Intelligent customer service question answering method and device and computer equipment

Abstract

The invention discloses an intelligent customer service question answering method and device and computer equipment. The method comprises the following steps: receiving question data input by a target object; performing word segmentation processing on the problem data to obtain a plurality of segmented words, and converting the plurality of segmented words into a plurality of word vectors; a natural language model is adopted to analyze the multiple word vectors to obtain context vectors, and the context vectors are used for representing context relations among the multiple word vectors; and determining user intention information of the question data based on the context vector, and generating reply information corresponding to the user intention information. The technical problem that the reply accuracy is low due to the fact that a dialogue robot cannot obtain the user intention in the related technology is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an intelligent customer service question-and-answer method, device, and computer equipment. Background Art

[0002] With the rapid development of artificial intelligence technology, conversational robots have gradually become part of people's daily lives. Using historical conversation data to generate intelligent question-and-answer customer service in a specific field has become a research topic of great value. In particular, the introduction of NLP language models has greatly improved the accuracy of conversational robots' answers while enhancing interactivity. Currently, operators, insurance companies, e-commerce platforms, and hotlines have all introduced conversational robots, which has greatly relieved customer service pressure and improved customer service efficiency.

[0003] The conversational robot mainly communicates with customers through program words. In the prior art, two methods are mainly used to construct customer service program words. One method is to use a single-round conversational customer service robot to answer users' questions or inquiries, and the other method is to respond to users' questions or inquiries based on a state machine jump mechanism (mainly relying on a deep learning mechanism) based on rule matching. However, the customer service words generated by the above methods need to be manually enumerated and classified in detail before building the customer service robot, or rules need to be manually set to achieve state transitions. These construction methods are complex and cumbersome, and it is difficult to ensure the comprehensiveness of the customer service words. For those special cases that are not listed, the system often cannot provide an effective response, or the response words are not precise enough, resulting in a poor user experience. Summary of the invention

[0004] The embodiments of the present application provide an intelligent customer service question-and-answer method, apparatus, and computer device to at least solve the technical problem in the related art that the conversation robot is unable to obtain the user's intention, resulting in low response accuracy.

[0005] According to one aspect of an embodiment of the present application, an intelligent customer service question and answer method is provided, including: receiving question data input by a target object; performing word segmentation processing on the question data to obtain multiple word segments, and converting the multiple word segments into multiple word vectors; using a natural language model to analyze the multiple word vectors to obtain a context vector, wherein the context vector is used to represent the context relationship between the multiple word vectors; determining user intent information of the question data based on the context vector, and generating reply information corresponding to the user intent information.

[0006] Optionally, the problem data is segmented to obtain multiple word segmentations, and the multiple word segmentations are converted into multiple word vectors, including: removing noise information in the problem data, segmenting the problem data to obtain multiple initial word segmentations, and normalizing the multiple initial word segmentations after part-of-speech restoration and part-of-speech tagging to obtain the multiple word segmentations; converting the multiple word segmentations into multiple dense vectors, and determining the multiple dense vectors as the multiple word vectors, each dimension in the word vector is used to represent a semantic feature of a word segmentation.

[0007] Optionally, a natural language model is used to analyze the multiple word vectors to obtain a context vector, including: using a natural language model to obtain a relationship weight matrix between the multiple word vectors, wherein the rows and columns of the relationship weight matrix are the multiple word vectors, and the elements in the relationship weight matrix are used to represent the relationship weights between each word vector; multiplying each word vector by the relationship weight matrix respectively to obtain a context word vector in the context vector; and aggregating the multiple word vectors according to the sentences to which the word vectors belong to obtain a context sentence vector in the context vector.

[0008] Optionally, generating answer data corresponding to the user intent information includes: identifying multiple entities from the user intent information, and determining the multiple entities as attributes of the user intent information, wherein the attributes of the user intent information include at least one of the following: time range, comparison object and question type; filling the attributes of the user intent information into preset slots to obtain intent attributes; filling the slot values ​​in the preset slots into the slots of the query template to generate a query statement; and using the query statement to query the reply information.

[0009] Optionally, using the query statement to query the reply information includes: sending the query statement to a preset knowledge graph database to perform a query operation; receiving question resolution data output by the preset knowledge graph database in response to the query operation, the question resolution data including at least: answer data corresponding to the question data; and generating reply information based on the question resolution data.

