Customer service robot automatic reply system based on large model and knowledge base

Abstract

The invention discloses a customer service robot automatic reply system based on a large model and a knowledge base, relates to the technical field of AI automatic reply, and solves the technical problem in the prior art that when a user inquires other related knowledge, reply is not performed in combination with inquiry records of the user because only problems are considered. According to the method, the problem text of the user and the comprehensive data of the enterprise are collected, the comprehensive data are preprocessed and then used for constructing the knowledge base, and the independent user data unit is designed in the knowledge base and used for storing the historical dialogue record of the user; after the question text sent by the user is obtained in the dialog box, wrongly written characters and grammar errors in the question text are detected and corrected, the question text is optimized, and an answer question text is obtained through screening from the knowledge base; and analyzing historical dialogue records in the independent user data unit, extracting key elements in the historical dialogue records, and combining the key elements with the matched answer text to generate a user reply text for replying the user.

Description

Technical Field

[0001] The present invention belongs to the field of AI automatic reply, and specifically relates to a customer service robot automatic reply system based on a large model and a knowledge base. Background Art

[0002] In today's digital age, with the rapid development of Internet technology and the popularization of big data applications, enterprises and organizations are increasingly relying on efficient customer service systems to meet user needs; the emergence of large-scale pre-trained language models (such as BERT, GPT series, etc.) has greatly promoted the development of the field of natural language processing (NLP). These models can learn rich language structures and semantic information through unsupervised pre-training on massive text data, providing a strong foundation for subsequent tasks. These models can be used for a variety of downstream tasks, including but not limited to question answering, text generation, sentiment analysis, etc., and they have shown great potential in improving the quality and efficiency of conversations; although large models can understand the context and generate coherent answers to a certain extent, in practical applications, especially in scenarios where precise information is required, relying solely on the knowledge of the model itself is often not enough. Therefore, it becomes crucial to build and maintain a structured knowledge base.

[0003] The invention patent with application number CN2024104446509 discloses a robot customer service automatic reply method and system based on user portrait. The invention obtains a text database of user portraits and constructs an automatic reply model. The automatic reply model includes a generative natural language analyzer, a retrieval dialogue manager and a natural language generator. It receives the question text input by the user, and uses the natural language generator to calculate the cumulative consistency value of similar dialogues with the question text and other similar dialogues; outputs the similar dialogue corresponding to the largest cumulative consistency value; and replies to the user with the answer corresponding to the user input question in the similar dialogue; when obtaining the reply text of the replying user, the invention obtains the answer corresponding to the similar dialogue by analyzing the text database of the user portrait and the question text. The obtained answer is relatively single. If the user asks for other relevant information before asking the current question, it is impossible to obtain a complete and more user-friendly reply answer in combination with the user's needs.

[0004] The present invention provides a customer service robot automatic reply system based on a large model and a knowledge base to solve the above technical problems. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a customer service robot automatic reply system based on a large model and a knowledge base, which is used to solve the technical problem in the prior art that when a user asks about other relevant knowledge, only the question itself is considered, and the reply is not combined with the user's inquiry record.

[0006] To achieve the above-mentioned purpose, the first aspect of the present invention provides a customer service robot automatic reply system based on a large model and a knowledge base, comprising: a data collection module, a knowledge base construction module, a question optimization module and a dialogue management module;

[0007] Data collection module: used to collect question texts sent by users and extract comprehensive data;

[0008] Knowledge base construction module: used to construct a knowledge base based on comprehensive data, and the knowledge base contains historical user data storing user information;

[0009] Question optimization module: used to optimize the question text to obtain the standard question text, and input the standard question text into the knowledge base to match the corresponding answer text;

[0010] Dialogue management module: used to reply to users based on matching answer text and record dialogue records.

