Search

CN-122019689-A - Pre-sale consultation assisting method and system based on chat robot

CN122019689ACN 122019689 ACN122019689 ACN 122019689ACN-122019689-A

Abstract

The invention discloses a pre-sale consultation auxiliary method and a system based on a chat robot, and relates to the field of language processing, wherein the method comprises the steps of training a large language model based on a Transformer according to historical dialogue records of a whole customer service and a customer, and determining to generate a dialogue model; training a lora model according to the historical dialogue record of each customer service and a customer to obtain a lora model of each customer service, training a classification model according to the historical dialogue record of the whole customer service and the customer and corresponding purchasing effect to obtain an evaluation model, determining a QA database according to an original knowledge base by using a large language model based on a Transformer, acquiring a question of the customer, determining three corresponding answers by using a generated dialogue model, the lora model and the QA database, determining a plurality of answers by using the generated dialogue model according to the three answers, sorting the plurality of answers by using the evaluation model, and transmitting the sorting result to the corresponding customer service. The invention can improve the auxiliary effect of the chat robot.

Inventors

  • LI HUANYU
  • WEI ZIKUN
  • ZHU LING
  • SHEN YUE

Assignees

  • 云南云科特色植物提取实验室有限公司
  • 上海贝芙汀科技有限公司
  • 李寰宇

Dates

Publication Date
20260512
Application Date
20231024

Claims (7)

  1. 1. A pre-sale consultation assisting method based on a chat robot, comprising: training a large language model based on a transducer according to the historical dialogue records of the whole customer service and the customer, and determining to generate a dialogue model; training a lora model according to the history dialogue record of each customer service and the customer to obtain a lora model of each customer service; Training a classification model according to a historical dialogue record of the whole customer service and the customer and corresponding purchasing effect to obtain an evaluation model, wherein the purchasing effect is whether the customer purchases or not; determining a QA database by using a large language model based on a transducer according to an original knowledge base, wherein the QA database is used for storing feature vectors of questions and answers; respectively determining three corresponding answers by utilizing a generated dialogue model, a lora model and a QA database; According to the three answers, adopting a generated dialogue model to determine a plurality of answers; and sequencing the answers by using the evaluation model, and sending the sequencing result to the corresponding customer service.
  2. 2. The pre-sales consultation assisting method based on the chat robot according to claim 1, wherein the assessment model takes as input a dialogue record of customer service and customers and as output a probability of customer purchase.
  3. 3. The pre-sales consultation assisting method based on the chat robot according to claim 1, characterized in that the determining the QA database by using a large language model based on a fransformer according to the original knowledge base specifically includes: Disassembling the original knowledge base into a plurality of sections of characters; Generating corresponding questions and answers for each text by using a large language model based on a Transformer; Extracting feature vectors of the questions and the answers; And storing the feature vectors of the questions and the answers into a vector database to obtain a QA database.
  4. 4. The pre-sale consultation assisting method based on the chat robot according to claim 1, wherein the ranking of the plurality of answers by the evaluation model and the sending of the ranking result to the corresponding customer service specifically comprise: and the customer service determines reply content according to the sorting result and sends the reply content to the customer.
  5. 5. The pre-sales consultation assisting method based on the chat robot according to claim 4, wherein when the reply content of the customer service is not in the ordering result, the questions of the customer and the answers of the customer service are added into the corresponding customer service and the historical dialogue record of the customer, and the lora model is retrained.
  6. 6. The pre-sales consultation assisting method based on the chat robot according to claim 4, characterized in that when the reply contents of the customer service are not in the ordering result, the feature vectors of the customer questions and the answers of the customer service are extracted and the QA database is updated.
  7. 7. A pre-sale consultation assistance system based on a chat robot, comprising: The conversation model generation module is used for training a large language model based on a transducer according to the historical conversation records of the whole customer service and the customer, and determining to generate a conversation model; The LORA model training module is used for training the LORA model according to the history dialogue record of each customer service and the customer to obtain the LORA model of each customer service; The evaluation model determining module is used for training the classification model according to the historical dialogue record of the whole customer service and the customer and the corresponding purchasing effect to obtain an evaluation model, wherein the purchasing effect is whether the customer purchases or not; The QA database determining module is used for determining a QA database by using a large language model based on a transducer according to an original knowledge base, wherein the QA database is used for storing feature vectors of questions and answers; the recognition module is used for acquiring the questions of the clients and respectively determining three corresponding answers by utilizing the generated dialogue model, the lora model and the QA database; the secondary identification module is used for determining a plurality of answers by adopting a generated dialogue model according to the three answers; And the ranking module is used for ranking the answers by using the evaluation model and sending the ranking result to the corresponding customer service.

