Dialogue reply text generation method and device, equipment and medium

Abstract

The application relates to the technical field of data analysis, and discloses a dialogue reply text generation method, device, equipment and medium, which comprises the following steps: collecting user dialogue input and converting the same into standardized text, performing semantic recognition, simultaneously analyzing emotional characteristic values, combining the two to complete intention recognition, obtaining an intention label; integrating a user portrait, product information and the intention label to construct a dialogue state vector, determining a reply type label through a preset strategy network model; extracting relationship entities in reliance on an insurance product knowledge graph, searching for a dialogue fragment in a dialogue skill library in combination with the reply type, and finally generating a natural and fluent reply text by using a large language model; and the application can be applied to a financial technology business system platform. The adaptability of the text and the user experience are improved when the reply text is generated.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method, apparatus, device, and medium for generating dialogue response text. Background Technology

[0002] With the continuous iteration of artificial intelligence and natural language processing technologies, the fintech industry is comprehensively promoting the transformation to digital and intelligent services. Intelligent dialogue customer service has become a core interactive platform in financial sectors such as insurance, banking, and wealth management. Relying on artificial intelligence technologies such as semantic recognition, intent classification, and text generation, traditional human-dominated financial consultation services are gradually upgrading to automated and intelligent human-computer interaction models. These services can handle high-frequency service demands such as product consultation, business Q&A, and problem feedback around the clock, effectively alleviating the pressure on financial institutions' human customer service and shortening user service response times. They are now widely used in intelligent consultation scenarios for various financial products and are an important support for the construction of a digital financial service system.

[0003] Current intelligent dialogue solutions in the fintech field still suffer from numerous technical shortcomings, making it difficult to adapt to the refined service needs of insurance product consultations. Most existing technologies rely solely on single text semantics for intent determination, ignoring the emotional characteristics of user dialogue, which easily leads to biased intent recognition and one-sided interpretation of needs. Furthermore, traditional solutions fail to integrate user profiles and insurance product-specific attributes to construct dynamic dialogue states, resulting in rigid and singular response strategies that are difficult to adapt to personalized consultation scenarios for different users. In addition, existing intelligent responses often use generic templates without combining insurance professional knowledge graphs to select precise dialogue materials, resulting in insufficient professionalism and poor adaptability in response content. Moreover, they cannot achieve multi-dimensional feature fusion decision-making, leading to low accuracy, awkward interaction, and weak scenario adaptability in intelligent consultations, severely impacting the service quality and user experience of intelligent consultations for insurance products. Summary of the Invention

[0004] This invention provides a method, apparatus, computer device, and medium for generating dialogue response text, in order to solve the problems of insufficient adaptability and poor user experience of existing dialogue response text generation methods on the market.

[0005] Firstly, a method for generating dialogue response text is provided, including: Obtain the dialogue input when a user inquires about a preset product, convert the dialogue input into text, and obtain the user dialogue text; Perform semantic recognition on the user dialogue text and generate semantic recognition results; Analyze the sentiment feature values ​​of the user dialogue text, and perform intent recognition on the dialogue text based on the semantic recognition results and the sentiment feature values ​​to obtain intent tags; Obtain user profile data and product information data of the preset product, and construct a dialogue state vector based on the profile data, the product information data and the intent tag; The response type label is determined based on the dialogue state vector using a preset policy network model; Obtain the relational entities of the preset products from the preset insurance product knowledge graph, and search for corresponding script fragments in the preset script library based on the relational entities and the response type tags to obtain a set of script fragments; The system uses a pre-defined large language model to generate natural language text based on the set of dialogue fragments, thereby obtaining the response text for the dialogue input.

[0006] Secondly, a device for generating dialogue response text is provided, comprising: The text conversion module is used to acquire the dialogue input when a user inquires about a preset product, convert the dialogue input into text, and obtain the user dialogue text. The intent recognition module is used to perform semantic recognition on the user dialogue text, generate semantic recognition results, analyze the sentiment feature values ​​of the user dialogue text, and perform intent recognition on the dialogue text based on the semantic recognition results and the sentiment feature values ​​to obtain intent tags; The type confirmation module is used to obtain user profile data and product information data of the preset product, construct a dialogue state vector based on the profile data, the product information data and the intent tag, and determine the response type tag based on the dialogue state vector using a preset policy network model. The script retrieval module is used to obtain the relationship entities of the preset products from the preset insurance product knowledge graph, and search for the corresponding script segments in the preset script library according to the relationship entities and the response type tags to obtain a set of script segments; The response generation module is used to generate natural language text based on the set of dialogue fragments using a preset large language model, thereby obtaining the response text of the dialogue input.

[0007] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described dialogue response text generation method.

[0008] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described dialogue response text generation method.

