An artificial intelligence customer service system based on natural language processing

Abstract

The application discloses an artificial intelligence customer service system based on natural language processing and concretely relates to the technical field of natural language processing, and comprises an input processing module, an intention cooperation module, an intention optimization module and an intention dynamic decision module.The input processing module is used for accepting user voice and text input and performing pretreatment, calculating context entropy values based on historical N-round conversation feature distribution and time attenuation weight;the intention cooperation module is used for outputting emotion intensity signals through a multi-modal emotion recognition model according to input information, and generating an intention correction vector in combination with a preset intention-emotion mapping matrix;the intention optimization module is used for constructing and optimizing a cooperative loss function, performing intention classification on the input of the user based on optimized model parameters, and dynamically adjusting according to context entropy and emotion analysis; and the intention dynamic decision module is used for adjusting and deciding the response strategy of the system based on the output of the input processing module, the intention cooperation module and the optimization module and according to the context of the current conversation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, more specifically, the present application relates to an artificial intelligence customer service system based on natural language processing. BACKGROUND

[0002] With the rapid development of information technology, artificial intelligence is increasingly widely used in various fields, especially in the customer service industry, the introduction of artificial intelligence customer service system has become one of the key technologies to improve service efficiency and customer experience. The traditional customer service mode usually relies on manual processing, when facing a large number of customer inquiries, it is easy to cause slow service response, high error rate and excessive work pressure of employees, etc., which directly affects customer satisfaction and the operation efficiency of enterprises.

[0003] Natural language processing, as an important branch of artificial intelligence, aims to enable computers to understand and generate natural language. Through the processing of human language, natural language processing enables computers to interact with humans more naturally, and when applied to artificial intelligence customer service systems, it can greatly improve the intelligent level of the system, helping to achieve automated, efficient and accurate customer service. Natural language processing technology includes speech recognition, semantic understanding, sentiment analysis and dialogue management, which enables computers to understand customer needs, automatically generate appropriate responses, and provide accurate help. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an artificial intelligence customer service system based on natural language processing to solve the problems raised in the above background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme, an artificial intelligence customer service system based on natural language processing, comprising an input processing module, an intent coordination module, an intent optimization module, and an intent dynamic decision module;

[0006] The input processing module accepts voice and text input from the user and performs preprocessing, and calculates the context entropy value based on the historical N-turn dialogue feature distribution and the time decay weight;

[0007] The intent coordination module outputs an emotional intensity signal according to the input information through a multi-modal emotion recognition model, and generates an intent correction vector in combination with a pre-set intent-emotion mapping matrix;

[0008] The intent optimization module constructs and optimizes a collaborative loss function, including an original intent loss term, a context entropy weighted penalty term and an emotion distribution alignment term, based on the optimized model parameters, classifies the user's input according to the context entropy and emotion analysis, and dynamically adjusts according to the context entropy and emotion analysis;

[0009] The intention dynamic decision module is based on the output of the input processing, intention coordination and optimization module, and adjusts and decides the response strategy of the system in real time according to the context and user emotion of the current dialogue.

[0010] Preferably, as a preferred scheme of the artificial intelligence customer service system based on natural language processing, the input processing module accepts voice and text input of the user and performs preprocessing, calculates context entropy value based on historical N round dialogue feature distribution and time decay weight, and quantifies complexity and context continuity of the dialogue, specifically including:

[0011] Receiving voice and text information input by the user, converting voice input into text through automatic speech recognition, and performing text preprocessing through word segmentation and stop word removal to obtain a processed text sequence;

[0012] The state of each round of dialogue is represented as a vector , the complexity of the current dialogue is calculated based on the feature distribution of the historical N round dialogue, and the context entropy of the current dialogue is obtained, and the specific calculation formula of the context entropy is , wherein is the probability distribution of the historical dialogue round , and represents the context entropy, N is the round of the historical dialogue, and t is the text representation of the current dialogue, which is used to simulate the influence of time on the context information.

