Intelligent customer service dialogue model construction method and application

Abstract

The invention discloses an intelligent customer service dialogue model construction method and application, and relates to the technical field of artificial intelligence. Comprising an intention recognition module, a knowledge graph management module, a multi-modal interaction module and a reply generation optimization module. Wherein the intention recognition module judges the intention category consulted by a client; performing semantic analysis on a text input by a client by adopting a natural language processing technology based on deep learning in combination with word vector representation and a bidirectional long-short term memory network; and training is performed through the tagged corpus, and feature expressions of different business intentions are autonomously learned, so that questions of clients are accurately classified into a predefined intention category system. The designed intelligent customer service template covers multiple personality dimensions, the score can be dynamically adjusted according to the feedback of the customer, and then the template most conforming to the personality requirement of the customer is recommended for the customer; therefore, the interactive experience between the customer and the customer service can be enhanced, and the recognition of the customer to the service is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular, to a method and application for constructing an intelligent customer service dialogue model. Background Art

[0002] In the current digital age, with the continuous development and application of artificial intelligence technology, intelligent customer service systems have become one of the important means for enterprises to provide customer service. The traditional customer service mode mainly relies on manual service, but this method has problems such as low efficiency and high cost. To solve these problems, intelligent customer service systems have emerged. Through technologies such as natural language processing and machine learning, they have realized automated customer service, improving service efficiency and customer satisfaction.

[0003] However, existing intelligent customer service systems still have some limitations. First, the accuracy of intent recognition needs to be improved. Although natural language processing technology based on deep learning has made certain progress, in actual applications, due to the complex and diverse problems consulted by customers, the accuracy rate of intent recognition is still not high. Second, the quality of response generation also needs to be improved. Most current intelligent customer service systems adopt a templated response method, lacking personalization and pertinence, and unable to meet the needs of different customers.

[0004] After retrieval, the application solution with the Chinese patent application number CN202411261151.2 discloses a method and device for realizing an intelligent customer service, specifically related to the field of intelligent customer service technology, including a data acquisition module for realizing an intelligent customer service, a data analysis module for realizing an intelligent customer service, an intelligent evaluation module for realizing an intelligent customer service, an intelligent comprehensive analysis module for realizing an intelligent customer service, and an intelligent human-computer interaction module for realizing an intelligent customer service; the data acquisition module for realizing an intelligent customer service is used to collect the intelligent performance data of the intelligent customer service, obtain various monitoring parameters for evaluating the intelligent performance, and transmit each monitoring parameter to the data analysis module for realizing an intelligent customer service. The solutions in the above patents have the following deficiencies: they cannot provide a more suitable personalized experience according to different users and still need to be improved. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a method and application for constructing an intelligent customer service dialogue model.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A method for constructing an intelligent customer service dialogue model includes an intent recognition module, a knowledge graph management module, a multi-modal interaction module, and a response generation optimization module;

[0008] Among them:

[0009] The intention recognition module determines the intention category of the customer's consultation; adopts natural language processing technology based on deep learning, combines word vector representation and bidirectional long short-term memory network, and performs semantic analysis on the text input by the customer; through training with labeled corpus, autonomously learns the feature expressions of different business intentions, so as to accurately classify the customer's questions into the predefined intention category system;

[0010] The reply generation optimization module comprehensively considers the intention recognition result and the knowledge graph query information, and adopts a hybrid generation strategy combining template and deep learning; for different intention categories, multiple intelligent customer service templates are pre-designed, and each template contains multiple personality tags and their corresponding scores, and the score represents the degree of strength of the personality feature in the template; according to the customer's question content and past interaction records, dynamically select the most suitable intelligent customer service template for reply generation; on the basis of the template, use RNN, LSTM or GRU combined with the attention mechanism to optimize the reply content; at the same time, according to the customer's feedback evaluation on the reply, such as satisfaction score, follow-up situation, etc., use reinforcement learning algorithm to adjust and optimize the reply generation strategy, and gradually improve the accuracy and personalization degree of the reply.

[0011] Preferably: The knowledge graph management module extracts entity, relationship and attribute information from multi-source heterogeneous data within the enterprise, including product information, service terms, frequently asked questions, historical customer cases; uses graph database technology to store and organize these knowledge elements to form a structured knowledge network; the model quickly retrieves and reasons in the knowledge system based on the defined hierarchical relationships and semantic association rules between entities, providing information basis for subsequent reply generation;

[0012] In terms of voice interaction, the multi-modal interaction module adopts a voice recognition engine to accurately convert the customer's voice instructions into text form; at the same time, uses text-to-speech technology to play the text reply generated by the model to the customer in clear and natural voice, realizing a seamless voice interaction process; in the image interaction scenario, the model receives and analyzes the picture or video information provided by the customer, combines the relevant knowledge in the knowledge graph, generates analysis text, and finally feeds back to the customer by generating corresponding image information.