[0009] The aforementioned method, apparatus, computer equipment, and storage medium for generating dialogue response text enable the collection of user input regarding inquiries about preset insurance products. This multimodal input is then uniformly converted into standardized user dialogue text, unifying the multimodal input data format and eliminating invalid data interference. This provides a standardized and reliable data foundation for accurate analysis of user inquiries. Subsequently, refined semantic recognition is performed on the user dialogue text to obtain corresponding semantic recognition results. This fully mines the entity information and semantic relationships within the user inquiry text, achieving accurate interpretation of the core content of the user inquiry. Simultaneously, the emotional feature values ​​of the user dialogue text are analyzed. Combining the semantic recognition results with the emotional feature values ​​collaboratively completes the dialogue intent recognition, obtaining accurate intent labels. The dual-dimensional feature fusion effectively avoids intent judgment bias caused by single semantic recognition, significantly improving the accuracy of user inquiry intent recognition. Based on this, user profile data, product information data of preset insurance products, and the identified intent are integrated. The system uses tags to construct a dialogue state vector that comprehensively represents user characteristics, product attributes, and dialogue needs. This enables a comprehensive quantitative representation of the current overall consultation scenario, providing complete feature support for subsequent intelligent response strategy decisions. A preset strategy network model adaptively determines the appropriate response type tag based on the dialogue state vector, dynamically matching the optimal response strategy according to different user and product scenarios. This breaks the limitations of traditional fixed response patterns. Furthermore, it leverages an insurance product knowledge graph to extract corresponding preset product relationship entities. Combining these relationship entities with the response type tag, it retrieves and matches effective dialogue fragment sets from a preset script library. This ensures that the retrieved dialogue content aligns with insurance product expertise and fits the current response needs, improving the professionalism and scenario adaptability of the response content. Finally, a preset large language model integrates and optimizes the dialogue fragment set, performing natural language generation processing to output a professional and fluent response text that fits the consultation scenario, improving the adaptability and user experience of the generated response text. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an application environment for a dialogue response text generation method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for generating dialogue response text in one embodiment of the present invention; Figure 3 yes Figure 2 A flowchart illustrating a text conversion process in step S1. Figure 4 yes Figure 2 A flowchart illustrating a semantic recognition process in step S2. Figure 5 yes Figure 2 A flowchart illustrating the sentiment feature analysis process in step S3. Figure 6 yes Figure 2 A flowchart illustrating an intent recognition process in step S3. Figure 7 yes Figure 2 A schematic diagram of a process for constructing a dialogue state vector in step S4; Figure 8 yes Figure 2 A flowchart illustrating a dialogue retrieval process in step S6. Figure 9 This is a schematic diagram of a dialogue reply text generation device according to an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 11 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The dialogue response text generation method provided in this embodiment of the invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can collect user input regarding consultations with preset insurance products from the client, and convert the multimodal dialogue input into standardized user dialogue text. Then, it performs refined semantic recognition on the user dialogue text to obtain corresponding semantic recognition results. Simultaneously, it analyzes the sentiment feature values ​​of the user dialogue text, and combines the semantic recognition results with the sentiment feature values ​​to collaboratively complete dialogue intent recognition, obtaining accurate intent tags. Based on this, the invention integrates user profile data, product information data of preset insurance products, and the identified intent tags to construct a dialogue state vector that comprehensively represents user characteristics, product attributes, and dialogue needs. A preset strategy network model adaptively determines the response type tag suitable for the current scenario based on the dialogue state vector. Furthermore, it relies on the insurance product knowledge graph to extract the corresponding preset product relationship entities, and combines the relationship entities and response type tags to retrieve and match effective dialogue fragment sets in a preset dialogue script library. Finally, a preset large language model integrates and optimizes the dialogue fragment set and performs natural language generation processing to output a professional and fluent response text that fits the consultation scenario, improving the adaptability and user experience of the generated response text. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will now be described in detail through specific embodiments.

[0014] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the dialogue response text generation method provided in this embodiment of the invention includes the following steps: S1. Obtain the dialogue input when the user consults a preset product, convert the dialogue input into text, and obtain the user dialogue text.

[0015] In the fintech field, the dialogue input refers to the raw interaction data generated when a user initiates an interaction with an intelligent customer service representative. The data format may include text messages, recorded audio, multimedia messages, etc. The content usually involves financial-related requests such as business inquiries, product inquiries, claims, complaint feedback, and business processing requests. It also carries conversation context data such as conversation identifiers, interaction sequence information, and user identity association information. It serves as the original data source for subsequent semantic parsing, sentiment analysis, intent recognition, and response generation processing.

[0016] In this embodiment of the invention, reference is made to Figure 3 The diagram illustrates a text conversion process provided by this solution. The step of converting the dialogue input into text to obtain the user dialogue text includes: S21. Identify the data type of the dialogue input; S22. If the data type is text, then the dialogue input is determined to be user dialogue text; S23. If the data type is audio, then perform speech recognition on the dialogue input to obtain the initial text; S24. Remove spaces and preset stop words from the initial text to obtain the user dialogue text.

[0017] In detail, identifying the data type of the dialogue input involves performing format verification and attribute determination on the received raw dialogue input data, identifying the data storage and transmission type of the current dialogue input, distinguishing between text type data and audio type data, and providing a basis for subsequent targeted text conversion processing.

[0018] In detail, the statement that if the data type is text, then the dialogue input is determined to be user dialogue text means that when the data type of the dialogue input is determined to be text format, the original dialogue input data is directly treated as unprocessed original text content and initially identified as user dialogue text to be used, without the need for format conversion processing.

[0019] In detail, if the data type is audio, then performing speech recognition on the dialogue input to obtain initial text means that when the data type of the dialogue input is determined to be audio format, speech recognition technology is used to parse and transcribe the audio-format dialogue input content, converting the speech semantic information carried by the audio into corresponding text content, generating unprocessed initial text.

[0020] In detail, the process of removing spaces and preset stop words from the initial text to obtain user dialogue text involves cleaning the initial text generated by speech recognition, uniformly removing various invalid space characters contained in the text, matching it against a pre-set stop word dictionary with no actual semantic meaning, eliminating stop word content in the text, filtering invalid and redundant information, and finally obtaining clean and effective standardized user dialogue text.

[0021] In this embodiment of the invention, by acquiring the dialogue input of a user when consulting a preset product, the dialogue input is converted into text to obtain the user dialogue text. This can unify the data format of the user's multimodal consultation input, remove invalid noise content, and provide clean and standardized text base data for subsequent semantic and sentiment analysis.

[0022] S2. Perform semantic recognition on the user dialogue text and generate semantic recognition results.