[0010] Optionally, generating reply information based on the question resolution data includes: obtaining a complexity score and a preset narrative order of the user intent information; generating a reply template according to the complexity score and the preset narrative order of the user intent information; extracting key information from the question resolution data, the key information including at least: entities in the question resolution data; filling the key information into the reply template to obtain reply information.

[0011] Optionally, the method also includes: obtaining keywords in the question data for characterizing emotional characteristics; analyzing the keywords using a preset sentiment analysis model to determine the emotional state of the target object; and modifying the reply information based on the emotional state of the target object to obtain final reply information.

[0012] According to another aspect of an embodiment of the present application, an intelligent customer service question and answer device is also provided, including: a receiving module for receiving question data input by a target object; a word segmentation module for performing word segmentation on the question data to obtain multiple word segments, and converting the multiple word segments into multiple word vectors; using a natural language model to analyze the multiple word vectors to obtain a context vector, wherein the context vector is used to represent the context relationship between the multiple word vectors; determining user intention information of the question data based on the context vector, and generating reply information corresponding to the user intention information.

[0013] According to another aspect of the embodiment of the present application, a computer device is also provided, including: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and is used to execute the above-mentioned intelligent customer service question and answer method.

[0014] According to another aspect of the embodiments of the present application, a non-volatile storage medium is also provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned intelligent customer service question and answer method by running the computer program.

[0015] According to another aspect of the embodiments of the present application, a computer program product is also provided, including computer instructions, which implement the above-mentioned intelligent customer service question and answer method when executed by a processor.

[0016] In an embodiment of the present application, question data input by a target object is received; the question data is segmented to obtain multiple word segments, and the multiple word segments are converted into multiple word vectors; a natural language model is used to analyze the multiple word vectors to obtain a context vector, wherein the context vector is used to represent the context relationship between the multiple word vectors; user intent information of the question data is determined based on the context vector, and reply information corresponding to the user intent information is generated, thereby achieving the purpose of determining the user intent information according to the context association relationship of the question data, thereby achieving the technical effect of improving the accuracy of the acquired user intent information, and further solving the technical problem in the related art that the conversation robot cannot obtain the user intent, resulting in low reply accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 It is a hardware structure block diagram of a computer terminal for implementing an intelligent customer service question and answer method according to an embodiment of the present application;

[0019] Figure 2 is a flow chart of an intelligent customer service question and answer method according to an embodiment of the present application;

[0020] Figure 3 It is a schematic diagram of the structure of an intelligent customer service question and answer system according to an embodiment of the present application;

[0021] Figure 4 is a schematic diagram of a natural language understanding module structure according to an embodiment of the present application;

[0022] Figure 5 is a schematic diagram of a decision-making reasoning module structure according to an embodiment of the present application;

[0023] Figure 6 It is a structural diagram of an intelligent customer service question and answer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] The information collected in the embodiments of the present application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with the relevant laws, regulations and standards of the relevant regions, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or reject automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered.

[0027] In order to solve the problems existing in the related art, the embodiment of the present application provides an intelligent customer service question and answer method, which can be run on Figure 1 In the computer terminal shown, the computer terminal is explained below.

[0028] The intelligent customer service question and answer method embodiment provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal for implementing the intelligent customer service question and answer method is shown. Figure 1 As shown, the computer terminal 10 may include one or more (102a, 102b, ..., 102n are used to illustrate) processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions connected via a wired and / or wireless network. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art can understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0029] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0030] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the intelligent customer service question and answer method in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned intelligent customer service question and answer method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0031] The transmission module 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0032] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0033] It should be noted that, in some optional embodiments, the above Figure 1 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0034] In the above-mentioned operating environment, an embodiment of the present application provides an embodiment of an intelligent customer service question and answer method. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0035] Figure 2 is a flow chart of an intelligent customer service question and answer method according to an embodiment of the present application. Figure 2As shown, the method comprises the following steps:

[0036] Step S202, receiving question data input by the target object;

[0037] In step S202, the problem data includes at least communication text information, communication voice information, communication picture information, communication video information, phone charge recharge, business consultation and processing, and balance inquiry.

[0038] Step S204, performing word segmentation processing on the question data to obtain multiple word segments, and converting the multiple word segments into multiple word vectors;

[0039] In step S204, when the question data includes non-text data, the non-text data may be firstly extracted and converted into text data. After the question data is converted into text data, the question data may be segmented.