[0013] Preferably: The intention recognition module includes:

[0014] The text preprocessing unit cleans the original text input by the customer, removes irrelevant characters, links, special symbol noise information; uses lexical analysis technology to perform word segmentation on the text, and splits the sentence into lexical units with actual semantics;

[0015] The feature extraction unit uses word vector representation technology to map each word in the segmented text into a low-dimensional vector space, capturing the semantic information and context relationships of the words; at the same time, it extracts the syntactic structure features of the text, such as the grammatical dependency relationships and subject-verb-object structures of the sentences, and represents them through syntactic analysis trees.

[0016] The intent classification unit constructs an intent classification model based on deep learning algorithms, using CNN, RNN or LSTM, GRU; takes the extracted word vector features and syntactic structure features as the input of the model, and through the training and learning of the model, enables it to automatically judge the intent category of the customer's consultation and output the corresponding probability distribution.

[0017] Preferably, the knowledge graph management module includes:

[0018] The knowledge extraction unit extracts knowledge entities, relationships and attribute information from various structured data sources and unstructured data sources within the enterprise; for structured data, relevant data is obtained by writing SQL query statements or using data scraping tools; for unstructured data, named entity recognition and relationship extraction techniques in natural language processing technology are used to identify the entities in the text and the relationships between them; the extracted knowledge is cleaned and standardized, removing duplicate and incorrect information, and unifying the formats and specifications of knowledge representation in different data sources to ensure the quality and consistency of knowledge.

[0019] The knowledge fusion unit uses entity linking technology to merge and deduplicate the same or similar entities from different data sources, establishing an equivalent relationship between the entities; based on ontology and semantic rules, the knowledge is fused and integrated to construct a complete knowledge network.

[0020] The knowledge storage unit stores the constructed knowledge graph based on a graph database management system. Based on the index structure in the graph database, knowledge queries are performed.

[0021] Preferably, the multimodal interaction module includes:

[0022] The voice interaction unit uses automatic speech recognition technology to convert the voice signal input by the customer into text information; uses text-to-speech technology to convert the text reply generated by the intelligent customer service model into voice output to the customer.

[0023] The image interaction unit receives and parses the image file uploaded by the customer, extracts the key features and information in the image; based on computer vision technology and image recognition algorithms, understands and analyzes the parsed image information, and combines the relevant knowledge in the knowledge graph to generate replies and feedback information for the image content.

[0024] Preferably, the reply generation optimization module adjusts the personality scoring formula based on customer feedback. Let the initial score of a certain intelligent customer service template on a certain personality dimension be S init , and the influence factor of customer feedback on this dimension is ΔS. A positive value indicates positive feedback, and a negative value indicates negative feedback. Then the adjusted score S adj is expressed as:

[0025] S adj = S init + a×ΔS

[0026] Where:

[0027] S adj is the adjusted score;

[0028] S init is the initial score;

[0029] a is the adjustment coefficient, which is used to control the influence degree of customer feedback on the score;

[0030] ΔS is the change in score caused by customer feedback;

[0031] Calculate the total score of the template by integrating the scores of each personality dimension. Suppose the intelligent customer service template has n personality dimensions, and the score of the i-th personality dimension is S i , then the comprehensive total score S total of the template is expressed as:

[0032]

[0033] Where:

[0034] S total is the comprehensive total score of the template;

[0035] n is the number of personality dimensions;

[0036] S i is the score of the i-th personality dimension.

[0037] Preferably, the reply generation optimization module includes:

[0038] A template matching unit, according to the result of intent recognition, searches for a matching reply template in a pre-constructed intelligent customer service template library. These templates are classified and sorted according to different business intents and scenarios, covering standardized answers for common questions and common scenarios; at the same time, the found templates are adaptively adjusted and modified, and according to the specific questions and context information of the customer, the variable parts in the template are filled or replaced to make the reply more in line with the actual situation of the customer;

[0039] The semantic generation unit constructs a semantic generation model based on a deep learning recurrent neural network. Through learning and training on the reply corpus of artificial customer service, the model autonomously learns the rules and patterns of text generation. When generating a reply, the model, according to the intention recognition result and the relevant information queried from the knowledge graph, combines the current context, and uses machine learning algorithms to model and analyze the customer's personality preferences and communication styles. Then, by combining the personality feature vectors of each template in the intelligent customer service template library, and calculating similarity or matching degree metrics, the most suitable intelligent customer service template that meets the customer's personality needs is recommended as the leading reply style for this interaction, and the reply text is generated.