[0023] In this embodiment of the invention, reference is made to Figure 4 The diagram illustrates a semantic recognition process provided by this solution. The step of performing semantic recognition on the user dialogue text and generating a semantic recognition result includes: S31. Perform word segmentation on the user dialogue text to generate a word sequence; S32. Perform part-of-speech tagging on each word in the word sequence to generate part-of-speech tagging results; S33. Based on the part-of-speech tagging results, perform entity recognition on the words in the word sequence to obtain an entity list; S34. Identify the entity relationships between entities in the entity list and generate an entity relationship list; S35. Perform vector encoding on the entity list to generate an entity vector list; S36. Summarize the entity list, the entity relationship list, and the entity vector list to obtain the semantic recognition result.

[0024] In detail, the step of segmenting the user dialogue text to generate a word sequence involves splitting the cleaned and standardized user dialogue text into individual word units according to linguistic semantic rules, removing meaningless characters, and arranging all valid words in an orderly manner to form a regular and ordered text word sequence.

[0025] In detail, the step of performing part-of-speech tagging on each word in the word sequence and generating part-of-speech tagging results involves traversing each word in the word sequence, classifying and marking the part-of-speech attributes of each word in accordance with language grammar rules, clearly distinguishing different part-of-speech categories such as nouns, verbs, adjectives, and pronouns, and providing an accurate grammatical basis for subsequent entity recognition.

[0026] In detail, the step of performing entity recognition on the words in the word sequence based on the part-of-speech tagging results to obtain an entity list involves filtering out key semantic entities with actual referential meaning in the word sequence based on the already tagged part-of-speech information, identifying and classifying various core entity contents contained in the text, integrating all valid entities and organizing them into a structured entity list.

[0027] In detail, the process of identifying entity relationships between entities in the entity list and generating an entity relationship list involves combining the overall semantic logic and syntactic structure of the user dialogue text, analyzing the association logic and semantic connections between different entities in the entity list, determining the corresponding relationship type between each entity, sorting out and recording all entity association information, and generating a complete entity relationship list.

[0028] In detail, the step of vector encoding the entity list to generate an entity vector list involves performing a unified vectorization transformation on the identified entities, mapping the semantic information of the entities in text form into numerical vectors that can be recognized and calculated by a computer, preserving the semantic features and attribute information of the entities themselves, and completing the encoding transformation of all entities one by one to form an entity vector list with unified dimensions.

[0029] In detail, the step of summarizing the entity list, the entity relationship list, and the entity vector list to obtain the semantic recognition result involves integrating and summarizing the entity list containing key text content, the entity relationship list reflecting entity association logic, and the entity vector list carrying quantitative semantic information, fusing shallow syntactic information and deep semantic quantitative information of the text to form a complete and structured text semantic recognition result.

[0030] In this embodiment of the invention, by performing semantic recognition on the user dialogue text and generating semantic recognition results, it is possible to accurately mine the entity content and relationships in the user's financial consultation text, and achieve refined analysis of the deep semantics of the user's consultation content.

[0031] S3. Analyze the sentiment feature values ​​of the user dialogue text, and perform intent recognition on the dialogue text based on the semantic recognition results and the sentiment feature values ​​to obtain intent tags.

[0032] In this embodiment of the invention, reference is made to Figure 5 The diagram shown illustrates a flowchart of a sentiment feature analysis method provided by this solution. The analysis of the sentiment feature values ​​of the user's dialogue text includes: S41. Identify preset sentiment words in the user dialogue text and generate a list of sentiment words; S42. Extract the linguistic features of the user dialogue text based on the list of sentiment words; S43. Obtain the historical dialogue content of the dialogue input, and generate a context semantic embedding vector based on the historical dialogue content; S44. Construct a comprehensive sentiment feature matrix based on the language features and the contextual semantic embedding vector; S45. Using a preset emotion computing model, the comprehensive emotion feature matrix is ​​mapped to emotion feature values.

[0033] In detail, the step of identifying preset sentiment words in the user dialogue text and generating a sentiment word list involves matching the standardized user dialogue text with a pre-built sentiment word dictionary word by word, retrieving and locating all core words in the text that have positive, negative, or neutral sentiment tendencies, filtering out effective sentiment words, and sorting, deduplicating, and summarizing them to form a sentiment word list containing all sentiment keywords.

[0034] In detail, the step of extracting the language features of the user dialogue text based on the sentiment word list involves using the compiled sentiment word list, combined with the sentence structure, number of sentiment words, distribution of sentiment words, and sentiment modification relationships of the user dialogue text, to mine explicit sentiment expression features of the text, statistically analyze the basic language parameters related to the text's sentiment, and extract multi-dimensional language features that can characterize the text's sentiment tendency and intensity.

[0035] In detail, obtaining the historical dialogue content of the dialogue input and generating a context semantic embedding vector based on the historical dialogue content involves retrieving the historical interactive dialogue text content corresponding to the current dialogue session, performing overall semantic association analysis in conjunction with the current round of user dialogue text, vectorizing the complete dialogue context information through semantic encoding, capturing implicit emotional information in the dialogue context, and generating a context semantic embedding vector that can carry context semantic association features.

[0036] In detail, the construction of a comprehensive sentiment feature matrix based on the language features and the contextual semantic embedding vector involves fusing and splicing the extracted explicit text language features with the contextual semantic embedding vector containing contextual information, integrating shallow text sentiment lexical features and deep contextual semantic features, and arranging and combining them according to fixed dimensions to construct a comprehensive sentiment feature matrix with regular dimensions and complete information.