[0040] Step S206, using a natural language model to analyze the multiple word vectors to obtain a context vector, wherein the context vector is used to represent a contextual relationship between the multiple word vectors;

[0041] Step S208: determining user intention information of the question data based on the context vector, and generating answer information corresponding to the user intention information.

[0042] Through the above steps S202 to S208, by receiving the question data input by the target object; performing word segmentation processing on the question data to obtain multiple word segments, and converting the multiple word segments into multiple word vectors; using a natural language model to analyze the multiple word vectors to obtain a context vector, wherein the context vector is used to represent the contextual relationship between the multiple word vectors; determining the user intention information of the question data based on the context vector, and generating the reply information corresponding to the user intention information, thereby achieving the purpose of determining the user intention information according to the contextual association relationship of the question data, thereby achieving the technical effect of improving the accuracy of the acquired user intention information, and then solving the technical problem in the related technology that the conversation robot cannot obtain the user intention, resulting in low reply accuracy. The following is a detailed description.

[0043] In some embodiments of the present application, the specific steps of performing word segmentation processing on the problem data to obtain multiple word segmentations, and converting the multiple word segmentations into multiple word vectors include: removing noise information in the problem data, performing word segmentation processing on the problem data to obtain multiple initial word segmentations, and performing part-of-speech restoration and part-of-speech tagging on the multiple initial word segmentations and then normalizing them to obtain the multiple word segmentations; converting the multiple word segmentations into multiple dense vectors, and determining the multiple dense vectors as the multiple word vectors, wherein each dimension in the word vector is used to represent a semantic feature of a word segmentation.

[0044] It should be noted that the normalization submodule performs word form restoration and part-of-speech tagging on the words after word segmentation, converting the words into basic forms, such as restoring "running" to "run", to reduce word variants and improve the consistency of word vectors. Part-of-speech tagging is to mark the grammatical function of each word in the sentence, such as noun, verb, adjective, etc., which helps with subsequent semantic analysis and understanding.

[0045] It should be noted that a pre-trained Word2Vec word vector model may be used to convert the multiple word segments into multiple dense vectors. It is understandable that the dense vectors are used to capture the semantic relationship between words.

[0046] Optionally, a natural language processing model can be used to construct contextual associations between words for dense vectors through a self-attention mechanism to generate context-sensitive word vectors and sentence vectors to obtain user intent information.

[0047] For example, the role and relationship of words in a sentence can be analyzed through the self-attention mechanism. The self-attention mechanism can capture the long-distance dependencies between words in a sentence and generate context-sensitive word vectors and sentence vectors. These word vectors and sentence vectors contain not only the meaning of the words themselves, but also the meaning and relationship between words in the current context, so as to understand the user's intention more accurately.

[0048] It is understandable that traditional word vector models usually generate a fixed vector representation for each word, and the vector is the same no matter what context the word appears in. This representation method is effective in some scenarios, but it often leads to confusion and loss of information when dealing with polysemous words or multiple contexts. Context-sensitive word vectors generate a dynamic vector representation for each word in a specific context, that is, the same word will have different vectors in different sentences or contexts. The generation of this word vector depends on the model's ability to understand the semantics of the surrounding words, so as to adjust the word vector to adapt to the meaning of the word in the current context. Such word vectors can more accurately reflect the diversity of word meanings, which is particularly important for understanding complex semantic relationships. Similar to word vectors, sentence vectors can also reflect more complex semantic information, but they encode the meaning of the entire sentence. Context-sensitive sentence vectors not only consider the meaning of the words in the sentence, but also the relationship between the words and the global structure of the sentence. Deep learning models can capture the long-range dependencies and contextual information of sentences through technologies such as self-attention mechanisms or gated recurrent units (GRUs), and generate vector representations that can reflect the specific meaning of sentences.

[0049] In a conversational robot or intelligent customer service system, using context-sensitive word vectors and sentence vectors helps to understand and generate conversation content more accurately. By analyzing the sentence vectors of user questions, the robot can better understand the intent and context of the question, thereby providing more accurate responses. At the same time, the context sensitivity of word vectors enables the robot to distinguish the different meanings of polysemous words and avoid ambiguity in semantic understanding. This technology is the key to improving the intelligence of the conversational system and user experience.