Description

Pre-sale consultation assisting method and system based on chat robot Technical Field The invention relates to the field of language processing, in particular to a pre-sale consultation assisting method and system based on a chat robot. Background The existing intelligent customer service chat robots focus more on answering the questions of the clients according to the existing knowledge or rules, and if the questions of the clients are not in the knowledge base of the chat robot, effective answers cannot be given. On the other hand, most of the existing intelligent chat robots have the defect of being hard and not lively enough, so that the consulting experience of clients is very poor. Disclosure of Invention The invention aims to provide a pre-sale consultation assisting method and system based on a chat robot, which can improve the assisting effect of the chat robot. In order to achieve the above object, the present invention provides the following solutions: A pre-sale consultation assisting method based on a chat robot, comprising: training a large language model based on a transducer according to the historical dialogue records of the whole customer service and the customer, and determining to generate a dialogue model; training a lora model according to the history dialogue record of each customer service and the customer to obtain a lora model of each customer service; Training a classification model according to a historical dialogue record of the whole customer service and the customer and corresponding purchasing effect to obtain an evaluation model, wherein the purchasing effect is whether the customer purchases or not; determining a QA database by using a large language model based on a transducer according to an original knowledge base, wherein the QA database is used for storing feature vectors of questions and answers; respectively determining three corresponding answers by utilizing a generated dialogue model, a lora model and a QA database; According to the three answers, adopting a generated dialogue model to determine a plurality of answers; and sequencing the answers by using the evaluation model, and sending the sequencing result to the corresponding customer service. Optionally, the evaluation model takes as input a dialogue record of customer service and customer, and as output a probability of customer purchase. Optionally, determining the QA database according to the original knowledge base by using a large language model based on a transducer specifically includes: Disassembling the original knowledge base into a plurality of sections of characters; Generating corresponding questions and answers for each text by using a large language model based on a Transformer; Extracting feature vectors of the questions and the answers; And storing the feature vectors of the questions and the answers into a vector database to obtain a QA database. Optionally, the ranking the plurality of answers by using the evaluation model, and sending the ranking result to the corresponding customer service, specifically including: and the customer service determines reply content according to the sorting result and sends the reply content to the customer. Optionally, when the reply content of the customer service is not in the ordering result, adding the questions of the customer and the answers of the customer service into the corresponding historical dialogue record of the customer service and the customer, and retraining the lora model. Optionally, when the reply content of the customer service is not in the ordering result, extracting the feature vector of the customer's question and the answer of the customer service, and updating the QA database. A pre-sale consultation assistance system based on chat robots, comprising: The conversation model generation module is used for training a large language model based on a transducer according to the historical conversation records of the whole customer service and the customer, and determining to generate a conversation model; The LORA model training module is used for training the LORA model according to the history dialogue record of each customer service and the customer to obtain the LORA model of each customer service; The evaluation model determining module is used for training the classification model according to the historical dialogue record of the whole customer service and the customer and the corresponding purchasing effect to obtain an evaluation model, wherein the purchasing effect is whether the customer purchases or not; The QA database determining module is used for determining a QA database by using a large language model based on a transducer according to an original knowledge base, wherein the QA database is used for storing feature vectors of questions and answers; the recognition module is used for acquiring the questions of the clients and respectively determining three corresponding answers by utilizing the generated dialogue model