[0011] Preferably, the constructing of a knowledge base based on the comprehensive data comprises:

[0012] Extracting comprehensive data, wherein the comprehensive data includes historical question texts, corresponding historical answer texts, related search data, and historical conversation records of each user;

[0013] Construct an initial knowledge base and set up an independent user data unit in the initial knowledge base; preprocess the comprehensive data to obtain standard data, input the historical question text, historical answer text and related search data in the standard data into the initial knowledge base, store the historical conversation records in the independent user data unit, and obtain a knowledge base.

[0014] Preferably, the step of optimizing the question text to obtain a standard question text includes:

[0015] Extract question text;

[0016] The question text is normalized through natural language processing tools to detect and correct typos and grammatical errors in the question text to obtain a standard question text.

[0017] Preferably, the step of inputting the standard question text into the knowledge base to match the corresponding answer text includes:

[0018] Extract standard question text;

[0019] Identify question keywords in the standard question text, match historical question texts containing question keywords in the knowledge base, and mark the historical question texts containing question keywords as matching question texts; convert the standard question text into a standard question vector through the Embedding model, and convert the matching question text into a matching question vector; calculate the vector similarity between the standard question vector and the matching question vector; sort the matching question texts from large to small according to the vector similarity, mark the matching question text with the largest vector similarity as the answer question text, and mark the historical answer text corresponding to the answer question text as the matching answer text.

[0020] Preferably, the calculating the vector similarity between the standard question vector and the matching question vector includes:

[0021] Extracting standard question vector and matching question vector;

[0022] The standard question vector is marked as A, and the matching question vector is marked as B; the vector similarity ABS between vector A and vector B is calculated by the cosine similarity formula ABS=(A×B) / (|A||B|).

[0023] Preferably, the step of replying to the user based on the matching answer text and recording the conversation record includes:

[0024] Extract matching answer text;

[0025] The historical conversation records in the independent user data unit are analyzed through natural language processing tools to obtain the key elements of the historical conversation records, combine the key elements with the matching answer text to generate the user reply text, and reply to the user reply text in the dialog box.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] 1. The present invention collects user question texts and comprehensive data of enterprises, pre-processes the comprehensive data and uses it to build a knowledge base, and designs an independent user data unit in the knowledge base to store the user's historical conversation records. The system can better understand the user's preferences and needs, thereby providing more personalized services and suggestions, which can help the system identify the nature of the problem more quickly, reduce repeated inquiries, and improve the speed and efficiency of problem solving; after obtaining the question text sent by the user in the dialog box, typos and grammatical errors in the question text are detected and corrected, and the question text is optimized to obtain a standard question text, which can ensure that the system correctly understands the user's intentions, avoid incorrect answers caused by misunderstandings, and help the system more accurately match appropriate questions and answers, thereby providing higher quality a large number of answers; identify and extract question keywords in standard question texts, match matching question texts containing question keywords in the knowledge base, and filter out answer question texts from them, and mark the historical answer texts corresponding to the answer question texts as matching answer texts. The system can understand the user's questions more accurately, thereby providing more relevant and accurate answers, which can shorten the time of finding answers, speed up the problem-solving process, and improve the efficiency and satisfaction of interaction; analyze the historical conversation records in independent user data units, extract the key elements therein, and combine the key elements with the matching answer text to generate a user reply text for replying to the user, which can help the system understand the user's preferences and needs, thereby generating replies that are more in line with user needs and can improve the user experience.

[0028] 2. The present invention converts the standard question text and the matching question text into text vectors, and calculates the vector similarity between the standard question vector and the matching question vector through the cosine similarity formula, screens the answer question text that best matches the standard question text in the knowledge base according to the vector similarity, and marks the corresponding historical answer text as the matching answer text to obtain the standard answer to the standard question text. The present invention can more accurately find the answer closest to the user's question, thereby improving the accuracy and relevance of the answer, and helping to build a more structured and efficient database. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 The following is a flowchart of an embodiment of the present invention.

[0031] Figure 2 It is a system composition diagram of the present invention.

