Search

CN-122027741-A - Outbound dialogue flow arrangement and business opportunity identification method

CN122027741ACN 122027741 ACN122027741 ACN 122027741ACN-122027741-A

Abstract

The invention discloses an outbound conversation process arranging and business machine identifying method, which relates to the technical field of natural language processing and comprises the steps of receiving outbound tasks and associated data, initializing core resources such as conversation identifications, context caches and the like, sending a start time combining the conversation identifications and the context caches through conversation connection, switching to a state of waiting for customer response, continuously acquiring a customer voice signal and a conversation event, converting the voice signal into a customer text, triggering timeout judgment when the voice signal does not respond to a first preset threshold, executing corresponding processing when the voice signal meets the preset conversation event, calling an online searching submodule to acquire a scene coping knowledge segment of the customer text semantic matching, integrating multidimensional information to construct prompt information, inputting a large language model to generate a reply and send, synchronously updating the context caches, combining the customer text and the updated cache, combining the large language model to assist judgment and execute business machine identification, and triggering a ending operation or returning to the state of waiting response according to the result.

Inventors

  • WANG QIAO
  • DENG CHENGGUI
  • PAN XUESONG
  • LIU MIN

Assignees

  • 成都乐云互动网络技术有限公司

Dates

Publication Date
20260512
Application Date
20260112

Claims (10)

