CN-121985071-A - Intelligent seat assisting method and system

CN121985071ACN 121985071 ACN121985071 ACN 121985071ACN-121985071-A

Abstract

The application provides an intelligent agent assisting method and system, and relates to the technical field of artificial intelligence. The method comprises the steps of monitoring a conversation in real time and calculating a conversation state measurement index, generating first auxiliary information based on rules, generating an evaluation result comprising task related value and user experience related value for candidate actions when the index exceeds a trigger threshold, selecting one candidate action as second auxiliary information, comparing the user experience related value of the second auxiliary information with a risk threshold when the type of the agent action suggested by the first auxiliary information is different from the type of the agent action suggested by the second auxiliary information, and determining a final auxiliary strategy and presenting the final strategy to the agent. According to the application, by introducing the double-target value evaluation and risk-based conflict arbitration mechanism, the task efficiency can be ensured, the emotion experience of the client can be considered, the man-machine cooperative trust degree is enhanced through strategy interpretation, and the timeliness of intervention and the security of decision making are improved.

Inventors

WANG JIANBING
Ren Peishi
CHEN SHUFEN
TIAN YUANYUAN

Assignees

上海浩宜信息科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260204

Claims (10)

1. An intelligent agent assisting method is characterized by comprising the following steps: Monitoring the conversation process of the seat and the client in real time, and calculating a conversation state measurement index representing the conversation state; Generating first auxiliary information based on rules; When the dialogue state measurement index exceeds a preset trigger threshold, generating an evaluation result aiming at least one candidate action, wherein the evaluation result comprises a task related value and a user experience related value, and selecting one candidate action as second auxiliary information according to the evaluation result; when the type of the action performed by the agent suggested by the first auxiliary information is different from the type of the action performed by the agent suggested by the second auxiliary information, comparing the user experience related value corresponding to the second auxiliary information with a preset risk threshold value to determine a final auxiliary strategy; when the type of the action performed by the agent suggested by the first auxiliary information is the same as the type of the action performed by the agent suggested by the second auxiliary information, selecting the first auxiliary information as a final auxiliary strategy; and presenting the final auxiliary strategy to the seat.
2. The method of claim 1, wherein the dialog state metric is a multidimensional dialog dead-office entropy, and wherein the computation of the multidimensional dialog dead-office entropy merges at least two of intent information entropy computed based on dialog intent distribution, abnormal silence duration duty ratio at client side, and variance of speech rate variation of client.
3. The method of claim 1, wherein the task related Value and the user experience related Value are related by a dual-objective conversational Value network, wherein the task related Value is a task completion long-term return Q Value, and wherein the user experience related Value is a future emotional Value expectation E-Value.
4. The method of claim 1, wherein comparing the user experience-related value corresponding to the second assistance information with a preset risk threshold to determine a final assistance policy, further comprises: if the user experience related value corresponding to the second auxiliary information is lower than the risk threshold, selecting the first auxiliary information as a final auxiliary strategy; and if the user experience related value corresponding to the second auxiliary information is equal to or higher than the risk threshold, selecting the second auxiliary information as a final auxiliary strategy.
5. The method according to any one of claims 1 to 4, further comprising: If the second auxiliary information is selected as a final auxiliary strategy, generating strategy explanation text for explaining the strategy selection reason based on the corresponding task related value and user experience related value, and presenting the strategy explanation text together with the final auxiliary strategy.
6. Intelligent agent assistance system, characterized by being adapted to perform the method according to any one of claims 1-5, comprising: The monitoring module is used for monitoring the conversation process of the seat and the client in real time and calculating conversation state measurement indexes representing conversation states; The prediction module is in telecommunication connection with the monitoring module and is used for receiving the dialogue state measurement index, generating first auxiliary information based on rules through an auxiliary model, generating an evaluation result for at least one candidate action based on a dialogue strategy evaluation model when the dialogue state measurement index exceeds a preset trigger threshold, wherein the evaluation result comprises a task related value and a user experience related value, and selecting one candidate action as the second auxiliary information according to the evaluation result; The arbitration module is in telecommunication connection with the prediction module and is used for receiving the first auxiliary information and the second auxiliary information, determining a final auxiliary strategy by comparing the user experience related value corresponding to the second auxiliary information with a preset risk threshold when the type of the action suggested by the first auxiliary information and the type of the action suggested by the second auxiliary information are different, selecting the first auxiliary information as the final auxiliary strategy when the type of the action suggested by the first auxiliary information and the type of the action suggested by the second auxiliary information are the same, and generating a source identifier indicating whether the final auxiliary strategy is derived from the first auxiliary information or the second auxiliary information; And the presentation module is in telecommunication connection with the arbitration module and is used for receiving and presenting the final auxiliary strategy to the seat.
7. The system of claim 6, wherein the monitoring module is configured to calculate a multidimensional dialogue impassal entropy as a measure of dialogue state, the calculation of the multidimensional dialogue impassal entropy fusing at least two of the characteristics of intent information entropy calculated based on dialogue intent distribution, abnormal silence period duty ratio at the client side, and variance of speech rate variation of the client.
8. The system of claim 6, wherein the predictive module includes a dual-objective conversational value network for generating the task related value and the user experience related value.
9. The system of claim 6, wherein the arbitration module is configured to: When the user experience related value corresponding to the second auxiliary information is lower than the risk threshold value, selecting the first auxiliary information as a final auxiliary strategy; And when the user experience related value corresponding to the second auxiliary information is equal to or higher than the risk threshold value, selecting the second auxiliary information as a final auxiliary strategy.
10. The system of any of claims 6 to 9, wherein the presentation module is further configured to: And when the source identification indicates that the final auxiliary strategy is derived from the second auxiliary information, generating strategy interpretation text based on the corresponding task related value and user experience related value, and presenting the strategy interpretation text with the final auxiliary strategy.