[0032] Figure 3 The following is a complete flowchart of obtaining user reply text in one embodiment of the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] See also Figure 1-Figure 3 , the first aspect of the present invention provides a customer service robot automatic reply system based on a large model and a knowledge base, including: a data collection module, a knowledge base construction module, a question optimization module and a dialogue management module;

[0035] Data collection module: used to collect question texts sent by users and extract comprehensive data.

[0036] Exemplarily, a front-end window is designed to allow users to ask questions through a text box, detect user submission status, and when a customer submits a question, extract the question text submitted by the customer; extract all historical question texts and corresponding historical answer texts from the historical database, as well as the user's historical conversation records in the front-end window, and integrate the historical question texts, historical answer texts and historical conversation records to obtain comprehensive data.

[0037] Knowledge base construction module: used to construct a knowledge base based on comprehensive data, and the knowledge base contains historical user data storing user information.

[0038] Exemplarily, abnormal data and missing data in the comprehensive data are detected and corrected through natural language processing tools (NLP); an initial knowledge base is constructed, and an independent user data unit is separately designed in the initial knowledge base, and the user's historical conversation records are stored in the independent user data unit. When the user enters the account password and passes the identity authentication of the independent user data unit, the independent user data unit calls the corresponding user's historical conversation records; the historical question text and the historical answer text are stored in the initial knowledge base, and the initial knowledge base is also connected to the network, and can search for network knowledge, and the stored initial knowledge base is marked as a knowledge base.

[0039] It should be noted that the knowledge base contains all uploaded material knowledge of the enterprise, including but not limited to relevant material knowledge of the enterprise's products, corporate culture, corporate background and other related materials.

[0040] Question optimization module: used to optimize the question text to obtain the standard question text, and input the standard question text into the knowledge base to match the corresponding answer text.

[0041] For example, in this embodiment, the question raised by the user is "What is the effect of using product A?" The question text is standardized through the natural language processing tool (NLP), and typos and grammatical errors in the question text are detected and corrected to obtain the standard question text "What is the effect of using product A?"

[0042] In some other embodiments, the question text may contain typos or missing characters, and it is necessary to use natural language processing tools to supplement and correct the typos or missing characters in the question text and the material knowledge in the knowledge base.

[0043] The standard question text "What is the effect of using product A" is input into the knowledge base, and the question keywords in the standard question text are extracted as "product A" and "use effect". The historical question text containing the question keywords is matched in the knowledge base to obtain matching question text 1 "What is the effect of using product A" and matching question text 2 "What effect will product A have after use"; the standard question text is converted into a text vector A[1, 1, 1, 0] through the Embedding model, and the matching question text 1 and matching question text 2 are converted into text vectors B[1, 1, 1, 0] and text vector C[1 , 0, 0, 0]; the vector similarity between vector A and vector B is calculated by the cosine similarity formula ABS = (A×B) / (|A||B|), and the vector similarity between vector A and vector C is ACS = 0.577. The matching question texts are sorted from large to small according to the size of the vector similarity, and the matching question text 2 with the largest vector similarity is marked as the answer question text, and the historical answer text corresponding to the matching question text 2 "Product A can maintain stable performance in various environments, and can guarantee a good experience whether it is daily use or under high load" is marked as the matching answer text.

[0044] Dialogue management module: used to reply to users based on matching answer text and record dialogue records.

[0045] Exemplarily, historical conversation records are analyzed through natural language processing tools (NLP) to identify key elements in historical conversation records. The key elements are the focus of users in historical conversation records, such as usage effects, frequently mentioned products or certain types of problems. In this embodiment, the key elements obtained from the user are "ease of getting started and usage effects". Combining the key elements "ease of getting started and usage effects" and the matching answer text "Product A can maintain stable performance in various environments, and can guarantee a good experience both in daily use and under high load conditions" to generate a user reply text of "Thank you very much for your inquiry! Regarding product A, it not only maintains stable performance in various environments, but also can guarantee a good experience both in daily use and under high load conditions. In addition, the design of product A also fully considers the ease of getting started for users. Even first-time users can quickly master its usage and enjoy a smooth operating experience", and the user reply text is restored to the user in the front window.