[0050] Specifically, the natural language model is used to analyze the multiple word vectors, and the specific steps for obtaining the context vector are as follows: a natural language model is used to obtain a relationship weight matrix between the multiple word vectors, wherein the rows and columns of the relationship weight matrix are the multiple word vectors, and the elements in the relationship weight matrix are used to represent the relationship weights between each word vector; each word vector is multiplied by the relationship weight matrix respectively to obtain a context word vector in the context vector; the multiple word vectors are aggregated according to the sentences to which the word vectors belong to obtain a context sentence vector in the context vector to obtain user intent information.

[0051] It should be noted that there are many ways to aggregate multiple word vectors, such as: the accumulation method: directly accumulate the word vectors belonging to the same sentence, the weighted summation method: directly accumulate the word vectors belonging to the same sentence according to the weight of each word vector.

[0052] In some embodiments of the present application, generating answer data corresponding to the user intent information includes: identifying multiple entities from the user intent information, and determining the multiple entities as attributes of the user intent information, wherein the attributes of the user intent information include at least one of the following: time range, comparison object, and question type; filling the attributes of the user intent information into the preset slots to obtain the intent attributes; filling the slot values ​​in the preset slots into the slots of the query template to generate a query statement; and using the query statement to query the reply information. When identifying entities, a named entity recognition (NER) model or a part-of-speech tagging (POS tagging) tool can be used to analyze sentences to identify key entities or attributes related to intent, such as time, location, service type, etc. For example, when a user asks "What is the weather like in place A tomorrow?", entity recognition will find "tomorrow" (time) and "place A" (place) as key information.

[0053] Optionally, the query template can be determined by: designing one or more query templates based on the user's intent attributes. Each template is based on the structure of the knowledge graph and contains a basic query grammar framework. Reserve slot parameters: reserve slots in the query statement, represented by placeholders.

[0054] It should be noted that each slot represents a specific type of information, such as "time", "location", "service type", etc. For example, when processing intents related to weather queries, two slots, "time" and "location", may be defined. Time range: refers to the time-related information involved in the user's question. This can include the specific date, time period, a point in the past or future, etc. that the user asks. For example, the user may ask "How was my data usage last month?" or "How much will my package cost for the next six months?" When building queries and generating responses, the time range is used to limit or specify the time dimension of the data to ensure that the information returned by the system matches the time range of the user's query. This helps improve the accuracy of queries and the pertinence of responses. Comparison object: refers to the object that needs to be compared or referenced in the user's question. This can be a comparison between different products, services, solutions, or a comparison with the user's past records. For example, the user may ask "How is my package different from your premium package?" or "How does my current data usage compare to the same period last year?" The comparison object is used to determine the comparison benchmark for the query and help the system generate comparative responses so that users can better understand and evaluate the information. Question type: refers to the nature or category of the user's question, such as inquiries, complaints, suggestions, consultations, and requests for operation guides. Question types help the system understand the type of user needs, so as to select the most appropriate response strategy and information organization method. For example, if the question is of a complaint nature, the system may need to first express apologies and then provide a solution; if it is a request for an operation guide, the system may need to provide a clear step-by-step guidance text. Question type is an important factor that the system needs to consider when responding to users to ensure that the response is consistent with business logic and meets the specific needs of users. In intelligent customer service systems, these concepts (time range, comparison object, question type) are usually identified and filled as slot values ​​to guide the generation of query and response information in the knowledge graph. They are the basis for understanding user intent, constructing query statements, and generating personalized response texts, and are crucial to improving system accuracy and user satisfaction. By accurately identifying and processing these slot values, the system can provide more intimate and efficient services.

[0055] In actual application scenarios, user intent information can also be refined by sub-intents and related attributes. For example, users raise questions or demands to the system through various channels (such as text, voice). Initial intent recognition: Use NLP technology to analyze user input and identify the main intent (such as checking phone bills, business consultation). Sub-intent refinement: Based on the initial intent, further analyze the specific needs or actions that users may have. For example, checking phone bills can be further subdivided into "checking this month's phone bills", "checking past month's phone bills", etc.

[0056] In some embodiments of the present application, the query statement is used to query the reply information, including: sending the query statement to a preset knowledge graph database to perform a query operation; receiving question resolution data output by the preset knowledge graph database in response to the query operation, the question resolution data including at least: answer data corresponding to the question data; and generating reply information based on the question resolution data.

[0057] In some embodiments of the present application, reply information is generated based on the question resolution data, including: obtaining a complexity score and a preset narrative order of the user intent information; generating a reply template according to the complexity score and the preset narrative order of the user intent information; extracting key information from the question resolution data, the key information including at least: entities in the question resolution data; filling the key information into the reply template to obtain reply information.