[0040] Preferably, the reply generation optimization module further includes:

[0041] The emotion integration unit analyzes the emotional tendency in the customer's consultation text, judges whether the customer is happy, dissatisfied, anxious or in other emotional states; according to the emotion analysis result, appropriately integrates emotional factors into the reply content to make the reply more in line with the customer's emotional needs;

[0042] The quality evaluation and optimization unit conducts quality evaluation on the generated reply content; according to the quality evaluation result, optimizes and adjusts the reply generation model.

[0043] Preferably, the reply generation optimization module, based on the customer's historical interaction records denoted as X hist , personal information denoted as X info , and the current consultation question type and context environment denoted as X context , predicts the probability of the customer on a certain personality preference dimension, denoted by y, where y = 1 means like and y = 0 means dislike. The formula of the logistic regression model is as follows:

[0044]

[0045] Where:

[0046] p(y = 1│X hist ,X info ,X context ) is the probability that the customer likes a certain personality preference given the historical interaction, personal information and context environment;

[0047] z = β0 + β1X hist +β2X info +β3X context , where β0 is the intercept term, and β1, β2, β3 are the regression coefficients of the corresponding features respectively.

[0048] An application of an intelligent customer service dialogue model, which applies the intelligent customer service dialogue model to the customer service of e-commerce platforms, e-commerce live broadcasts, customer service centers of financial institutions, smart home devices, and customer service systems of telecom operators.

[0049] The beneficial effects of the present invention are:

[0050] 1. The intelligent customer service template designed by the present invention covers multiple personality dimensions, and can dynamically adjust the score according to customer feedback, and then recommend the template that best suits the customer's individual needs; this can enhance the interactive experience between customers and customer service and improve customers' sense of identity with the service.

[0051] 2. The reply generation module that combines template matching with semantic generation in the present invention can not only quickly respond to common questions based on templates, but also generate logical and coherent personalized replies based on deep learning; through the emotional integration unit, it can also adjust replies according to customer emotions to improve customer satisfaction.

[0052] 3. The intent recognition module of the present invention integrates text preprocessing, feature extraction and deep learning algorithms, which can effectively respond to diverse customer expressions, accurately determine the category of consultation intentions, and provide a solid foundation for subsequent replies; for example, in e-commerce customer service scenarios, it can quickly distinguish intentions such as product information query, order processing or after-sales complaints, thereby improving service efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of an intelligent customer service dialogue model proposed by the present invention. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0055] Embodiment 1:

[0056] A method for constructing an intelligent customer service dialogue model. The model is mainly composed of an intent recognition module, a knowledge graph management module, a multimodal interaction module, and a reply generation optimization module. The modules work together to achieve efficient communication with customers.

[0057] in:

[0058] The intent recognition module determines the intent category of customer consultation; it uses natural language processing technology based on deep learning, combined with word vector representation and bidirectional long short-term memory network (Bi-LSTM), to perform semantic analysis on the text entered by the customer; it is trained through annotated corpus and autonomously learns the characteristic expressions of different business intents, so as to accurately classify the customer's questions into the predefined intent category system. For example, in the e-commerce customer service scenario, common intent types such as product information consultation, order inquiry, return and exchange processing, complaint suggestions, etc. can be distinguished;

[0059] Among them, the intention recognition module performs semantic analysis based on the natural language processing technology of deep learning combined with a bidirectional long short-term memory network, and is implemented based on the CBOW word vector training model in Word2Vec. The specific formula is as follows:

[0060]

[0061] Among them: J(θ) is the objective function, which is used to maximize the probability of predicting the central word by the context;

[0062] θ is the parameter of the model;

[0063] N is the number of training samples;

[0064] w i is the input training word;

[0065] P(Context(w i )│w i ) is the probability of generating its context Context(w i ) given the central word w i ); by optimizing this objective function, the word vector representation of each word is trained, and these word vectors can capture the semantic information of the words and provide a basis for subsequent semantic analysis;

[0066] For the bidirectional long short-term memory network, taking the calculation of a single time step as an example, the formula is as follows:

[0067] f t =σ(W f ·[h t-1 ,x t +b f )

[0068] i t =σ(W i ·[h t-1 ,x t +b i )

[0069]

[0070] o t =σ(W o ·[h t-1 ,x t +b o )

[0071] t t =o t *tanh(C t )

[0072] Wherein:

[0073] f t 、i t 、o t respectively represent the outputs of the forget gate, input gate, and output gate;

[0074] σ is the activation function;

[0075] W f 、W i 、W C 、W o are the weight matrices of the forget gate, input gate, candidate memory cell, and output gate respectively;

[0076] b f 、b i 、b C 、b o are the corresponding bias terms;

[0077] b t-1 is the hidden state of the previous time step;

[0078] x t is the input of the current time step;

[0079] C t is the memory cell state of the current time step;

[0080] h t is the hidden state output of the current time step.

