CN-121996765-A - Speaking recommendation method, device, equipment and medium based on context and intention prediction

CN121996765ACN 121996765 ACN121996765 ACN 121996765ACN-121996765-A

Abstract

The application provides a context and intention prediction-based speaking operation recommendation method, which is characterized in that by introducing intention recognition and a logic map-based potential subsequent intention prediction mechanism, and fusing current input information, business knowledge context and multiple rounds of dialogue context corresponding to multiple rounds of dialogue history data, the method responds to the current intention of a user and generates a more accurate target speaking operation. And then predicting the intention evolution of the potential follow-up intention of the current user based on logic patterns, so as to predict the potential requirement of the user, and generating a target guided call for the potential requirement, thereby improving the pertinence of the generated call in a complex dialogue scene by combining the recommended mode of the reply call and the guided call, improving the accuracy of the generated call, and improving the user experience and the message reply efficiency. The method can be applied to the seat conversation recommendation function in the financial field or the medical field, so that the seat conversation generation accuracy is improved.

Inventors

LI CHANGXIAN
QU CHENG
LEI BING

Assignees

平安健康互联网股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. A method of speaking recommendation based on context and intent prediction, the method comprising the steps of: Acquiring current input information of a current user and multi-round dialogue historical data of the current user and an agent, wherein the multi-round dialogue historical data comprises historical input information and agent reply information corresponding to the historical input information; Performing intention recognition on the current input information to obtain a current intention recognition result, and generating a target answer phone corresponding to the current input information and the multi-round dialogue history data based on a phone generation model; Predicting and obtaining potential follow-up intentions of the current user based on a preset logic map and the current intention recognition result, and generating target guide dialogs corresponding to the multiple rounds of dialogue history data, the business knowledge context and the potential follow-up intentions based on the dialog generation model; and based on the target answer phone operation and/or the target guide phone operation, obtaining the recommended phone operation of the current user.
2. The speaking recommendation method as claimed in claim 1, wherein before generating the preset business knowledge context, the current input information, the current intention recognition result, and the target reply speaking corresponding to the multi-round dialogue history data based on the speaking generation model, the method further comprises: carrying out semantic validity check on the current input information to obtain a first check result, and determining the emotion intensity and the problem type of the current input information based on the first check result, wherein the first check result comprises an integrity score, an emotion polarity label and a query word detection result; generating a message feature vector of the current user based on the first checking result, the current intention recognition result and the current input information; Acquiring the current customer service system load, and generating a general large model, a domain fine tuning model and a small parameter model based on the routing decision model, wherein the probability distribution of selection is under the current customer service system load and the message feature vector; when the current client system load is lower than a preset load threshold, taking the model with the highest probability of the selected probability distribution as the speaking operation generation model; and when the current client system load is not lower than a preset load threshold, using the small parameter model as the speech generating model.
3. The speaking recommendation method of claim 2, wherein the generating a message feature vector for the current user based on the current intent recognition result and the current input information comprises: extracting message length characteristics and key entity quantity from the current input information, and analyzing the intention complexity and intention category of the current user from the current intention recognition result; and generating the message feature vector based on the emotion strength, the question type, the message length feature, the number of key entities, the intention complexity and the intention category.
4. The speaking recommendation method as claimed in claim 2, wherein said performing intention recognition on the current input information to obtain a current intention recognition result comprises: Performing clause processing on the current input information to obtain each clause, performing semantic role recognition on each clause to obtain a semantic role recognition result of each vocabulary in each clause, and calculating the semantic relevance between each clause and the query word detection result based on the semantic role recognition result; Calculating the attention weight of each clause based on an attention mechanism, the multi-round dialogue history data and the semantic association degree of each clause, and generating a key text of the current input information based on the clause of which the attention weight exceeds a preset weight threshold; And carrying out intention classification on the key text to obtain a plurality of candidate intentions and probability distributions corresponding to the candidate intentions, and generating the current intention recognition result based on the emotion polarity labels, the question types and the candidate intentions and probability distributions corresponding to the candidate intentions.
5. The speaking recommendation method as claimed in claim 1, wherein predicting a potential follow-up intention of the current user based on a preset logic map and the current intention recognition result comprises: Based on the current intention recognition result, carrying out matching query in the intention node index of the logic atlas to obtain a current intention node; Traversing all the outgoing edges from the current intention node through the query of a graph database, and acquiring the follow-up intention node pointed by each outgoing edge and the transition probability weight attached to the edge; and screening out the node with the highest transition probability from the follow-up intention nodes based on the user portrait and the historical behavior data of the current user, and taking the node as the potential follow-up intention.
6. The speaking recommendation method of claim 1, wherein the generating the multiple rounds of dialogue history data, the business knowledge context, and the target guided speaking corresponding to the potential follow-up intention based on the speaking generation model comprises: Performing data fusion splicing on the multi-round dialogue historical data and the business knowledge context to generate a dialogue state representation vector; Encoding the potential subsequent intention as an intention embedding vector, and performing potential intention fusion decoding on the intention embedding vector and the dialogue state representation vector to generate a context representation of target guidance; Generating a candidate conversation text sequence corresponding to the context representation of the target guide in an autoregressive manner through a decoder based on the conversation generation model; And carrying out fluency check and context consistency check on the candidate phone text sequence, and determining the target guided phone in the candidate phone text sequence based on a check result and a consistency check result.
7. The speaking recommendation method as claimed in any one of claims 1 to 6, wherein generating a target answer speaking corresponding to the current input information, and the multi-turn conversation history data based on the speaking generation model comprises: In a preset multi-mode knowledge graph, retrieving the multi-mode business knowledge related to the key text of the current input information and the current intention recognition result as the business knowledge context; Generating a target message based on the key text, the business knowledge context and the multi-turn dialogue history data, wherein the target message comprises the comprehensive state of the current input information, information background knowledge and historical dialogue state; And determining a task target corresponding to the current input information based on the target message, and generating a speaking operation containing target problem answering or operation guiding information based on the task target as the target replying speaking operation.
8. A speech recommendation apparatus based on context and intent prediction, the speech recommendation apparatus comprising: The context acquisition module is used for acquiring current input information of a current user and multi-round dialogue historical data of the current user and an agent, wherein the multi-round dialogue historical data comprises historical input information and agent reply information corresponding to the historical input information; the reply phone operation generating module is used for carrying out intention recognition on the current input information to obtain a current intention recognition result, and generating a target reply phone operation corresponding to the current input information and the multi-round dialogue history data based on a phone operation generating model; The guided phone generation module is used for predicting and obtaining potential follow-up intentions of the current user based on a preset logic map and the current intention recognition result, and generating target guided phones corresponding to the multi-round dialogue historical data, the business knowledge context and the potential follow-up intentions based on the phone generation model; and the customer service call recommendation module is used for obtaining the recommended call of the current user based on the target answer call and/or the target guide call.
9. A computer device comprising a processor, a memory, and a context and intent prediction based session recommendation program stored on the memory and executable by the processor, wherein the context and intent prediction based session recommendation program, when executed by the processor, implements the steps of the context and intent prediction based session recommendation method of any of claims 1 to 7.
10. A computer readable storage medium, wherein a context and intent prediction based speech recommendation program is stored on the computer readable storage medium, wherein the context and intent prediction based speech recommendation program, when executed by a processor, implements the steps of the context and intent prediction based speech recommendation method according to any of claims 1 to 7.