[0058] Specifically, the complexity score of the user's intent can be determined based on whether the question data involves multiple sub-intents or requires cross-domain information. For example, a user may ask for package details and upgrade suggestions at the same time.

[0059] In another optional approach, you can also evaluate the richness of the question parsing data to determine whether it covers all aspects of the user's intent. If the data is incomplete or ambiguous, you may need to make further queries or request more information from the user. Understand the characteristics of the current interaction environment, such as user preferences, historical interaction records, contextual information, etc., to determine the degree of personalization and format of the response. Based on the above analysis, determine the level of detail and hierarchy of the response. For example, for complex intents, you may need to answer in steps, providing general information first, and then providing details based on user feedback.

[0060] The preset narrative order can be determined in advance. It is understood that in the process of information transmission or text generation, the information points are organized according to a certain logical order and correlation to form a smooth and organized overall narrative. In the intelligent customer service system, the construction of the preset narrative order is based on user intention and question parsing data, and the key information points are sorted and associated according to their inherent logical relationship and importance, ensuring that the system reply information not only contains all the necessary information, but also the order and way of presenting this information can be easily understood by the user to form a complete story or answer. For example, if the user asks about package upgrades, the coherent narrative structure may include the following steps: Confirm the current package details: First, the system needs to provide the user with basic information such as the type, price, and included services of the current package as the starting point of the conversation. List the upgrade options: Then, the system will provide the upgradeable package options according to the user's needs and the limitations of the current package, including their respective features and advantages. Compare the new and old packages: In this link, the system compares the new package with the user's current package, focusing on the additional services or cost savings of the new package, etc., to help users make decisions. Explain the upgrade process: The system will explain in detail how to upgrade the package, including any possible steps, required documents or operation instructions, to ensure that users understand the entire process. Answer potential questions: Based on the system's understanding of the user, predict and answer possible follow-up questions, such as bill changes after the upgrade, service effective time, etc., to provide a complete answer experience.

[0061] In actual application scenarios, the reply information can also be modified by obtaining sentiment words in the question data, for example: obtaining keywords in the question data used to characterize sentiment characteristics; using a preset sentiment analysis model to analyze the keywords to determine the emotional state of the target object; and modifying the reply information based on the emotional state of the target object to obtain the final reply information.

[0062] Figure 3 An intelligent customer service question-answering system is shown. Figure 3 As shown, it includes: a user interface module, which is used to receive user input request data and output the input request data to the natural language processing module and the knowledge management module;

[0063] The natural language understanding module is used to use the Word2Vec word vector model to perform semantic analysis on the input request data to obtain user intent information;

[0064] Specifically, the present embodiment also includes a data receiving submodule, which is used to receive user input request data, perform data cleaning on the user input request data, and obtain unified input request data, wherein the input request data at least includes communication text information, communication voice information, communication picture information, communication video information, phone bill recharge, business consultation and processing, and balance inquiry; a data segmentation submodule, which is used to segment the unified input request data according to linguistic rules to obtain word unit input data; a normalization submodule, which is used to perform word form restoration and part-of-speech tagging on the word unit input data, and finally perform normalization processing to obtain normalized vocabulary data; a vector conversion submodule, which is used to use a pre-trained Word2Vec word vector model to convert the normalized vocabulary data into a dense vector, and the dense vector is used to capture the semantic relationship between words; an intention analysis submodule, which is used to construct context associations between words for dense vectors based on an NLP natural language processing model through a self-attention mechanism, generate context-sensitive word vectors and sentence vectors, and obtain user intention information.

[0065] The knowledge management module is used to search for data based on user intent information based on knowledge graph technology to obtain problem analysis data;

[0066] Specifically, the present embodiment also includes an intent acquisition submodule, which is used to obtain user intent information, refine the user intent information into sub-intents and related attributes, identify and fill in slot values ​​related to the intent, and obtain user intent attributes, where the slot values ​​at least include a time range, a comparison object, and a question type; a template construction submodule, which is used to construct a query template for the knowledge graph based on the user intent information and user intent attributes, where the query template includes the basic framework of the SPARQL query statement and query parameters for reserved slots; a slot mapping submodule, which is used to map the slot values ​​in the user intent information to the slots corresponding to the query template to complete the concretization of the query statement; and an interface query submodule, which is used to send the query template to the knowledge graph query interface, and perform query operations on the knowledge graph database to obtain question resolution data.