[0081] The intention recognition module includes:

[0082] A text preprocessing unit that cleans the original text input by the customer, removes irrelevant characters, links, and special symbol noise information. Using lexical analysis technology, the text is tokenized, splitting the sentence into lexical units with actual semantics. For example, for "I want to know how the camera function of this mobile phone is", after tokenization, a lexical sequence such as "I / want / to / know / how / the / camera function / of / this / mobile phone / is" is obtained.

[0083] The feature extraction unit uses word vector representation technology to map each word in the segmented text into a low-dimensional vector space, capturing the semantic information and context relationships of the words. For example, the word vectors obtained by training with the Word2Vec model can reflect the similarity and semantic associations between words. For instance, the distance between "mobile phone" and "cell phone" in the vector space is relatively close. At the same time, syntactic structure features of the text are extracted, such as the grammatical dependency relationships and subject-verb-object structures of sentences, which are represented by syntactic analysis trees. For example, for the sentence "I bought a piece of clothing in the mall yesterday", it can be clearly determined that "I" is the subject, "bought" is the predicate, "clothing" is the object, "yesterday" is the time adverbial, and "in the mall" is the place adverbial.

[0084] The intent classification unit constructs an intent classification model based on deep learning algorithms, using CNN, RNN or LSTM, GRU; taking the extracted word vector features and syntactic structure features as the input of the model, and through the training and learning of the model, enabling it to automatically judge the intent category of the customer's consultation and output the corresponding probability distribution. For example, for different types of questions input by the customer, such as order query, product consultation, complaint and suggestion, etc., the model outputs the probability values belonging to the corresponding intent categories respectively. For example, the probability of the order query intent is 0.8, and the probability of the product consultation intent is 0.6, etc. Finally, the main intent category is determined according to the probability size.

[0085] Wherein:

[0086] The knowledge graph management module extracts entity, relationship, and attribute information from multi-source heterogeneous data within the enterprise, including product information, service terms, frequently asked questions, historical customer cases, etc. Using graph database technology to store and organize these knowledge elements to form a structured knowledge network. The model quickly retrieves and reasons in the knowledge system based on the defined hierarchical relationships and semantic association rules between entities, providing information basis for subsequent response generation. Taking the financial customer service field as an example, the knowledge graph covers knowledge such as the characteristics, rate structures, risk level assessments of various financial products, and relevant policies and regulations, which can effectively support the customer service in answering professional questions.

[0087] The knowledge graph management module includes:

[0088] The knowledge extraction unit extracts knowledge entities, relationships, and attribute information from various structured data sources (such as data tables in databases and business systems) and unstructured data sources (such as product documents, technical manuals, web texts, etc.) within the enterprise. For structured data, relevant data is obtained by writing SQL query statements or using data scraping tools; for unstructured data, named entity recognition (NER) and relation extraction (RE) techniques in natural language processing are used to identify entities (such as product names, customer names, business terms, etc.) in the text and the relationships between them (such as the purchase relationship between products and customers, the composition relationship between products and product accessories, etc.). The extracted knowledge is cleaned and standardized to remove duplicate and incorrect information, and to unify the formats and specifications of knowledge representations in different data sources to ensure the quality and consistency of knowledge. For example, the name expressions of the same product in different documents are unified and standardized so that it has a unique identifier in the knowledge graph.

[0089] The knowledge fusion unit uses entity linking technology to merge and deduplicate the same or similar entities from different data sources and establish equivalent relationships between entities. For example, the same product mentioned in different channels may have different numbers or name expressions, and through entity linking, they can be identified as the same entity object. Based on ontology and semantic rules, the knowledge is fused and integrated to construct a complete knowledge network. The domain and range of the attributes and relationships of entities are defined to ensure the logical consistency and correctness of the knowledge. For example, it is stipulated that the attributes of the product entity include name, model, price, production date, etc., and the value ranges of these attributes should conform to the actual business rules and data type requirements.

[0090] The knowledge storage unit stores the constructed knowledge graph based on a graph database management system (such as Neo4j, OrientDB, etc.). Knowledge queries are performed based on the index structure in the graph database. For example, indexes are established for the key attributes of knowledge entities (such as product names, customer IDs, etc.) to quickly locate and retrieve relevant knowledge information. At the same time, the knowledge graph is partitioned for storage and management to meet the storage and processing requirements of large-scale data.

[0091] In terms of voice interaction, the multimodal interaction module uses a speech recognition engine to accurately convert the customer's voice commands into text form. At the same time, it uses text-to-speech technology to play the text response generated by the model to the customer in clear and natural voice, realizing a seamless voice interaction process. In the image interaction scenario, when the customer consults questions about the appearance of the product, operation demonstration, etc., the model receives and analyzes the picture or video information provided by the customer, combines the relevant knowledge in the knowledge graph, generates analysis text, and finally feedbacks to the customer by generating corresponding image information. In addition, through the integration of natural language processing technology with image and voice information, the model can better understand the customer's comprehensive intention. For example, from the product failure photos taken by the customer and the voice description, it can more accurately judge the problem and provide targeted solutions.