[0037] In detail, the process of mapping the comprehensive emotional feature matrix into emotional feature values ​​using a preset emotional computing model involves inputting the constructed comprehensive emotional feature matrix into a pre-trained and converged emotional computing model. Through feature operations and weight mapping relationships within the model, the fused multi-dimensional emotional features are normalized and calculated, and finally, quantified emotional feature values ​​are output, thereby achieving accurate quantitative representation of the emotional state of user dialogue text.

[0038] In this embodiment of the invention, reference is made to Figure 6 The diagram illustrates a process for intent recognition provided by this solution. The step of performing intent recognition on the dialogue text based on the semantic recognition result and the sentiment feature value to obtain intent tags includes: S51. Extract semantic elements from the semantic recognition result, and generate a semantic dimension feature vector of the user dialogue text based on the semantic elements; S52. Establish the sentiment feature vector of the user dialogue text based on the sentiment feature values; S53. Generate weight correction coefficients based on the emotional feature values; S54. Based on the weight correction coefficient, the semantic dimension feature vector and the sentiment feature vector are weighted and fused to obtain a fused feature vector; S55. Calculate the cosine similarity between the fused feature vector and the semantic vector of each tag in the preset semantic tag library; S56. Select the label with the highest cosine similarity as the intent label of the dialogue text.

[0039] In detail, the step of extracting semantic elements from the semantic recognition result and generating a semantic dimension feature vector of the user dialogue text based on the semantic elements involves extracting core semantic elements from the complete and structured semantic recognition result, including key semantic information such as entity content, entity relationships, and entity vectors. Various discrete semantic elements are uniformly standardized and feature-mapped to transform the text semantic information into vector data with unified dimensions and quantifiable computation, thereby generating a semantic dimension feature vector that can fully represent the core semantic content of the text.

[0040] In detail, the process of establishing the emotional feature vector of the user's dialogue text based on the emotional feature values ​​involves using the quantified emotional feature values ​​as basic dimensional data, and arranging and combining them in an orderly manner according to quantitative indicators such as positive emotion, negative emotion, neutral emotion, and emotional intensity to construct a vector structure with fixed dimensions and complete feature information, thereby forming an emotional feature vector that can accurately represent the emotional state of the user's text.

[0041] In detail, the step of generating weight correction coefficients based on the emotional feature values ​​is to combine the numerical magnitude and emotional tendency of emotional feature values ​​of different dimensions, and dynamically generate corresponding weight correction coefficients according to preset weight matching rules. This differentiates the feature influence corresponding to different emotional states, so that the weight coefficients can adapt to the current user's true emotional expression, and provide a quantitative correction basis for subsequent feature fusion.

[0042] In detail, the step of weightedly fusing the semantic dimension feature vector and the sentiment feature vector based on the weight correction coefficient to obtain a fused feature vector involves matching the semantic dimension feature vector and the sentiment feature vector with corresponding weight correction coefficients, adjusting the weights of each dimension component of the two types of feature vectors through weighted operations, and then fusing and concatenating the weighted dual-path feature vectors to integrate textual semantic information and user sentiment information, thereby obtaining a fused feature vector that takes into account both semantic content and emotional state.

[0043] In detail, the calculation of the cosine similarity between the fused feature vector and the semantic vector of each tag in the preset semantic tag library involves retrieving a pre-built semantic tag library containing various business intent tags, performing cosine similarity calculations between the fused feature vector and the standard semantic vector corresponding to each intent tag in the tag library, and quantifying the semantic matching degree between the current dialogue features and each preset intent tag by using the numerical value of the angle between the vector spaces.

[0044] In detail, selecting the label with the highest cosine similarity as the intent label of the dialogue text involves sorting and comparing the cosine similarity values ​​corresponding to all preset semantic labels, filtering out the intent label with the highest similarity value and the highest semantic matching degree, and determining this label as the final intent label corresponding to the current user's dialogue text, thereby completing the intent recognition and determination of the dialogue text.

[0045] In this embodiment of the invention, by analyzing the emotional feature values ​​of the user's dialogue text, and performing intent recognition on the dialogue text based on the semantic recognition results and the emotional feature values, an intent tag is obtained. This method can accurately distinguish the user's true consultation intent by integrating semantic information and the user's emotional state, effectively avoiding the problem of intent misjudgment caused by single semantic recognition.

[0046] S4. Obtain user profile data and product information data of the preset product, and construct a dialogue state vector based on the profile data, the product information data and the intent tag.

[0047] In this embodiment of the invention, reference is made to Figure 7 The diagram illustrates a process for constructing a dialogue state vector according to this solution. The step of constructing the dialogue state vector based on the profile data, the product information data, and the intent tag includes: S61. Perform feature encoding on the profile data, the product information data, and the intent tag respectively to generate user profile sub-vector, product information sub-vector, and intent sub-vector; S62. The user profile sub-vector, the product information sub-vector, and the intent sub-vector are weighted and fused based on preset weight coefficients to obtain an initial fused vector; S63. Perform nonlinear mapping on the initial fusion vector to obtain a nonlinear feature vector; S64. Normalize the nonlinear feature vector to generate a dialogue state vector.

[0048] In detail, the step of performing feature encoding on the profile data, the product information data, and the intent tags respectively to generate user profile sub-vectors, product information sub-vectors, and intent sub-vectors involves performing unified feature encoding processing on the structured user profile data, product information data, and the identified intent tags respectively. This process converts discrete attribute data and tag data of different types and dimensions into a dense vector form with unified dimensions, resulting in user profile sub-vectors representing long-term user preference characteristics, product information sub-vectors representing the attribute characteristics of currently associated products, and intent sub-vectors representing the target characteristics of this round of dialogue.