[0067] Decision reasoning module, used to generate system response information based on user intent information and question analysis data;

[0068] Specifically, the present embodiment also includes an intent parsing unit, which is used to obtain the system response detail level and the system response hierarchy according to the complexity of the user's intent, the richness of the question parsing data and the characteristics of the interactive environment; a logical organization unit, which is used to organize the key information points in the question parsing data through logical reasoning to obtain a coherent narrative structure; a template generation unit, which is used to obtain a natural language generation template according to the coherent narrative structure, the system response hierarchy and the system response detail level; and a language calibration unit, which is used to use natural language processing tools to detect grammatical errors and correct spellings on the response text generated by the natural language generation template.

[0069] The sentiment analysis module is used to detect the sentiment of the input request data and the human-computer interaction history using the sentiment recognition model to obtain the user's emotional state;

[0070] Specifically, the present embodiment also includes a sentiment analysis unit, which is used to perform lexical sentiment tendency analysis, sentiment keyword recognition, sentiment phrase mining, and semantic role labeling on the input request data to extract sentiment features at the text level; an information input unit, which is used to input feature vectors into a sentiment recognition model to obtain sentiment category prediction results for the input request and the human-computer interaction history; and a sentiment recognition unit, which is used to identify whether there are mixed or contradictory sentiment signals, and to evaluate the user's emotional baseline level, common sentiment triggers, and coping strategies in combination with the long-term sentiment features in the human-computer interaction history to obtain the user's emotional state.

[0071] The multimodal interaction module is used to use the RAG search enhanced dialogue model to obtain the multi-information data input by the user and integrate the multi-information data into the human-computer interaction process.

[0072] Specifically, the present embodiment also includes a model building submodule, which is used to load a pre-trained RAG (Retrieval-Augmented Generation) search enhancement dialogue model and initialize the parameters of the RAG search enhancement dialogue model; an information reasoning submodule, which is used to drive the generation module of the RAG model to reason about the information input by the user and generate a preliminary answer or dialogue response to the user's question; a content analysis submodule, which is used to perform content analysis on the response generated by the RAG model, extract key information therein, and obtain analyzed interaction content; an interaction mode submodule, which is used to convert the analyzed interaction content into an output form suitable for different interaction modes according to the requirements of the interaction process update.

[0073] Its beneficial effects are that, through the user interface module, it is used to receive user input request data and output the input request data to the natural language processing module and the knowledge management module; the natural language understanding module is used to use the Word2Vec word vector model to perform semantic analysis on the input request data to obtain user intention information; the knowledge management module is used to search data based on user intention information based on the knowledge graph technology to obtain problem analysis data; the decision reasoning module is used to generate system reply information based on user intention information and problem analysis data; the sentiment analysis module is used to use the sentiment recognition model to perform sentiment detection on the input request data and the human-computer interaction history to obtain the user's emotional state; the multimodal interaction module is used to use the RAG search enhanced dialogue model to obtain the multivariate information data input by the user, and integrate the multivariate information data into the human-computer interaction process. It realizes real-time response to user questions in the field of mobile communications, efficient answers, improved customer experience, reduced costs, and enhanced business continuity.

[0074] Figure 4 A natural language understanding module is shown, including: a data receiving submodule, which is used to receive user input request data, perform data cleaning on the user input request data, and obtain unified input request data, wherein the input request data at least includes communication text information, communication voice information, communication picture information, communication video information, telephone charge recharge, business consultation and handling, and balance inquiry;

[0075] The data segmentation submodule is used to segment the unified input request data according to linguistic rules to obtain word unit input data;

[0076] The normalization submodule is used to perform word form restoration and part-of-speech tagging on the word unit input data, and finally perform normalization processing to obtain normalized vocabulary data;

[0077] The vector conversion submodule is used to convert normalized vocabulary data into dense vectors using the pre-trained Word2Vec word vector model. The dense vector is used to capture the semantic relationship between words.

[0078] The intent analysis submodule is used to build contextual associations between words for dense vectors based on the NLP natural language processing model through the self-attention mechanism, generate context-sensitive word vectors and sentence vectors, and obtain user intent information.