[0013] Preferably, as a preferred scheme of the artificial intelligence customer service system based on natural language processing, the intention coordination module outputs an emotion intensity signal through a multi-modal emotion recognition model according to the input information, and generates an intention correction vector combined with a preset intention-emotion mapping matrix, specifically including:

[0014] A multi-modal emotion recognition model is constructed, and the input voice and text are feature extracted, including acoustic features and emotional features, the features of different modalities are weighted and averaged to be fused to obtain a global emotion feature vector, and the fused feature vector is used for classification to obtain the probability distribution of the emotion category;

[0015] The input information including voice and text input is analyzed by the multi-modal emotion recognition model, and an emotion intensity signal is extracted, the emotion intensity signal representing the intensity of the current emotion of the user, different intensity values representing different emotion categories, and the emotion intensity signal vector is represented as , wherein represents the intensity value of the th emotion category, is the number of emotion categories, and is in the range of ; and

[0016] Setting an intention-emotion mapping matrix Each row of the matrix represents an emotion category, and each column represents an intention associated therewith, calculating an emotion intensity signal and the intention-emotion mapping matrix using matrix multiplication, generating an intention correction vector wherein, represents the influence of emotion u on intention v, represents the correction value of each intention, adjusting the emotional influence on the user's intention through the intention correction vector to achieve the coordinated adjustment of emotion and intention.

[0017] Preferably, as a preferred scheme of the artificial intelligence customer service system based on natural language processing, it comprises the step of constructing a multi-modal emotion recognition model, extracting features of input voice and text, including acoustic features and emotional features, weighting and averaging fusion of features of different modalities to obtain a global emotion feature vector, and using the fused feature vector for classification to obtain the probability distribution of emotion categories, further comprising:

[0018] The acoustic feature vector extracted from the voice is represented as The emotional feature vector extracted from the text is represented as The fused emotion feature vector is obtained as wherein, is the fused feature, and are weight coefficients representing the contribution of voice and text features to the emotion recognition result;

[0019] The fused emotion feature vector is input into an emotion classifier using a softmax function, and an emotion category probability distribution is output: wherein, , and , are the weights and biases of the classifier, is the predicted probability of the kth emotion category, and K is the total number of emotion categories.

[0020] Preferably, as a preferred scheme of the artificial intelligence customer service system based on natural language processing, it comprises the step of constructing and optimizing a collaborative loss function, including an original intention loss term, a context entropy weighted penalty term, and an emotion distribution alignment term, based on the optimized model parameters, classifying the user's input according to the context entropy and emotion analysis, and dynamically adjusting, specifically comprising:

[0021] Constructing a collaborative loss function, the specific formula is wherein, a loss term representing original intention classification, a context entropy weighted penalty term, an emotion distribution alignment term, and are hyperparameters controlling the weights of each loss term;

[0022] by optimizing the collaborative loss function , the optimal parameters of the model are obtained Based on the optimized model parameters, the user's input is classified, and the intention is dynamically adjusted according to the results of context entropy and emotion analysis, and the specific formula is wherein, is the probability of the intention category predicted by the model, represents the context entropy, and f is a function based on the context entropy and emotion analysis, is the prediction result after dynamic adjustment, is the result of emotion analysis.

[0023] Preferably, as a preferred scheme of the artificial intelligence customer service system based on natural language processing, the original intention loss term is measured by using cross-entropy loss to measure the difference between the predicted intention of the model and the true intention, and the intention label of the user input is represented as The probability distribution predicted by the model is represented as The original intention loss term is obtained as wherein, Q is the number of intention categories, is the true label of the qth intention, is the probability of the qth intention predicted by the model;

[0024] The context entropy weighted penalty term is used to measure the entropy value of the model under different contexts, and the output distribution of the model is limited by the weighted entropy penalty term, and the specific formula is wherein, is the probability distribution of intention q under a given context, is a weighting coefficient, representing the influence degree of context entropy on the model;

[0025] The emotion distribution alignment term is used to make the intention classification result of the model consistent with the emotion distribution, and the Kullback-Leibler divergence is used to measure the difference between the predicted emotion distribution and the true emotion distribution The specific formula is wherein, and are the true distribution and predicted distribution of emotion respectively, and K is the number of emotion categories.