[0049] In detail, the initial fusion vector is obtained by weighting and fusing the user profile sub-vector, the product information sub-vector, and the intent sub-vector based on preset weight coefficients. This involves retrieving pre-configured weight coefficients that satisfy normalization constraints, which respectively match the user profile sub-vector, product information sub-vector, and intent sub-vector. Vector weighting operations are then used to distribute weights to the feature components of each sub-vector, balancing user preferences, product attributes, and dialogue intent. Figure 3 The feature proportions of the participants are used to fuse and superimpose the weighted vectors to generate an initial fusion vector containing multi-dimensional dialogue information.

[0050] In detail, the nonlinear mapping of the initial fusion vector to obtain a nonlinear feature vector involves performing a nonlinear transformation on the initial fusion vector obtained by linear fusion to mine user profiles, product information, and dialogue intent. Figure 3 By leveraging the implicit cross-correlation features between features, this approach overcomes the limitations of linear feature fusion, strengthens the semantic coupling between multiple features, and generates nonlinear feature vectors with high-order feature expression capabilities.

[0051] In detail, the normalization process of the nonlinear feature vector to generate the dialogue state vector is to perform overall numerical scaling and regularization on the nonlinear feature vector obtained by nonlinear mapping, unify the numerical distribution range of each dimension of the vector, eliminate the calculation deviation caused by the difference in feature amplitude of different dimensions, ensure the stability of the overall numerical distribution of the vector and the regularization of dimensions, and finally generate a dialogue state vector that can accurately represent the current overall dialogue state.

[0052] In this embodiment of the invention, by acquiring user profile data and product information data of the preset product, a dialogue state vector is constructed based on the profile data, the product information data and the intent tag. This fully integrates user personalized features, financial product attributes and current dialogue intent, and achieves a refined and global quantitative representation of the dialogue state.

[0053] S5. Determine the response type label based on the dialogue state vector using a preset strategy network model.

[0054] In this embodiment of the invention, the step of determining the response type label based on the dialogue state vector using a preset strategy network model involves first inputting the dialogue state vector, which is constructed by fusing user profile data, product information data, and intent labels, into a pre-trained strategy network model. The strategy network model performs deep feature extraction and nonlinear operations on the user's long-term preference features, associated product features, and current user dialogue intent features contained in the dialogue state vector, and calculates the output probability score corresponding to each candidate response type. The appropriate response direction in the current dialogue scenario is determined by the probability score, and the item with the highest probability is selected from the preset response type set as the response type label corresponding to this dialogue. This completes the decision on the response strategy based on the current dialogue state and outputs a response type label that can be used for subsequent dialogue retrieval.

[0055] S6. Obtain the relational entity of the preset product from the preset insurance product knowledge graph, and search for the corresponding speech fragment in the preset speech library according to the relational entity and the response type tag to obtain a speech fragment set.

[0056] In this embodiment of the invention, reference is made to Figure 8 The diagram illustrates a script retrieval process provided by this solution. The step of searching for corresponding script segments in a preset script database based on the relational entity and the response type tag to obtain a script segment set includes: S71. Extract keywords from the response type tags to generate first search keywords; S72. Establish a second search keyword based on the relational entity; S73. Based on the first search keyword and the second search keyword, a search is performed in the speech database to obtain an initial search result set; S74. Identify conflicting statements in the initial search result set; S75. Remove conflicting statements from the initial search result set to obtain a set of statement fragments.

[0057] In detail, the step of extracting keywords from the response type tags to generate first search keywords involves semantic parsing and keyword extraction of the response type tags corresponding to the current dialogue, removing redundant modifying information, retaining core words that can accurately represent the attributes of the response strategy, and organizing and summarizing these core words to form first search keywords used to limit the dimensions of response style and response category.

[0058] In detail, the step of establishing the second search keywords based on the relational entities involves extracting core entity semantic information such as entity names, entity categories, and entity relationships contained in the text from the identified entity list and entity relationship list, organizing various entity-related semantic elements into standardized search terms, and constructing the second search keywords used to limit the dimensions of the dialogue content.

[0059] In detail, the initial search result set obtained by searching the script database based on the first search keyword and the second search keyword is to use the first search keyword representing the response strategy type and the second search keyword representing the dialogue entity content as joint search conditions to match and search the script database that stores various standardized script fragments in a pre-built database, filter out all script fragments that are suitable for both the response type and the dialogue entity scenario, and summarize them to form an initial search result set without filtering.

[0060] In detail, identifying conflicting statements in the initial search result set involves performing semantic content and business logic verification on all statement fragments within the initial search result set, comparing the expressive semantics, business orientation, and applicable scenarios of different statement fragments, identifying statement fragments with contradictory semantic content, mutually exclusive business orientations, and conflicting applicable scenarios, and completing the marking and location of conflicting statements.

[0061] In detail, the step of removing conflicting statements from the initial search result set to obtain a set of statement fragments involves uniformly removing the identified and marked conflicting statements from the initial search result set, retaining valid statement fragments that are semantically consistent, business-adaptable, and scenario-matching, completing the purification and screening of the search results, and finally obtaining a clean, conflict-free, and standardized set of statement fragments that is adapted to the current dialogue scenario.

[0062] In this embodiment of the invention, by obtaining the relational entities of the preset products from a preset insurance product knowledge graph, and searching for corresponding script fragments in a preset script library based on the relational entities and the response type tags, a set of script fragments is obtained, which effectively improves the professionalism and accuracy of the subsequently generated response content.

[0063] S7. Using a preset large language model, generate natural language text based on the set of dialogue fragments to obtain the response text of the dialogue input.