[0079] Figure 5 A decision reasoning module is shown, including: an intention acquisition submodule, which is used to obtain user intention information, refine the user intention information into sub-intentions and related attributes, identify and fill in slot values ​​related to the intention, and obtain user intention attributes, where the slot values ​​at least include a time range, a comparison object, and a question type;

[0080] The template construction submodule is used to construct a query template for the knowledge graph based on user intent information and user intent attributes. The query template includes the basic framework of the SPARQL query statement and query parameters with reserved slots.

[0081] The slot mapping submodule is used to map the slot value in the user intent information to the slot corresponding to the query template to complete the concretization of the query statement;

[0082] The interface query submodule is used to send the query template to the knowledge graph query interface and perform query operations on the knowledge graph database to obtain question resolution data.

[0083] Figure 6 An intelligent customer service question-and-answer device according to an embodiment of the present application includes:

[0084] A receiving module 60, for receiving question data input by a target object;

[0085] A word segmentation module 62, configured to perform word segmentation processing on the question data to obtain a plurality of word segments, and convert the plurality of word segments into a plurality of word vectors;

[0086] An analysis module 64 analyzes the multiple word vectors using a natural language model to obtain a context vector, wherein the context vector is used to represent a contextual relationship between the multiple word vectors;

[0087] The generating module 66 determines the user intention information of the question data based on the context vector, and generates answer information corresponding to the user intention information.

[0088] Through the above-mentioned intelligent customer service question and answer device, the question data input by the target object is received; the question data is segmented to obtain multiple word segments, and the multiple word segments are converted into multiple word vectors; the multiple word vectors are analyzed by a natural language model to obtain a context vector, wherein the context vector is used to represent the context relationship between the multiple word vectors; based on the context vector, the user intention information of the question data is determined, and the reply information corresponding to the user intention information is generated, thereby achieving the purpose of determining the user intention information according to the context association relationship of the question data, thereby achieving the technical effect of improving the accuracy of the acquired user intention information, and then solving the technical problem in the related technology that the dialogue robot cannot obtain the user intention, resulting in low reply accuracy.

[0089] The word segmentation module 62 includes: a word segmentation submodule, which is used to perform word segmentation processing on the problem data to obtain multiple word segmentations, and convert the multiple word segmentations into multiple word vectors, including: removing noise information in the problem data, performing word segmentation processing on the problem data to obtain multiple initial word segmentations, and performing part-of-speech restoration and part-of-speech tagging on the multiple initial word segmentations and then normalizing them to obtain the multiple word segmentations; converting the multiple word segmentations into multiple dense vectors, and determining the multiple dense vectors as the multiple word vectors, each dimension in the word vector is used to represent a semantic feature of the word segmentation.

[0090] The word segmentation submodule includes: a word segmentation unit, which is used to analyze the multiple word vectors using a natural language model to obtain a context vector, including: using a natural language model to obtain a relationship weight matrix between the multiple word vectors, wherein the rows and columns of the relationship weight matrix are the multiple word vectors, and the elements in the relationship weight matrix are used to represent the relationship weights between each word vector; multiplying each word vector by the relationship weight matrix respectively to obtain a context word vector in the context vector; aggregating the multiple word vectors according to the sentences to which the word vectors belong to obtain a context sentence vector in the context vector.

[0091] The generation module 66 includes: a generation submodule for generating answer data corresponding to the user intention information, including: identifying multiple entities from the user intention information, and determining the multiple entities as attributes of the user intention information, wherein the attributes of the user intention information include at least one of the following: time range, comparison object and question type; filling the attributes of the user intention information into preset slots to obtain intention attributes; filling the slot values ​​in the preset slots into the slots of the query template to generate a query statement; and using the query statement to query the reply information.

[0092] The generation submodule includes: a generation unit for querying the reply information using the query statement, including: sending the query statement to a preset knowledge graph database to perform a query operation; receiving question analysis data output by the preset knowledge graph database in response to the query operation, wherein the question analysis data at least includes: answer data corresponding to the question data; and generating reply information based on the question analysis data.

[0093] The generation unit includes: a reply unit, which is used to generate reply information based on the question resolution data, including: obtaining a complexity score and a preset narrative order of the user intention information; generating a reply template according to the complexity score and the preset narrative order of the user intention information; extracting key information from the question resolution data, the key information including at least: entities in the question resolution data; filling the key information into the reply template to obtain reply information.

[0094] The reply unit includes: an emotional sub-unit, which is used to obtain keywords in the question data for characterizing emotional characteristics; use a preset emotional analysis model to analyze the keywords to determine the emotional state of the target object; modify the reply information based on the emotional state of the target object to obtain final reply information.