[0092] The multimodal interaction module includes:

[0093] A voice interaction unit that uses automatic speech recognition (ASR) technology to convert the voice signal input by the customer into text information. Using text-to-speech (TTS) technology, it converts the text response generated by the intelligent customer service model into voice and outputs it to the customer. During the speech synthesis process, different voice timbres, intonations, speech rates and other parameters can be selected according to the customer's needs to provide personalized voice services. For example, it provides a gentle and sweet female voice or a steady and grand male voice for the customer, and can adjust the volume and speech rate of the voice to meet the customer's auditory needs in different scenarios.

[0094] An image interaction unit that receives and parses the image file uploaded by the customer, extracts the key features and information in the image. Based on computer vision technology and image recognition algorithms, it understands and analyzes the parsed image information, combines the relevant knowledge in the knowledge graph, and generates response and feedback information for the image content. For example, if the customer uploads a picture of a math problem, the image understanding and feedback sub-unit can use optical character recognition (OCR) technology to extract the problem content, and provide the customer with detailed problem-solving ideas and answer guidance according to the math knowledge and problem-solving methods in the knowledge graph.

[0095] Among them:

[0096] The reply generation optimization module comprehensively considers the intent recognition results and the knowledge graph query information, and adopts a hybrid generation strategy that combines templates and deep learning. For different intent categories, multiple intelligent customer service templates are pre-designed. Each template contains multiple personality tags and their corresponding scores, and the scores indicate the degree of strength of the personality characteristics in the template. For example, for the intelligent customer service personality template A, its personality tags are lively (60 points), humorous (20 points), and vivid (20 points); for the intelligent customer service personality template B, its personality tags are rational (50 points), mature (30 points), and gentle (20 points). In the actual application process, according to the customer's question content and past interaction records, the most suitable intelligent customer service template is dynamically selected for reply generation. If the customer's question is about product usage skills and the customer group is more youthful, the personality template A may be selected for reply to provide information in a more lively and vivid manner, increasing the customer's acceptance and satisfaction; while for more serious questions such as those involving financial investment risk assessment, the personality template B is inclined to be selected to convey professional knowledge and advice to the customer in a rational and mature style. On the basis of the template, RNN, LSTM or GRU combined with the attention mechanism is used to optimize the reply content. By learning from a large number of high-quality artificial customer service reply corpora, the model can automatically learn the language expression habits and information organization methods under different intents, and generate more natural, fluent, accurate and detailed reply texts. At the same time, according to the customer's feedback evaluation on the reply, such as the satisfaction score, follow-up question situation, etc., the reinforcement learning algorithm is used to adjust and optimize the reply generation strategy, gradually improving the accuracy and personalization of the reply.

[0097] If the customer shows a high degree of satisfaction and positive emotional feedback to the reply of a template with a lively personality tag, and believes that the reply style meets their expectations, then the lively personality score of this template will be appropriately increased; conversely, if the customer indicates that the reply is too lively and not professional or accurate enough, the lively personality score will be correspondingly decreased, and the scores of other personality dimensions such as rational or professional may be increased. This dynamic adjustment mechanism based on customer feedback enables the intelligent customer service template to continuously adapt to the needs and preferences of different customers, and gradually optimize the reply style and personality performance.

[0098] The reply generation optimization module adjusts the personality score formula based on customer feedback. Let the initial score of a certain intelligent customer service template in a certain personality dimension (such as lively) be S init , and the influence factor of customer feedback on this dimension be ΔS (a positive value indicates positive feedback, and a negative value indicates negative feedback), then the adjusted score S adj is expressed as:

[0099] S adj = S init + a×ΔS

[0100] Where:

[0101] S adj is the adjusted score;

[0102] S init is the initial score;

[0103] a is the adjustment coefficient, which is used to control the influence degree of customer feedback on the score (the value range can be set according to specific situations, for example, between 0 and 1);

[0104] ΔS is the change in score caused by customer feedback;

[0105] Calculate the total score of the comprehensive score calculation template for each personality dimension. Suppose there are n personality dimensions in the intelligent customer service template, and the score of the i-th personality dimension is S i , then the comprehensive total score S total of this template is expressed as:

[0106]

[0107] where:

[0108] S total is the comprehensive total score of the template;

[0109] n is the number of personality dimensions;

[0110] S i is the score of the i-th personality dimension.

[0111] The reply generation optimization module includes:

[0112] A template matching unit, according to the result of intent recognition, searches for a matching reply template in a pre-constructed intelligent customer service template library. These templates are classified and sorted according to different business intents and scenarios, covering common questions and standardized answers in common scenarios. For example, for a customer's question about the order status, there is a corresponding order query template; for a customer's inquiry about product functions, there is a corresponding product function introduction template. Each template contains rich text content and reply examples, which can quickly respond to the common needs of customers. At the same time, the found template is adaptively adjusted and modified. According to the specific questions and context information of the customer, the variable parts in the template are filled or replaced to make the reply more in line with the actual situation of the customer. For example, in the order query template, according to the order number and other information provided by the customer, specific order status reply content is dynamically generated.