[0064] In this embodiment of the invention, the step of generating natural language text based on the set of dialogue fragments using a preset large language model to obtain the response text for the dialogue input involves inputting the retrieved set of dialogue fragments into the preset large language model. The large language model reads multiple candidate dialogue contents, matching criteria, and associated entity information within the dialogue fragment set. Using these dialogue fragments as generation materials and content constraints, and combining the semantics of the current dialogue context, the model selects, splices, polishes, and semantically rewrites each dialogue fragment, adjusts the word order, adds transitional sentences, eliminates semantic gaps between dialogue fragments, and ensures that the generated content fits the dialogue scenario and is logically coherent. Finally, it outputs fluent and complete natural language text, which serves as the response text generated for the current dialogue input.

[0065] In the fintech field, this solution can be widely applied to intelligent dialogue service scenarios in various financial businesses such as banking, insurance, wealth management, and lending. By standardizing and converting user financial dialogue input into text, accurately recognizing semantics, quantitatively analyzing emotional features, intelligently determining user intent, dynamically modeling dialogue states, and matching and generating scenario-based intelligent dialogue, it can achieve fully automated intelligent responses in financial interaction scenarios such as financial consultation, business processing, problem feedback, dispute resolution, and product consultation. It effectively solves the technical shortcomings of traditional financial intelligent customer service, such as one-sided semantic understanding, inaccurate intent recognition, templated response content, and inability to dynamically adapt response strategies based on user profiles and business product characteristics. It significantly improves the professionalism, accuracy, and humanization of financial human-computer interaction, reduces the service pressure and operating costs of financial human customer service, and enhances the financial user consultation and processing experience and the level of intelligent financial services.

[0066] As can be seen, the above-mentioned solution proposes a dialogue response text generation method, aiming to solve the technical problems of traditional intelligent dialogue systems, such as one-sided semantic understanding, low accuracy of intent recognition, single response strategy, insufficient content professionalism, and inability to dynamically adapt responses based on user characteristics and product information. This solution first obtains the user's dialogue input for a preset insurance product, uniformly converting it into standardized user dialogue text to eliminate data differences and text noise from multimodal inputs. Then, it performs refined semantic recognition on the user dialogue text, extracting text entities and entity relationships to obtain complete semantic recognition results, and simultaneously analyzes the emotional feature values ​​of the user dialogue. Combining semantic information and emotional features, it jointly completes intent recognition, effectively avoiding judgment bias caused by single-dimensional recognition. This solution further integrates user profile data, insurance product information data, and the identified intent tags to construct a dialogue state vector that comprehensively represents user preferences, product attributes, and current dialogue needs. A policy network model intelligently decides the appropriate response type tag based on the dialogue state. Simultaneously, relying on the insurance product knowledge graph, it extracts corresponding relational entities, and combines the response type tags to accurately retrieve matching dialogue fragments from a professional dialogue database, filtering conflicting content to obtain a valid dialogue set. Finally, the retrieved text fragments are integrated, polished, and reorganized using a large language model to generate natural, fluent, context-appropriate, and professionally-content intelligent response text.

[0067] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0068] In one embodiment, a dialogue response text generation device is provided, which corresponds one-to-one with the dialogue response text generation method in the above embodiments. For example... Figure 9 As shown, the dialogue response text generation device includes a text conversion module 101, an intent recognition module 102, a type confirmation module 103, a dialogue script retrieval module 104, and a response generation module 105. Detailed descriptions of each functional module are as follows: Text conversion module 101 is used to acquire the dialogue input when a user inquires about a preset product, convert the dialogue input into text, and obtain the user dialogue text; The intent recognition module 102 is used to perform semantic recognition on the user dialogue text, generate semantic recognition results, analyze the sentiment feature values ​​of the user dialogue text, and perform intent recognition on the dialogue text based on the semantic recognition results and the sentiment feature values ​​to obtain intent tags; The type confirmation module 103 is used to obtain user profile data and product information data of the preset product, construct a dialogue state vector based on the profile data, the product information data and the intent tag, and determine the response type tag based on the dialogue state vector using a preset policy network model. The script retrieval module 104 is used to obtain the relational entities of the preset products from the preset insurance product knowledge graph, and search for the corresponding script segments in the preset script library according to the relational entities and the response type tags to obtain a set of script segments. The response generation module 105 is used to generate natural language text based on the set of dialogue fragments using a preset large language model, so as to obtain the response text of the dialogue input.

[0069] In one embodiment, the text conversion module 101, when performing the conversion of the dialogue input into text to obtain the user dialogue text, is specifically used for: Identify the data type of the dialogue input; If the data type is text, then the dialogue input is determined to be user dialogue text; If the data type is audio, then speech recognition is performed on the dialogue input to obtain the initial text; Remove spaces and preset stop words from the initial text to obtain the user dialogue text.

[0070] In one embodiment, the intent recognition module 102, when performing semantic recognition on the user dialogue text and generating a semantic recognition result, is specifically used for: The user dialogue text is segmented to generate a word sequence; Each word in the word sequence is tagged with its part of speech, generating part-of-speech tagging results; Based on the part-of-speech tagging results, entity recognition is performed on the words in the word sequence to obtain an entity list; Identify the entity relationships between entities in the entity list and generate an entity relationship list; The entity list is vector-encoded to generate an entity vector list; By summing the entity list, the entity relationship list, and the entity vector list, the semantic recognition result is obtained.

[0071] In one embodiment, the intent recognition module 102, when performing the analysis of the sentiment feature values ​​of the user dialogue text, is specifically used for: Identify preset sentiment words in the user's dialogue text and generate a list of sentiment words; Extract linguistic features from the user's dialogue text based on the list of sentiment words; Obtain the historical dialogue content of the dialogue input, and generate a context semantic embedding vector based on the historical dialogue content; A comprehensive sentiment feature matrix is ​​constructed based on the language features and the contextual semantic embedding vectors; The comprehensive emotional feature matrix is ​​mapped to emotional feature values ​​using a pre-defined emotional computing model.