[0095] It should be noted that Figure 6 The intelligent customer service question-answering device shown is used to perform Figure 2 The intelligent customer service question and answer method shown, therefore the relevant explanations in the above-mentioned intelligent customer service question and answer method are also applicable to the intelligent customer service question and answer device, and will not be repeated here.

[0096] An embodiment of the present application also provides a computer device, including: a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the above-mentioned intelligent customer service question and answer method.

[0097] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned intelligent customer service question and answer method by running the computer program.

[0098] An embodiment of the present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the intelligent customer service question and answer method in the present application.

[0099] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0100] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0102] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0103] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0104] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or all or part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0105] The above are only preferred implementations of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An intelligent customer service question and answer method, characterized in that: include: Receive question data input by the target object; Perform word segmentation on the question data to obtain multiple word segments, and convert the multiple word segments into multiple word vectors; Analyze the multiple word vectors using a natural language model to obtain a context vector, wherein the context vector is used to represent a contextual relationship between the multiple word vectors; The user intention information of the question data is determined based on the context vector, and answer information corresponding to the user intention information is generated.

2. The method according to claim 1, characterized in that The problem data is segmented to obtain a plurality of segmented words, and the plurality of segmented words are converted into a plurality of word vectors, including: removing noise information in the question data, performing word segmentation processing on the question data to obtain a plurality of initial word segmentations, and performing part-of-speech restoration and part-of-speech tagging on the plurality of initial word segmentations and performing normalization processing to obtain the plurality of word segmentations; The multiple word segments are converted into multiple dense vectors, and the multiple dense vectors are determined as the multiple word vectors, where each dimension in the word vector is used to represent a semantic feature of the word segment.

3. The method according to claim 2, characterized in that The multiple word vectors are analyzed using a natural language model to obtain a context vector, including: A natural language model is used to obtain a relationship weight matrix between the multiple word vectors, wherein the rows and columns of the relationship weight matrix are the multiple word vectors, and the elements in the relationship weight matrix are used to represent the relationship weights between each word vector; Multiply each word vector by the relationship weight matrix to obtain a context word vector in the context vector; The multiple word vectors are aggregated according to the sentences to which the word vectors belong, to obtain a context sentence vector in the context vector.

4. The method according to claim 1, characterized in that Generating answer data corresponding to the user intention information, including: Identifying a plurality of entities from the user intent information, and determining the plurality of entities as attributes of the user intent information, wherein the attributes of the user intent information include at least one of the following: a time range, a comparison object, and a question type; Fill the preset slots according to the attributes of the user intention information to obtain the intention attributes; Fill the slot value in the preset slot into the slot of the query template to generate a query statement; The query statement is used to query the reply information.

5. The method according to claim 4, characterized in that Using the query statement to query the reply information includes: Send the query statement to a preset knowledge graph database to perform a query operation; Receiving question analysis data output by the preset knowledge graph database in response to the query operation, wherein the question analysis data at least includes: answer data corresponding to the question data; Reply information is generated based on the question analysis data.

6. The method according to claim 5, characterized in that Generating answer information based on the question parsing data includes: Obtaining a complexity score and a preset narrative order of the user intention information; Generate a response template based on the complexity score of the user's intention information and a preset narrative order; Extract key information from the problem analysis data, the key information at least including: entities in the problem analysis data; Fill the key information into the reply template to obtain reply information.

7. The method according to claim 6, characterized in that The method further comprises: Obtaining keywords for representing emotional characteristics in the question data; Analyze the keywords using a preset sentiment analysis model to determine the sentiment state of the target object; The reply information is modified based on the emotional state of the target object to obtain final reply information.

8. An intelligent customer service question and answer device, characterized in that: include: A receiving module, used for receiving question data input by a target object; A word segmentation module, used to perform word segmentation processing on the question data to obtain multiple word segments, and convert the multiple word segments into multiple word vectors; An analysis module, configured to analyze the plurality of word vectors using a natural language model to obtain a context vector, wherein the context vector is used to represent a contextual relationship between the plurality of word vectors; A generation module is used to determine user intention information of the question data based on the context vector and generate answer information corresponding to the user intention information.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory is used to store program instructions; The processor is connected to the memory and is used to execute the intelligent customer service question and answer method described in any one of claims 1-7.

10. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by the processor, the intelligent customer service question and answer method described in any one of claims 1 to 7 is implemented.

Images