[0113] The semantic generation unit builds a semantic generation model based on a deep learning recurrent neural network. Through the training of the manual customer service reply corpus, the model can autonomously learn the rules and patterns of text generation. When generating a reply, the model uses the machine learning algorithm to model and analyze the customer's personality preferences and communication style based on the intention recognition results and the relevant information queried by the knowledge graph, combined with the current context. Then, combined with the personality feature vectors of each template in the intelligent customer service template library, by calculating the similarity or matching index, the intelligent customer service template that best meets the customer's personality needs is recommended as the dominant reply style for this interaction; and a coherent, accurate and logical reply text is generated. For example, when a customer asks a complex question, the semantic generation unit can organize and generate detailed and comprehensive answer content based on the semantic connotation of the question and the relevant knowledge fragments in the knowledge graph. For a customer who has shown a preference for concise, clear, professional and efficient communication methods in many past consultations, when he asks a new business question, the model will automatically select a template with a higher rational and professional personality score to reply; for another customer who likes a relaxed and friendly atmosphere, a template with a more prominent lively and humorous style will be selected to interact with him. This personalized service model can significantly improve customer engagement and satisfaction, and enhance the stickiness and trust between customers and intelligent customer service.

[0114] The emotional integration unit analyzes the emotional tendencies in customer consultation texts to determine whether the customer is happy, dissatisfied, anxious, or in other emotional states. For example, through sentiment analysis technology, it can be identified that the customer may be angry or dissatisfied when complaining about a problem. Based on the results of sentiment analysis, emotional factors are appropriately incorporated into the reply content to make the reply more in line with the customer's emotional needs. For example, when a customer expresses dissatisfaction, the reply should reflect an attitude of understanding and apology, and add soothing words and positive solutions to improve customer satisfaction and acceptance.

[0115] The quality assessment and optimization unit conducts quality assessment on the generated reply content, and checks and evaluates it from multiple dimensions such as grammatical correctness, semantic accuracy, information completeness, and logical coherence. For example, check whether there are typos in the reply, whether the sentences are fluent, whether the information provided meets customer needs, and whether the transition between each part of the content is natural. According to the quality assessment results, the reply generation model is optimized and adjusted. If it is found that some replies have quality problems or poor customer feedback, analyze the reasons and conduct targeted training and improvement on the model to continuously improve the quality and effectiveness of the replies. For example, when it is found that a certain type of reply often has grammatical errors, strengthen the grammar checking and correction mechanism for this part of the content; if the customer is not satisfied with a certain type of answer, further optimize the model's processing logic and answer strategy for this type of problem.

[0116] The reply generation optimization module predicts the probability of a customer on a certain personality preference dimension (such as whether they like a humorous style, denoted by y, where y = 1 means like and y = 0 means dislike) based on the customer's historical interaction records (feature denoted as X hist ), personal information (feature denoted as X info ), and the current consultation question type and context environment (feature denoted as X context ). The formula of the logistic regression model is as follows:

[0117]

[0118] Where:

[0119] p(y = 1│X hist ,X info ,X context ) is the probability that the customer likes a certain personality preference given the historical interaction, personal information, and context environment;

[0120] z = β0 + β1X hist + β2X info + β3X context , where β0 is the intercept term, and β1, β2, and β3 are the regression coefficients of the corresponding features respectively.

[0121] Example 2:

[0122] An application of an intelligent customer service dialogue model specifically includes:

[0123] E-commerce field:

[0124] In the customer service of e-commerce platforms, the intelligent customer service dialogue model can handle various problems of customers such as product information query, order tracking, logistics consultation, after-sales return and exchange, etc. Through the multi-modal interaction function, customers can directly upload product pictures or shoot product videos to ask relevant questions. The model combines the product knowledge graph and multi-modal data analysis to quickly and accurately provide customers with product details, usage suggestions, matching recommendations, etc., effectively improving the customer shopping experience and purchase conversion rate.

[0125] In the e-commerce live broadcast scenario, the anchor can use the intelligent customer service model to answer the questions of the audience about the live broadcast products in real time. At the same time, the model can select a suitable intelligent customer service template to give humorous or professional and detailed answers according to the atmosphere of the live broadcast room and the interaction situation of the audience, increasing the fun and interactivity of the live broadcast and promoting product sales.