[0072] In one embodiment, the intent recognition module 102, when performing intent recognition on the dialogue text based on the semantic recognition result and the sentiment feature value to obtain intent tags, is specifically used for: Extract semantic elements from the semantic recognition results, and generate a semantic dimension feature vector of the user dialogue text based on the semantic elements; Based on the emotional feature values, an emotional feature vector of the user dialogue text is established; Generate weight correction coefficients based on the emotional feature values; Based on the weight correction coefficient, the semantic dimension feature vector and the sentiment feature vector are weighted and fused to obtain a fused feature vector; Calculate the cosine similarity between the fused feature vector and the semantic vector of each tag in the preset semantic tag library; The label with the highest cosine similarity is selected as the intent label of the dialogue text.

[0073] In one embodiment, the type confirmation module 103, when performing the step of constructing a dialogue state vector based on the profile data, the product information data, and the intent tag, is specifically used for: The user profile data, the product information data, and the intent tag are respectively feature-encoded to generate user profile sub-vectors, product information sub-vectors, and intent sub-vectors; The user profile sub-vector, the product information sub-vector, and the intent sub-vector are weighted and fused based on preset weight coefficients to obtain an initial fused vector; The initial fusion vector is nonlinearly mapped to obtain a nonlinear feature vector; The nonlinear feature vector is normalized to generate a dialogue state vector.

[0074] In one embodiment, the script retrieval module 104, when performing the step of searching for corresponding script segments in a preset script library based on the relational entity and the response type label to obtain a set of script segments, is specifically used for: Extract keywords from the response type tags to generate the first search keywords; A second search keyword is established based on the aforementioned relational entities; Based on the first and second search keywords, a search is performed in the speech database to obtain an initial set of search results; Identify conflicting statements in the initial search result set; Remove conflicting statements from the initial search result set to obtain a set of statement fragments.

[0075] This invention provides a dialogue response text generation device, aiming to solve the technical problems of traditional intelligent dialogue systems, such as one-sided semantic understanding, low accuracy of intent recognition, single response strategy, insufficient content professionalism, and inability to dynamically adapt responses based on user characteristics and product information. This solution first acquires user dialogue input for a preset insurance product, uniformly converting it into standardized user dialogue text to eliminate data differences and text noise from multimodal inputs. Then, it performs refined semantic recognition on the user dialogue text, extracting text entities and entity relationships to obtain complete semantic recognition results. Simultaneously, it analyzes the emotional feature values ​​of the user dialogue, combining semantic information and emotional features to jointly complete intent recognition, effectively avoiding judgment bias caused by single-dimensional recognition. This solution further integrates user profile data, insurance product information data, and the identified intent tags to construct a dialogue state vector that comprehensively represents user preferences, product attributes, and current dialogue needs. A policy network model intelligently decides on the appropriate response type tag based on the dialogue state. Simultaneously, it extracts corresponding relational entities based on an insurance product knowledge graph, and accurately retrieves matching dialogue fragments from a professional dialogue database based on the response type tags, filtering conflicting content to obtain a set of effective dialogues. Finally, the retrieved text fragments are integrated, polished, and reorganized using a large language model to generate natural, fluent, context-appropriate, and professionally-content intelligent response text.

[0076] Specific limitations regarding the dialog reply text generation device can be found in the limitations of the dialog reply text generation method described above, and will not be repeated here. Each module in the aforementioned dialog reply text generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0077] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When executed by the processor, the computer program implements the functions or steps of a dialog response text generation method on the server side.

[0078] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 11 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements client-side functions or steps of a dialog response text generation method.

[0079] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain the dialogue input when a user inquires about a preset product, convert the dialogue input into text, and obtain the user dialogue text; Perform semantic recognition on the user dialogue text and generate semantic recognition results; Analyze the sentiment feature values ​​of the user dialogue text, and perform intent recognition on the dialogue text based on the semantic recognition results and the sentiment feature values ​​to obtain intent tags; Obtain user profile data and product information data of the preset product, and construct a dialogue state vector based on the profile data, the product information data and the intent tag; The response type label is determined based on the dialogue state vector using a preset policy network model; Obtain the relational entities of the preset products from the preset insurance product knowledge graph, and search for corresponding script fragments in the preset script library based on the relational entities and the response type tags to obtain a set of script fragments; The system uses a pre-defined large language model to generate natural language text based on the set of dialogue fragments, thereby obtaining the response text for the dialogue input.

[0080] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain the dialogue input when a user inquires about a preset product, convert the dialogue input into text, and obtain the user dialogue text; Perform semantic recognition on the user dialogue text and generate semantic recognition results; Analyze the sentiment feature values ​​of the user dialogue text, and perform intent recognition on the dialogue text based on the semantic recognition results and the sentiment feature values ​​to obtain intent tags; Obtain user profile data and product information data of the preset product, and construct a dialogue state vector based on the profile data, the product information data and the intent tag; The response type label is determined based on the dialogue state vector using a preset policy network model; Obtain the relational entities of the preset products from the preset insurance product knowledge graph, and search for corresponding script fragments in the preset script library based on the relational entities and the response type tags to obtain a set of script fragments; The system uses a pre-defined large language model to generate natural language text based on the set of dialogue fragments, thereby obtaining the response text for the dialogue input.

[0081] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0082] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0083] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0084] It should be noted that in the technical solutions disclosed in this invention, the acquisition of user information (personal image data (e.g., facial videos or pictures, facial feature videos or pictures, etc.) and personal privacy information (e.g., name, ID number, occupation, address, etc.)) is all completed with the user's knowledge and consent, and the acquisition of the relevant user information is legal and compliant.