  1. 1. An outbound dialogue flow arrangement and business opportunity identification method is characterized by comprising the following steps: step 1, receiving outbound tasks and associated data, initializing session identification, context cache, an online cable sub-module and a timer to enter an initial state; Step 2, sending a preset text generated or called based on the session identifier and the context cache to the client through call connection, and switching to a state of waiting for the response of the client; step 3, starting a timer and continuously acquiring a client voice signal and a call event in a state of waiting for a client response, converting the client voice signal into a client text through a real-time voice-to-text technology if the client voice signal is acquired, triggering timeout judgment if the client voice signal is not acquired and the timer reaches a first preset threshold, and executing a corresponding event processing flow if the preset call event is acquired; Step 4, calling an online retrieval sub-module, integrating a context cache, a client text, a scenario coping knowledge fragment, a preset system role and an action instruction to construct prompt information based on the scenario coping knowledge fragment with the client text retrieval semantic matching, inputting the prompt information into a large language model to generate a reply text, and sending the reply text to the client through call connection; And 5, combining the client text and the updated context cache, executing business opportunity identification by adopting a mode of combining rule matching with large language model auxiliary judgment, registering business opportunity information and triggering ending voice transmission if the business opportunity is positive, directly triggering ending voice transmission if the business opportunity is negative, and returning to the step 2 if the business opportunity judgment is not formed.
  2. 2. The outgoing call conversation process arrangement and business opportunity identification method of claim 1 wherein the context handling knowledge segments based on the semantic matching of the client text retrieval comprise the following: extracting core semantic features based on the client text generated in the step 3; Extracting associated semantic information of the history dialogue from the context cache initialized in the step 1, distributing first weight for the core semantic features, and distributing second weight for the associated semantic information, wherein the first weight is greater than the second weight; Fusing the core semantic features and the associated semantic information to generate a comprehensive retrieval vector; And inputting the comprehensive retrieval vector into a vector database, and retrieving a scene coping knowledge segment with the semantic similarity of the comprehensive retrieval vector meeting the preset requirement.
  3. 3. The outgoing call conversation process arrangement and business opportunity identification method of claim 2, further comprising, before retrieving the context answer knowledge piece: identifying customer group labels based on the outbound task associated data received in step 1; the method comprises the steps of calling pre-stored scenario coping knowledge segments in a knowledge base, wherein the scenario coping knowledge segments are configured with group adaptation coefficients, and the group adaptation coefficients are determined by counting the acceptance rates of different client groups to the corresponding knowledge segments based on historical dialogue data; Matching the identified customer group labels with group adaptation coefficients of the knowledge segments corresponding to each scene, and calculating to obtain dynamic adaptation scores of the knowledge segments corresponding to each scene by combining the successful application rate of the knowledge segments corresponding to each scene; And (3) when the comprehensive retrieval vector retrieval operation in the step (2) is executed, the knowledge segments are processed by the scenes meeting the preset requirement of the semantic similarity, and are sorted according to the descending order of the dynamic adaptation scores, and the first-order scene processing knowledge segments are returned in priority.
  4. 4. The method for outbound dialog flow arrangement and business identification of claim 3 wherein before sending the reply text to the client via the call connection further comprises: Analyzing the emotion state of the client through the real-time voice signals obtained in the step 3; And optimizing the expression structure and information density of the reply text based on the emotion state of the client, and generating the optimized reply text.
  5. 5. The method for arranging and identifying business machine according to claim 4, wherein the method for identifying business machine by adopting rule matching and large language model auxiliary judgment comprises the following steps: If clear intention keywords appear in the client text, triggering the rule matching weight to be temporarily maximized, and judging that the client text is a forward business opportunity; And if the rule matching result is inconsistent with the large language model auxiliary judging result, calling a preset semantic verification model, inputting a client text and an updated context cache to carry out secondary judgment, and outputting a final business opportunity identification result.
  6. 6. The method for outbound dialog flow arrangement and business identification of claim 5 wherein the trigger timeout determination in step 3 comprises the following: based on historical dialogue data, counting a client thinking time length reference value corresponding to each question type; The method comprises the steps of configuring corresponding dynamic timeout thresholds for different problem types, playing a simplified version default response and resetting a timer if the timer reaches the dynamic timeout threshold of the corresponding problem type and no client voice signal is acquired, and executing a ending flow if the timer reaches preset duration again and no client voice signal is acquired after the timer is reset.
  7. 7. The method for outbound dialog flow arrangement and business identification of claim 6 wherein the updating of the context cache in step 4 comprises the following: extracting core information from the client text and the generated reply text by adopting a lightweight semantic distillation algorithm, wherein the core information comprises client requirements, objection points and confirmed information; maintaining the dialogue core information and dialogue abstract vectors of preset rounds, and deleting redundant data; And adding a time stamp and an associated tag for the core information, wherein the associated tag comprises a demand class, an objection class and a confirmation class and is used for subsequent retrieval and call in reply generation.
  8. 8. The method for arranging and recognizing business machine according to claim 7, wherein starting a timer and continuously acquiring the client voice signal and the call event in the state of waiting for the client response, and converting the client voice signal into the client text by the real-time voice-to-text technology if the client voice signal is acquired comprises the following steps: The method comprises the steps of comparing the semantic similarity between a current client text and a preset round client text recorded in a context cache, judging repeated response if the similarity reaches a second preset threshold, not executing a complete search flow at the moment, calling a scene corresponding knowledge segment with the highest dynamic adaptation score in a history search result to generate a variant reply text, marking a client objection point corresponding to repeated response, and improving the judgment weight of the objection point in the subsequent business recognition process.
  9. 9. The method for arranging and identifying business machine according to claim 8, wherein if a call interruption event is obtained in step 3, executing the corresponding event processing procedure includes the following steps: if successful call reconnection is detected within preset time, generating an engagement session based on the dialogue progress label, and switching into a dialogue node before interruption; If the successful reconnection of the call is not detected within the preset time, information comprising the core value point and the secondary communication offer is sent to the client associated contact way, and the business machine state is synchronously updated to be called again.
  10. 10. The method for outbound dialog flow arrangement and business machine identification of claim 9, further comprising the following: after each communication is finished, based on the business opportunity judging result, if the business opportunity is judged to be forward, calling the scenario corresponding knowledge segments used in the communication, and judging whether the scenario corresponding knowledge segments are not recorded by the knowledge base; if not, automatically extracting a customer scenario and a successfully-handled text pair in the call, and checking the validity of the extracted customer scenario and the successfully-handled text pair; Correlating the client scenario vectors, the corresponding successful coping text, the client group labels identified in the step 3 and the initial successful application rate to form new scenario coping knowledge segments, and writing the new scenario coping knowledge segments into a vector database; And (3) dynamically adapting the average value of the scores of all the scenario coping knowledge fragments in the regular statistical vector database, and eliminating the scenario coping knowledge fragments with scores lower than a third preset threshold.