[0026] Preferably, as a preferred solution of the artificial intelligence customer service system based on natural language processing, the intention dynamic decision module is based on the output of the input processing, intention coordination and optimization module, and adjusts and decides the response strategy of the system in real time according to the context and user emotion of the current dialogue, specifically including:

[0027] According to the dynamically adjusted intention prediction result, and combined with the context and user emotion of the current dialogue, the response strategy is determined, the response strategy is the best response behavior selected by the system based on the input intention, context and emotion, and the specific formula is , wherein, is the response option, is the matching degree between the intention category, the response option and the emotion analysis, is the weight of the response option, and the selected response strategy A is the response that maximizes the matching degree;

[0028] According to the feedback satisfaction of the user, self-adjustment is carried out, when the feedback satisfaction is negative, the model is further optimized by adjusting the loss function, the weight is updated to adapt to different dialogue contexts and user emotions, and further including:

[0029] Display feedback: the user clicks the "satisfied" and "dissatisfied" buttons to give feedback scores, and the feedback satisfaction of the user is , wherein, , represents complete satisfaction, represents neutral, represents complete dissatisfaction, and are the minimum value and the maximum value of the score respectively;

[0030] Implicit feedback: according to the frequency of the user continuing the dialogue, the stay time, the tone of the interaction and the attitude of the response, the emotion value of the user in the current dialogue is obtained;

[0031] The feedback satisfaction of the user is obtained by combining the explicit and implicit feedback, and self-adjustment is carried out, when the user feedback is negative, the loss function is adjusted and optimized to adapt to different dialogue contexts and user emotions.

[0032] On the other hand, the present application provides a computer device comprising a memory and a processor, the memory storing a computer program, wherein: the computer program is executed by the processor to realize the function modules of the artificial intelligence customer service system based on natural language processing as described above.

[0033] In another aspect, the present application provides a computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements an artificial intelligence customer service system based on natural language processing as described above.

[0034] In the above technical solution, the present application provides the following technical effects and advantages:

[0035] By accepting the voice and text input of the user and preprocessing, calculating the context entropy value based on the historical N round of conversation feature distribution and time decay weight, outputting the emotion intensity signal through the multi-modal emotion recognition model according to the input information, and combining the preset intention-emotion mapping matrix to generate the intention correction vector, a collaborative loss function is constructed and optimized, including the original intention loss term, the context entropy weighted penalty term and the emotion distribution alignment term, based on the optimized model parameters, the input of the user is classified, and the intention is dynamically adjusted according to the context entropy and emotion analysis, according to the context and user emotion of the current conversation, the response strategy of the system is adjusted and decided in real time, through the collaborative work of the input processing module, the intention collaborative module, the intention optimization module and the intention dynamic decision module, the service efficiency can be significantly improved, the customer experience can be optimized, and the enterprise operation cost can be reduced. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0037] Figure 1 The method flow chart of the artificial intelligence customer service system based on natural language processing of the present application.

[0038] Table 1 is a data record table of the simulation experiment of the artificial intelligence customer service system based on natural language processing of the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0040] In the description of the present application, the terms "first", "second" are only for descriptive purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0041] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or description". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given in order to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can recognize that the present application can be implemented without using these specific details. In other examples, well-known structures and processes will not be described in detail in order to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope consistent with the principles and features disclosed in the present application.

[0042] Embodiment 1

[0043] The present embodiment provides an artificial intelligence customer service system based on natural language processing as shown in Figure 1 which specifically comprises an input processing module, an intent coordination module, an intent optimization module, and an intent dynamic decision module;

[0044] The input processing module is to accept the voice and text input of the user and perform preprocessing, calculate the context entropy value based on the historical N round of conversation feature distribution and time decay weight;

[0045] The intent coordination module is to output the emotion intensity signal through the multi-modal emotion recognition model according to the input information, and generate the intent correction vector in combination with the preset intent-emotion mapping matrix;

[0046] The intent optimization module is to construct and optimize the coordination loss function, including the original intent loss term, the context entropy weighted penalty term and the emotion distribution alignment term, based on the optimized model parameters, to classify the input of the user, and dynamically adjust according to the context entropy and emotion analysis;

[0047] The intent dynamic decision module is based on the output of the input processing, intent coordination and optimization modules, and adjusts and decides the response strategy of the system in real time according to the context and user emotion of the current conversation.

