record inquiry interface is used for executing single-round chat conversations, and the chat method comprises the steps of splicing a history chat record and inquiry text, forwarding the history chat record through a multi-round conversation chat model, and returning to the model for outputting; the chat record resetting interface is used for resetting the history chat record module to be empty.
8. The multi-scenario, multi-round dialog system for virtual customer service according to claim 7, wherein the dispatch response module in the dialog dispatch component is specifically configured to: the query text is used as input, a business handling query interface in the business handling component is called to carry out text matching, a first response text is obtained according to a matching result, if the first response text is not a 'matching failure' signal, the intention of a dialogue is confirmed to be business handling, the first response text is used as output text, a context reset interface of a knowledge base query component and a chat record reset interface of a chat component are called, otherwise, a knowledge base query interface in the knowledge base query component is called to carry out text matching; The knowledge base query component obtains a second response text according to a matching result, if the second response text is not 'matching failure' or 'searching failure', the dialogue intention is confirmed to be the knowledge base query, the second response text is taken as an output text, a progress reset interface of the business handling component and a chat record reset interface of the chat component are called, if the second response text is a 'matching failure' signal, a context reset interface of the knowledge base query component is called to reset the historical state of the knowledge base query, the business handling query interface in the business handling component is called to carry out text matching, if the second response text is 'searching failure', the dialogue intention is confirmed to be a general chat, the chat record query interface in the chat component is called, a third response text is obtained as the output text, and the progress reset interface of the business handling component and the context reset interface of the knowledge base query component are called.
9. A multi-scene multi-round dialogue method for virtual customer service, which is implemented based on a multi-scene multi-round dialogue system for virtual customer service according to any one of claims 1-8, and is characterized by comprising the following steps: step1, receiving query text or query voice input by a user, if the input is voice, executing step 2, otherwise, executing step 3; step 2, adopting a voice-to-text algorithm to analyze the query voice into a query text; step 3, taking the query text as input, calling a query interface of the business handling component to obtain a first response text, if the first response text is not a 'failure' signal, confirming that the round of dialogue is intended to be business handling, temporarily storing the first response text as output text, calling a context reset interface of the knowledge base query component, calling a chat record reset interface of the chat component, executing step 6, otherwise executing step 4; Step 4, taking the query text as input, calling a query interface of a knowledge base query component to obtain a second response text, if the second response text is not a 'matching failure' or 'search failure' signal, confirming that the round of dialogue intention is knowledge base query, taking the second response text as output text temporary storage, calling a progress reset interface of a business handling component, calling a chat record reset interface of a chat component, executing step 6, if the second response text is a 'matching failure' signal, namely, indicating that the previous round of dialogue intention is knowledge base query, but the current round of query progress is not matched with the previous round of dialogue intention, calling a context reset interface of the knowledge base query component to reset the history state of knowledge base query, then executing step 3, and if the second response text is 'search failure', namely, indicating that the query text is not matched with the knowledge base item, confirming that the round of dialogue intention is not knowledge base query, and executing step 5; Step 5, confirming that the round of dialogue intention is general chat, taking a query text as input, calling a query interface of a chat component to obtain a third response text as temporary storage of an output text, calling a progress reset interface of a business handling component, calling a context reset interface of a knowledge base query component, and executing step 6; And 6, if the input of the user is text, returning the temporarily stored output text to the user, otherwise, synthesizing the output text into audio by adopting a text-to-speech algorithm and returning the audio to the user.

Description

Multi-scene multi-round dialogue method and system for virtual customer service Technical Field The invention relates to the technical field of natural language processing, in particular to a multi-scene multi-round dialogue method and system for virtual customer service. Background Virtual customer service is the application of virtual digital people in an intelligent customer service scene. The traditional customer service generally only needs to have the service inquiry handling capability, and the virtual customer service often bears the tasks of natural dialogue, boring, question and answer and the like due to the characteristics of vivid image, more interaction capability and the like. This requires not only that the virtual customer service has more general dialogue capabilities than business rules, but also that it can judge the current dialogue scene according to the user input sentence and context, and can naturally switch between scenes. Some work has been proposed to implement switching of multi-scene, multi-round conversations. Xu Min et al in the patent No. 201811626785.8, a session scene switching method and apparatus are disclosed, which are a method for determining whether to stay in a first scene or switch to a second scene and whether to restore to the first scene by the similarity of two adjacent rounds of user input messages. Li Chunlan et al in the patent No. 201911166714.9, "a multi-round dialogue system construction method and system based on a service scenario" propose a multi-round dialogue method for multi-scenario service, preset keywords and predictions for different service scenarios, and use an intent judgment model and a rule matching model to perform scenario matching and conversion. Wang Zihao et al in patent number 202010809597.X, the invention of "method and apparatus for intelligent customer service multi-round session management" propose a multi-round intent model that allows the system to perform different business logic according to potential user intent categories by identifying user intent in multi-round communications with the user. The problem with the above methods is that they are designed to switch only between two scenes or multiple scenes of similar function and they mostly only give "intention" or "signal" of scene switch, requiring specific dialog logic to be implemented by other methods. The methods are more suitable for the traditional customer service robots, and the service types required to be transacted by the user are identified through two or more rounds of conversations and then are transferred to a system component to implement the service flow. For the application of the virtual digital man technology in the intelligent customer service scene, the virtual man needs to have complete dialogue logic similar to a real man, and can naturally and flexibly switch between different dialogue scenes, which is not possessed by the existing method. Disclosure of Invention Aiming at the technical problems existing in the prior art, the invention provides a multi-scene multi-round dialogue method and a system for virtual customer service, provides multi-round dialogue capability for three different scenes of business handling, knowledge base inquiry and boring, and intelligently identifies user semantics and switches among contexts of different scenes and completes dialogue through a general scene dialogue scheduling and method and a dialogue method of three independent scenes. The technical scheme for solving the technical problems is as follows: in one aspect, the invention provides a multi-scenario multi-round dialog system for virtual customer service, wherein the multi-scenario comprises a business-oriented transaction scenario, a knowledge base query scenario and a boring scenario. The service handling scene refers to a process that a user selects among various preset service handling lines according to the guidance to advance, retreat, jump or exit among the service handling steps. The knowledge base inquiry scene refers to a process of searching the corresponding content of a specific item in a database containing knowledge items and knowledge content by a user. The chatting scene refers to a conversation flow of non-business inquiry and non-specific content retrieval of the user without purposeful chatting. The dialogue system comprises a dialogue scheduling component, a business handling component, a knowledge base inquiring component and a chat component; The dialogue scheduling component is used for managing the operation sequence and the calling relation of each scene dialogue component and providing an interface for interaction with the outside, and in each round of dialogue, receiving the text or the voice input from the outside, returning the response text or the voice after processing and waiting for the next input; the business handling component is used for processing the dialogue related to business handling, maintaining all business rule h