[0046] It is understandable that as users use the system more deeply, more historical conversation records are accumulated, and the system has a deeper profile of the user. When providing services to users, it can better meet the users' specific requirements and enhance the users' personalized experience.

[0047] Working principle of the present invention:

[0048] The present invention collects user's question text and enterprise's comprehensive data, constructs a knowledge base after preprocessing the comprehensive data, and designs an independent user data unit in the knowledge base for storing the user's historical conversation records; extracts the question text sent by the user, and optimizes the question text to obtain a standard question text; identifies keywords in the standard question text, and matches them with the historical question text in the knowledge base to obtain a matching question text, converts the standard question text and the matching question text into vectors, calculates the vector similarity, screens the matching question text according to the vector similarity to obtain the answer question text, marks the corresponding historical answer text as the matching answer text, identifies and extracts the key elements of the historical conversation record in the independent user data unit, combines the key elements with the matching answer text to generate a user reply text, and replies the user with the user reply text in the front-end window.

[0049] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. The customer service robot automatic reply system based on the big model and knowledge base is characterized by: include: Data collection module, knowledge base construction module, question optimization module and dialogue management module; Data collection module: used to collect question texts sent by users and extract comprehensive data; Knowledge base construction module: used to construct a knowledge base based on comprehensive data, and the knowledge base contains historical user data storing user information; Question optimization module: used to optimize the question text to obtain the standard question text, and input the standard question text into the knowledge base to match the corresponding answer text; Dialogue management module: used to reply to users based on matching answer text and record dialogue records.

2. The customer service robot automatic reply system based on a large model and a knowledge base according to claim 1 is characterized in that: The step of constructing a knowledge base based on comprehensive data includes: Extracting comprehensive data, wherein the comprehensive data includes historical question texts, corresponding historical answer texts, related search data, and historical conversation records of each user; Construct an initial knowledge base and set up an independent user data unit in the initial knowledge base; pre-process the comprehensive data to obtain standard data, input the historical question text, historical answer text and related search data in the standard data into the initial knowledge base, store the historical conversation records in the independent user data unit, and obtain a knowledge base.

3. The customer service robot automatic reply system based on a large model and a knowledge base according to claim 1 is characterized in that: The step of optimizing the question text to obtain a standard question text includes: Extract question text; The question text is normalized through natural language processing tools to detect and correct typos and grammatical errors in the question text to obtain a standard question text.

4. The customer service robot automatic reply system based on a large model and a knowledge base according to claim 1 is characterized in that: The step of inputting the standard question text into the knowledge base to match the corresponding answer text includes: Extract standard question text; Identify question keywords in the standard question text, match historical question texts containing question keywords in the knowledge base, and mark the historical question texts containing question keywords as matching question texts; convert the standard question text into a standard question vector through the Embedding model, and convert the matching question text into a matching question vector; calculate the vector similarity between the standard question vector and the matching question vector; sort the matching question texts from large to small according to the vector similarity, mark the matching question text with the largest vector similarity as the answer question text, and mark the historical answer text corresponding to the answer question text as the matching answer text.

5. The customer service robot automatic reply system based on a large model and a knowledge base according to claim 4 is characterized in that: The calculating the vector similarity between the standard question vector and the matching question vector includes: Extracting standard question vector and matching question vector; The standard question vector is marked as A, and the matching question vector is marked as B; the vector similarity ABS between vector A and vector B is calculated by the cosine similarity formula ABS=(A×B) / (|A||B|).

6. The customer service robot automatic reply system based on a large model and a knowledge base according to claim 1 is characterized in that: The step of replying to the user based on the matching answer text and recording the conversation record includes: Extract matching answer text; The historical conversation records in the independent user data unit are analyzed through natural language processing tools to obtain the key elements of the historical conversation records, combine the key elements with the matching answer text to generate the user reply text, and reply to the user reply text in the dialog box.

Images