[0048] In this embodiment, the input processing module needs to be specifically explained, which accepts the voice and text input of the user and performs preprocessing, calculates the context entropy value based on the historical N round conversation feature distribution and time decay weight, quantifies the complexity and context coherence of the conversation, and specifically includes:

[0049] The voice and text information input by the user are received, the voice input is converted into text through automatic speech recognition, and the text is preprocessed through word segmentation and stop word removal to obtain a processed text sequence, which can quickly understand the input of the user, extract key information including questions, emotions, and context, and greatly accelerate the response time of customer service;

[0050] The state of each round of conversation is represented as a vector The complexity of the current conversation is calculated based on the feature distribution of the historical N round conversation to obtain the context entropy of the current conversation, and the specific calculation formula of the context entropy is , wherein is the probability distribution of the historical conversation round , and represents the context entropy, N is the round of historical conversation, t is the text representation of the current conversation, and is used to simulate the influence of time on context information.

[0051] In this embodiment, the intent coordination module needs to be specifically explained, which outputs an emotion intensity signal through a multi-modal emotion recognition model according to the input information, and generates an intent correction vector combined with a preset intent-emotion mapping matrix, which can accurately identify the user's intent by integrating information from different sources, and specifically includes:

[0052] A multi-modal emotion recognition model is constructed to extract features from the input voice and text, including acoustic features and emotional features, to obtain a global emotion feature vector by weighted average fusion of features of different modalities, and to classify using the fused feature vector to obtain a probability distribution of emotion categories, further including:

[0053] The acoustic feature vector extracted from the voice is represented as , the emotional feature vector extracted from the text is represented as , and the fused emotional feature vector is , wherein is the fused feature, and are weight coefficients, representing the contribution of voice and text features to the emotion recognition result;

[0054] The fused emotional feature vector is input into an emotion classifier using a softmax function to output an emotion category probability distribution: , wherein, , and , These are the classifier's weights and biases, respectively. It is the predicted probability of the k-th emotion category, where K is the total number of emotion categories;

[0055] The input information, including voice and text input, is analyzed using a multimodal emotion recognition model to extract emotion intensity signals. These signals represent the intensity of the user's current emotion, with different intensity values ​​indicating different emotion categories. The emotion intensity signal vector is represented as follows: ,in, Indicates the first The intensity value of the emotion category. It is the number of emotion categories, in Within the range;

[0056] Set up an intention-emotion mapping matrix Each row of the matrix represents an emotion category, and each column represents an associated intention. Matrix multiplication is used to calculate the emotion intensity signal and the intention-emotion mapping matrix, generating an intention correction vector. ,in, This indicates the influence of emotion u on intention v. This represents the correction value for each intention. The intention correction vector is used to adjust the emotional impact on the user's intention, so as to achieve coordinated adjustment of emotion and intention.

[0057] In this embodiment, the intent optimization module is specifically described. This module constructs and optimizes a collaborative loss function, including an original intent loss term, a contextual entropy-weighted penalty term, and a sentiment distribution alignment term. Based on the optimized model parameters, it classifies the user's input intent and dynamically adjusts it according to contextual entropy and sentiment analysis. This allows for optimization of the response based on the user's historical data, preferences, and current needs, avoiding mechanical answers and increasing personalized experience. Specifically, it includes:

[0058] The specific formula for constructing the collaborative loss function is as follows: ,in, The loss term represents the original intent classification. This represents the context entropy weighted penalty term. Indicates the alignment of sentiment distribution. and It is a hyperparameter that controls the weights of various losses;

[0059] The original intent loss term uses cross-entropy loss to measure the difference between the model's predicted intent and the true intent, representing the user's input intent label as follows: The probability distribution predicted by the model is expressed as , the original intention loss item is obtained as , wherein Q is the number of intention categories, is the true label of the qth intention, is the probability of the qth intention predicted by the model;

[0060] The context entropy weighted penalty term is used to measure the entropy value of the model under different contexts, and the output distribution of the model is limited through the weighted entropy penalty term, and the specific formula is , wherein is the probability distribution of intention q under a given context, is a weighting coefficient, indicating the influence degree of context entropy on the model;

[0061] The emotion distribution alignment term is used to make the intention classification result of the model consistent with the emotion distribution, and the Kullback-Leibler divergence is used to measure the difference between the predicted emotion distribution and the true emotion distribution , and the specific formula is , wherein and are the true distribution and the predicted distribution of emotion respectively, and K is the number of emotion categories;

[0062] By optimizing the collaborative loss function , the optimal parameters of the model are obtained, based on the optimized model parameters, the input of the user is classified, and the intention is dynamically adjusted according to the results of the context entropy and the emotion analysis, and the specific formula is , wherein is the intention category probability predicted by the model, represents the context entropy, and f is a function based on the context entropy and the emotion analysis, is the predicted result after dynamic adjustment, is the result of emotion analysis.