Description

Speaking recommendation method, device, equipment and medium based on context and intention prediction Technical Field The present application relates to the field of intelligent decision making, and in particular, to a method, apparatus, computer device, and computer readable storage medium for speaking recommendation based on context and intent prediction. Background In a traditional intelligent customer service system, a keyword matching and regular expression method is adopted to process user input, and the method is difficult to capture consistency between conversations, so that faults exist in understanding of user demands, and a dynamic evolution process of user intention cannot be accurately identified, so that generated conversation lacks pertinence and logical consistency, and the accuracy of conversation is reduced. For example, in the medical field, a patient may go from the inquiry "does a hypertensive patient can eat canna. Or in the financial field, the customer might ask "how to transact credit card sessions"? then the inquiry is made about how do the stage commission calculate? next query "what effect of advance payouts. Therefore, how to improve the accuracy of customer service answering operation becomes a technical problem to be solved at present. Disclosure of Invention The application mainly aims to provide a context and intention prediction-based conversation recommendation method, a device, computer equipment and a computer-readable storage medium, aiming at improving the accuracy of customer service answering conversation. In order to achieve the above object, the present application provides a speaking recommendation method based on context and intent prediction, the speaking recommendation method comprising the steps of: Acquiring current input information of a current user and multi-round dialogue historical data of the current user and an agent, wherein the multi-round dialogue historical data comprises historical input information and agent reply information corresponding to the historical input information; Performing intention recognition on the current input information to obtain a current intention recognition result, and generating a target answer phone corresponding to the current input information and the multi-round dialogue history data based on a phone generation model; Predicting and obtaining potential follow-up intentions of the current user based on a preset logic map and the current intention recognition result, and generating target guide dialogs corresponding to the multiple rounds of dialogue history data, the business knowledge context and the potential follow-up intentions based on the dialog generation model; and based on the target answer phone operation and/or the target guide phone operation, obtaining the recommended phone operation of the current user. In addition, in order to achieve the above object, the present application also provides a speaking recommendation device based on context and intention prediction, the speaking recommendation device comprising: The context acquisition module is used for acquiring current input information of a current user and multi-round dialogue historical data of the current user and an agent, wherein the multi-round dialogue historical data comprises historical input information and agent reply information corresponding to the historical input information; the reply phone operation generating module is used for carrying out intention recognition on the current input information to obtain a current intention recognition result, and generating a target reply phone operation corresponding to the current input information and the multi-round dialogue history data based on a phone operation generating model; The guided phone generation module is used for predicting and obtaining potential follow-up intentions of the current user based on a preset logic map and the current intention recognition result, and generating target guided phones corresponding to the multi-round dialogue historical data, the business knowledge context and the potential follow-up intentions based on the phone generation model; and the customer service call recommendation module is used for obtaining the recommended call of the current user based on the target answer call and/or the target guide call. In addition, to achieve the above object, the present application also provides a computer device including a processor, a memory, and a context-based and intent-prediction session recommendation program stored on the memory and executable by the processor, wherein the context-based and intent-prediction session recommendation program, when executed by the processor, implements the steps of the context-based and intent-prediction session recommendation method as described above. In addition, in order to achieve the above object, the present application further provides a computer readable storage medium having stored thereon a context and intent predic