Description

Intelligent seat assisting method and system Technical Field The application relates to the technical field of artificial intelligence, in particular to an intelligent agent assisting method and system. Background In the field of intelligent customer service and call centers, intelligent agent auxiliary systems aim to improve service efficiency and quality. The prior art mainly provides assistance in two ways, namely, based on rules or simple models, recommending instant dialogs according to the current words of the user, and introducing models such as reinforcement learning and the like to generate a speaking strategy by optimizing a single target (such as task completion rate) which is directly related to the service for a long term. However, the prior art has significant drawbacks in practical applications. First, most reinforcement learning-based systems are directed only to task completion, ignoring the emotional experience of the customer during the conversation, which may lead to the system recommending some "cool" or "hard" strategies to achieve business goals, and even if the task is finally completed, may result in a reduced customer experience. Secondly, some systems attempt to introduce a dynamic trigger mechanism, but the trigger condition is usually based on isolated events or simple thresholds, so that it is difficult to accurately and early identify a complex "conversation tie" consisting of multiple factors together, and the system intervention time is too late. Finally, when suggestions given by different modules within the system contradict each other, the prior art generally lacks an intelligent arbitration mechanism to resolve such conflicts, and cannot explain to the seat personnel why to choose a seemingly "counter-intuitive" strategy, such a "black box" decision process makes it difficult for the seat to establish trust with the system, thereby reducing the adoption rate of effective suggestions. Disclosure of Invention The application aims to solve the technical problems that in the prior art, a dialogue strategy is single in target, the intervention time of a complex dialogue scene is not accurately judged, and intelligent and safe arbitration cannot be carried out when multi-module strategy suggestions conflict. In order to solve the technical problems, the application provides an intelligent agent assisting method, which comprises the steps of monitoring the conversation process of an agent and a client in real time, calculating conversation state measurement indexes representing conversation states, generating first auxiliary information based on rules, generating an evaluation result aiming at least one candidate action when the conversation state measurement indexes exceed a preset trigger threshold, selecting one candidate action as second auxiliary information according to the evaluation result, comparing the user experience correlation value corresponding to the second auxiliary information with a preset risk threshold when the type of the action suggested by the first auxiliary information is different from the type of the action suggested by the second auxiliary information, and presenting the final auxiliary strategy to the agent. Optionally, the dialogue state measurement index is multidimensional dialogue impartial entropy, and the computation of the multidimensional dialogue impartial entropy integrates at least two of the following characteristics of intent information entropy computed based on dialogue intent distribution, abnormal silence duration duty ratio of a client side and speech speed variation variance of the client. By fusing characteristics of multiple dimensions such as semantics, acoustics and behaviors, the sign of the conversation falling into the dead office can be accurately and early identified, and timely intervention is realized. Optionally, the task related Value and the user experience related Value are generated by a dual-target dialogue Value network, wherein the task related Value is a task completion long-term return Q Value, and the user experience related Value is a future emotion Value expected E-Value. By the method, parallel quantitative evaluation can be carried out on task completion and user experience, and a more comprehensive value reference is provided for subsequent decisions. Optionally, the comparing the user experience related value corresponding to the second auxiliary information with a preset risk threshold to determine a final auxiliary policy further includes selecting the first auxiliary information as the final auxiliary policy if the user experience related value corresponding to the second auxiliary information is lower than the risk threshold, and selecting the second auxiliary information as the final auxiliary policy if the user experience related value corresponding to the second auxiliary information is equal to or higher than the risk threshold. The arbitration mechanism realizes the aim of maximizing long-term