[0126] Financial service field:

[0127] After introducing this model into the customer service centers of financial institutions such as banks and securities, it can handle various business intents such as customer account inquiries, fund transfers, consultation on wealth management products, and loan application. When dealing with complex issues such as financial risk assessment and investment advice, the model makes cautious responses based on the financial knowledge graph and strict risk control rules to ensure the accuracy and compliance of information. At the same time, through the personalized service mode, it provides customized financial service solutions for customers with different asset sizes and risk preferences, improving customer satisfaction and loyalty.

[0128] Smart home and Internet of Things field:

[0129] Users of smart home devices can interact with the intelligent customer service model through voice or mobile applications to query device status, control device operation, obtain home environment monitoring data, etc. The model realizes precise control and intelligent management of smart home devices by invoking the home device knowledge graph and understanding the user's voice commands. For example, the user can say "Help me set the air conditioner temperature in the living room to 26 degrees", and the model can parse the command and complete the temperature adjustment operation through the connection with the home automation system; or when the user asks "How was the electricity consumption at home last night", the model can extract relevant information from the smart home electricity meter data and generate a detailed electricity consumption report to reply to the user.

[0130] Telecom and communication field:

[0131] The customer service department of telecom operators uses this model to handle customer issues such as package consultation, phone bill inquiry, data package change, and network fault repair. When customers consult about package combinations, the model analyzes and compares based on the telecom service knowledge graph and the customer's historical consumption data to recommend the most preferential and suitable package plan for customers; when customers report network faults, the model can obtain information such as the fault location and estimated repair time through data interaction with the network management system and promptly feedback it to customers, improving the fault handling efficiency and customer satisfaction.

[0132] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for constructing an intelligent customer service dialogue model, characterized in that: It includes intent recognition module, knowledge graph management module, multimodal interaction module and reply generation optimization module; in: The intention recognition module determines the intention category of the customer's consultation; uses natural language processing technology based on deep learning, combined with word vector representation and bidirectional long short-term memory network, to perform semantic analysis on the text input by the customer; through training with annotated corpus, it autonomously learns the characteristic expressions of different business intentions, so as to accurately classify the customer's questions into the predefined intention category system; The reply generation optimization module comprehensively considers the intent recognition results and the knowledge graph query information, and adopts a hybrid generation strategy based on templates and deep learning. For different intent categories, multiple intelligent customer service templates are pre-designed, and each template contains multiple personality tags and their corresponding scores, and the score indicates the degree of the personality characteristics in the template. According to the content of the customer's question and past interaction records, the most suitable intelligent customer service template is dynamically selected for reply generation. Based on the template, RNN, LSTM or GRU is used in combination with the attention mechanism to optimize the reply content. At the same time, according to the customer's feedback evaluation on the reply, such as satisfaction score, follow-up questioning, etc., the reinforcement learning algorithm is used to adjust and optimize the reply generation strategy, and the accuracy and personalization of the reply are gradually improved.

2. The method for constructing an intelligent customer service dialogue model according to claim 1, characterized in that: The knowledge graph management module extracts entity, relationship and attribute information from multi-source heterogeneous data within the enterprise, including product information, terms of service, FAQs, and historical customer cases; uses graph database technology to store and organize these knowledge elements to form a structured knowledge network; the model quickly retrieves and infers in the knowledge system based on the hierarchical relationship and semantic association rules between the defined entities, providing information basis for subsequent response generation; In terms of voice interaction, the multimodal interaction module uses a voice recognition engine to accurately convert the customer's voice instructions into text form; at the same time, using speech synthesis technology, the text replies generated by the model are played to the customer in a clear and natural voice, realizing a seamless voice interaction process; In the image interaction scenario, the model receives and analyzes the picture or video information provided by the customer, combines it with the relevant knowledge in the knowledge graph, generates analysis text, and finally feeds back to the customer by generating corresponding image information.

3. The method for constructing an intelligent customer service dialogue model according to claim 1, characterized in that: The intention recognition module includes: The text preprocessing unit cleans the original text input by the customer to remove irrelevant characters, links, special symbols and noise information; it uses lexical analysis technology to segment the text and split the sentences into vocabulary units with actual semantics; The feature extraction unit uses word vector representation technology to map each word in the segmented text into a low-dimensional vector space to capture the semantic information and contextual relationship of the word. At the same time, it extracts the syntactic structure features of the text, such as the grammatical dependency relationship of the sentence, the subject-verb-object structure, etc., and represents them through a syntactic analysis tree. The intent classification unit builds an intent classification model based on a deep learning algorithm, using CNN, RNN or LSTM, GRU; the extracted word vector features and syntactic structure features are used as the input of the model, and through the training and learning of the model, it automatically determines the intent category of the customer's consultation and outputs the corresponding probability distribution.