[0085] Finally, it should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application is authorized (with knowledge and consent) by the relevant parties or fully authorized by all parties, and the executing entity can obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals. The above-described embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for generating dialogue response text, characterized in that, include: Obtain the dialogue input when a user inquires about a preset product, convert the dialogue input into text, and obtain the user dialogue text; Perform semantic recognition on the user dialogue text and generate semantic recognition results; Analyze the sentiment feature values ​​of the user dialogue text, and perform intent recognition on the dialogue text based on the semantic recognition results and the sentiment feature values ​​to obtain intent tags; Obtain user profile data and product information data of the preset product, and construct a dialogue state vector based on the profile data, the product information data and the intent tag; The response type label is determined based on the dialogue state vector using a preset policy network model; Obtain the relational entities of the preset products from the preset insurance product knowledge graph, and search for corresponding script fragments in the preset script library based on the relational entities and the response type tags to obtain a set of script fragments; The system uses a pre-defined large language model to generate natural language text based on the set of dialogue fragments, thereby obtaining the response text for the dialogue input.

2. The dialogue response text generation method as described in claim 1, characterized in that, The step of converting the dialogue input into text to obtain the user dialogue text includes: Identify the data type of the dialogue input; If the data type is text, then the dialogue input is determined to be user dialogue text; If the data type is audio, then speech recognition is performed on the dialogue input to obtain the initial text; Remove spaces and preset stop words from the initial text to obtain the user dialogue text.

3. The dialogue response text generation method as described in claim 1, characterized in that, The step of performing semantic recognition on the user dialogue text and generating semantic recognition results includes: The user dialogue text is segmented to generate a word sequence; Each word in the word sequence is tagged with its part of speech, generating part-of-speech tagging results; Based on the part-of-speech tagging results, entity recognition is performed on the words in the word sequence to obtain an entity list; Identify the entity relationships between entities in the entity list and generate an entity relationship list; The entity list is vector-encoded to generate an entity vector list; By summing the entity list, the entity relationship list, and the entity vector list, the semantic recognition result is obtained.

4. The dialogue response text generation method as described in claim 1, characterized in that, The analysis of the sentiment feature values ​​of the user dialogue text includes: Identify preset sentiment words in the user's dialogue text and generate a list of sentiment words; Extract linguistic features from the user's dialogue text based on the list of sentiment words; Obtain the historical dialogue content of the dialogue input, and generate a context semantic embedding vector based on the historical dialogue content; A comprehensive sentiment feature matrix is ​​constructed based on the language features and the contextual semantic embedding vectors; The comprehensive emotional feature matrix is ​​mapped to emotional feature values ​​using a pre-defined emotional computing model.

5. The dialogue response text generation method as described in claim 1, characterized in that, The step of performing intent recognition on the dialogue text based on the semantic recognition result and the sentiment feature value to obtain intent tags includes: Extract semantic elements from the semantic recognition results, and generate a semantic dimension feature vector of the user dialogue text based on the semantic elements; Based on the emotional feature values, an emotional feature vector of the user dialogue text is established; Generate weight correction coefficients based on the emotional feature values; Based on the weight correction coefficient, the semantic dimension feature vector and the sentiment feature vector are weighted and fused to obtain a fused feature vector; Calculate the cosine similarity between the fused feature vector and the semantic vector of each tag in the preset semantic tag library; The label with the highest cosine similarity is selected as the intent label of the dialogue text.

6. The dialogue response text generation method as described in claim 1, characterized in that, The step of constructing a dialogue state vector based on the profile data, the product information data, and the intent tag includes: The user profile data, the product information data, and the intent tag are respectively feature-encoded to generate user profile sub-vectors, product information sub-vectors, and intent sub-vectors; The user profile sub-vector, the product information sub-vector, and the intent sub-vector are weighted and fused based on preset weight coefficients to obtain an initial fused vector; The initial fusion vector is nonlinearly mapped to obtain a nonlinear feature vector; The nonlinear feature vector is normalized to generate a dialogue state vector.

7. The dialogue response text generation method as described in claim 1, characterized in that, The step of searching for corresponding script segments in a preset script library based on the relation entity and the response type label to obtain a set of script segments includes: Extract keywords from the response type tags to generate the first search keywords; A second search keyword is established based on the aforementioned relational entities; Based on the first and second search keywords, a search is performed in the speech database to obtain an initial set of search results; Identify conflicting statements in the initial search result set; Remove conflicting statements from the initial search result set to obtain a set of statement fragments.

8. A device for generating dialogue response text, characterized in that, include: The text conversion module is used to acquire the dialogue input when a user inquires about a preset product, convert the dialogue input into text, and obtain the user dialogue text. The intent recognition module is used to perform semantic recognition on the user dialogue text, generate semantic recognition results, analyze the sentiment feature values ​​of the user dialogue text, and perform intent recognition on the dialogue text based on the semantic recognition results and the sentiment feature values ​​to obtain intent tags; The type confirmation module is used to obtain user profile data and product information data of the preset product, construct a dialogue state vector based on the profile data, the product information data and the intent tag, and determine the response type tag based on the dialogue state vector using a preset policy network model. The script retrieval module is used to obtain the relationship entities of the preset products from the preset insurance product knowledge graph, and search for the corresponding script segments in the preset script library according to the relationship entities and the response type tags to obtain a set of script segments; The response generation module is used to generate natural language text based on the set of dialogue fragments using a preset large language model, thereby obtaining the response text of the dialogue input.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the dialogue response text generation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the dialogue response text generation method as described in any one of claims 1 to 7.

Images