Description

Outbound dialogue flow arrangement and business opportunity identification method Technical Field The invention relates to the technical field of natural language processing, in particular to an outbound dialogue flow arrangement and business opportunity identification method. Background In the outbound business scenario of enterprise obtaining, the outbound system needs to interact with clients for multiple rounds based on sales clues, and the core objective is to identify potential business opportunities through effective conversations. The prior outbound technology generally adopts a fixed-line operation playing mode or a simple keyword triggering reply mode, lacks a systematic flow arrangement and context integration mechanism, on one hand, has no definite state management of a conversation flow, can not dynamically adjust interaction logic according to conversation progress, leads to reply generation to depend on a single client statement or a preset template, and is difficult to combine historical conversation information with a business knowledge base to form accurate response, and on the other hand, business machine identification is mostly dependent on single rule matching, conversation context and semantic understanding capability are not fully fused, business machine misjudgment or missed judgment easily occurs, and effective business machines can not be efficiently screened. Disclosure of Invention In order to solve the technical problems in the prior art, the invention provides an outbound dialogue flow arrangement and business opportunity identification method. The technical scheme adopted by the invention is as follows: an outbound dialogue flow arrangement and business opportunity identification method comprises the following steps: Step 1, receiving outbound tasks and associated data, initializing session identification, context cache, on-line retrieval sub-module and timer to enter an initial state, Step 2, sending a preset text generated or called based on the session identifier and the context cache to the client through call connection, and switching to a state of waiting for the response of the client; step 3, starting a timer and continuously acquiring a client voice signal and a call event in a state of waiting for a client response, converting the client voice signal into a client text through a real-time voice-to-text technology if the client voice signal is acquired, triggering timeout judgment if the client voice signal is not acquired and the timer reaches a first preset threshold, and executing a corresponding event processing flow if the preset call event is acquired; Step 4, calling an online retrieval sub-module, integrating a context cache, a client text, a scenario coping knowledge fragment, a preset system role and an action instruction to construct prompt information based on the scenario coping knowledge fragment with the client text retrieval semantic matching, inputting the prompt information into a large language model to generate a reply text, and sending the reply text to the client through call connection; And 5, combining the client text and the updated context cache, executing business opportunity identification by adopting a mode of combining rule matching with large language model auxiliary judgment, registering business opportunity information and triggering ending voice transmission if the business opportunity is positive, directly triggering ending voice transmission if the business opportunity is negative, and returning to the step 2 if the business opportunity judgment is not formed. The method has the advantages that at least one of the following steps is realized by combining the cooperative use of the session identifier, the context buffer and the timer, the orderly control and the dynamic adjustment of the outbound dialogue flow are realized, the continuity of dialogue interaction logic is effectively improved, and the communication interruption problem caused by disordered flow is reduced. In the reply generation process, context cache, client text and scenario coping knowledge segments are integrated to construct prompt information, so that generated replies can be jointed with dialogue histories and business scene demands, pertinence and suitability of reply contents are improved, and reply deviation caused by single dependence on client current sentences or preset templates is reduced. The business opportunity recognition mode of combining rule matching with large language model auxiliary judgment is adopted, the high efficiency of rule matching and the semantic understanding capability of the large language model are fully utilized, the accuracy of business opportunity recognition is improved, the misjudgment or missed judgment condition caused by a single recognition mode is reduced, and support is provided for enterprise screening of effective business opportunities. Drawings FIG. 1 is a block diagram of a system architecture corresponding to a