4. The method for constructing an intelligent customer service dialogue model according to claim 2, characterized in that: The knowledge graph management module includes: The knowledge extraction unit extracts knowledge entities, relationships, and attribute information from various structured and unstructured data sources within the enterprise. For structured data, relevant data is obtained by writing SQL query statements or using data crawling tools. For unstructured data, named entity recognition and relationship extraction technologies in natural language processing technology are used to identify entities in the text and the relationships between them. The extracted knowledge is cleaned and standardized to remove duplicate and erroneous information, unify the format and specifications of knowledge representation in different data sources, and ensure the quality and consistency of knowledge. The knowledge fusion unit uses entity linking technology to merge and deduplicate the same or similar entities from different data sources, and establish an equivalent relationship between entities; based on ontology and semantic rules, it fuses and integrates knowledge to build a complete knowledge network; The knowledge storage unit stores the constructed knowledge graph based on the graph database management system and performs knowledge query based on the index structure in the graph database.

5. The method for constructing an intelligent customer service dialogue model according to claim 2, characterized in that: The multimodal interaction module includes: The voice interaction unit uses automatic speech recognition technology to convert the voice signals input by customers into text information; it uses text-to-speech technology to convert the text responses generated by the intelligent customer service model into voice output to customers; The image interaction unit receives and parses image files uploaded by customers, extracts key features and information from the images; based on computer vision technology and image recognition algorithms, it understands and analyzes the parsed image information, and combines relevant knowledge in the knowledge graph to generate responses and feedback information for the image content.

6. The method for constructing an intelligent customer service dialogue model according to claim 1, characterized in that: The response generation optimization module adjusts the personality scoring formula based on customer feedback. Suppose the initial score of a certain intelligent customer service template on a certain personality dimension is S init , the impact factor of customer feedback on this dimension is ΔS, a positive value indicates positive feedback, and a negative value indicates negative feedback. The adjusted score S adj It is expressed as: S adj =S init +a×ΔS in: S adj is the adjusted score; S init is the initial score; a is the adjustment coefficient, which is used to control the impact of customer feedback on the score; ΔS is the amount of change in ratings caused by customer feedback; The total score of the template is calculated by combining the scores of each personality dimension. Assume that the intelligent customer service template has n personality dimensions and the score of the i-th personality dimension is S i , then the comprehensive score of the template is S total It is expressed as: in: S total is the overall comprehensive score of the template; n is the number of personality dimensions; S i is the score of the ith personality dimension.

7. The method for constructing an intelligent customer service dialogue model according to claim 1, characterized in that: The reply generation optimization module includes: The template matching unit searches for matching reply templates in the pre-built intelligent customer service template library based on the intent recognition results. These templates are classified and organized according to different business intentions and scenarios, covering common questions and standardized answers in common scenarios. At the same time, the found templates are adaptively adjusted and modified, and the variable parts in the templates are filled or replaced according to the customer's specific questions and context information, so that the reply is more in line with the customer's actual situation. The semantic generation unit builds a semantic generation model based on a deep learning recurrent neural network. Through learning and training on the manual customer service response corpus, the model can autonomously learn the rules and patterns of text generation. When generating a response, the model uses the machine learning algorithm to model and analyze the customer's personality preferences and communication style based on the intent recognition results and the relevant information queried from the knowledge graph, combined with the current context. Then, combined with the personality feature vectors of each template in the intelligent customer service template library, by calculating the similarity or matching index, the model recommends the intelligent customer service template that best meets the customer's individual needs as the dominant response style for this interaction; and generates a response text.

8. The method for constructing an intelligent customer service dialogue model according to claim 7, characterized in that: The reply generation optimization module also includes: The emotional integration unit analyzes the emotional tendencies in customer consultation texts to determine whether the customer is happy, dissatisfied, anxious, or in other emotional states; based on the results of the emotional analysis, the unit appropriately incorporates emotional factors into the reply content to make the reply more in line with the customer's emotional needs; The quality assessment and optimization unit conducts quality assessment on the generated reply content; based on the quality assessment results, the reply generation model is optimized and adjusted.

9. The method for constructing an intelligent customer service dialogue model according to claim 8, characterized in that: The reply generation optimization module is denoted as X according to the customer's historical interaction record. hist , personal information is recorded as X info , and the current consulting question type and context environment are recorded as X context , to predict the probability of a customer on a certain personality preference dimension, represented by y, y = 1 means like, y = 0 means dislike, the formula of the logistic regression model is as follows: in: p(y=1│X hist ,X info ,X context ) is the probability that a customer likes a certain personality preference given historical interactions, personal information, and context; z=β0+β1X hist +β2X info +β3X context , where β0 is the intercept term, and β1, β2, and β3 are the regression coefficients of the corresponding features.

10. An application of an intelligent customer service dialogue model, characterized in that: The intelligent customer service dialogue model is applied in customer service of e-commerce platforms, e-commerce live broadcasting, customer service centers of financial institutions, smart home devices, and customer service systems of telecommunications operators.

Images