[0063] In the embodiment, it is specifically necessary to explain the intention dynamic decision module, the intention dynamic decision module is based on the output of the input processing, intention collaboration and optimization module, and adjusts and decides the response strategy of the system in real time according to the context and user emotion of the current dialogue, extracts valuable data through the analysis of the user input, provides user behavior analysis and business insight for enterprises, improves the service quality and customer satisfaction, and specifically includes:

[0064] According to the intention prediction result after dynamic adjustment, and combining the context and user emotion of the current dialogue, the response strategy is decided, the response strategy is the best response behavior selected by the system based on the input intention, context and emotion, and the specific formula is , wherein, is the first response option, is the degree of matching between the intent category, the response option and the emotion analysis, is the weight of the response option, and the selected response strategy A is the response that maximizes the degree of matching;

[0065] Self-adjustment according to the feedback satisfaction of the user, when the feedback satisfaction is negative, further optimize the model by adjusting the loss function, update the weight to adapt to different dialogue context and user emotion, further comprising:

[0066] Display feedback: the user clicks the "satisfied" and "dissatisfied" buttons to give feedback scores, and the feedback satisfaction of the user is , wherein, , represents complete satisfaction, neutral, completely dissatisfied, and are the minimum and maximum values of the score respectively;

[0067] Implicit feedback: according to the frequency of the user continuing the dialogue, the stay time, the tone of the interaction, and the attitude of the response, the emotion value of the user in the current dialogue is obtained;

[0068] The feedback satisfaction of the user is obtained by combining the explicit and implicit feedback, and self-adjustment is performed, when the user feedback is negative, the loss function is adjusted and optimized to adapt to different dialogue context and user emotion.

[0069] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used for wired or wireless communication with external terminals, and the wireless communication can be realized through WIFI, operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0070] The embodiment also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to realize the function module of the artificial intelligence customer service system based on natural language processing proposed in the above embodiment; the storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk or an optical disk.

[0071] Embodiment 2

[0072] In another embodiment of the present application, an artificial intelligence customer service system based on natural language processing is provided, and a simulation experiment is performed to scientifically demonstrate the beneficial effects of the present application.

[0073] The present experiment aims to verify the effectiveness of the artificial intelligence customer service system based on natural language processing in real-time dynamic response and user emotion understanding, and to improve the understanding and response accuracy of the system to user needs through multi-modal input, emotion recognition, intent correction and optimization technology.

[0074] The simulation experiment steps are implemented according to the content of the artificial intelligence customer service system based on natural language processing provided in Embodiment 1, and the specific steps include,

[0075] The conversation collection frequency is set to 1000 Hz, and the collection time is 30 minutes;

[0076] A synchronous trigger line is set, and conversation data of 100 users is collected, including 20 users with different emotional states and 80 ordinary users;

[0077] Based on the historical conversation data, virtual voice text input data is synthesized for label classification, and an intent classification model is fitted;

[0078] The collected conversation data is divided into a short time window of 1 minute each;

[0079] The maximum context entropy value in the short time window is calculated, and this value is used for weight normalization of the conversation round.

[0080] The pre-processed text features are extracted, combined with the emotional intensity signal, and corrected by the intent coordination module;

[0081] Through the optimized collaborative loss function, the intent in each window is classified, and the classification result is output;

[0082] Based on the optimized model output, the system response strategy is dynamically adjusted to generate the optimal response behavior;

[0083] The accuracy of the response behavior is evaluated, and the system is self-adjusted according to the user feedback satisfaction.

[0084] The specific data of the above simulation experiment is:

[0085] Time / Minute User emotional state Input features (text / voice) Contextual entropy value Emotional intensity Intention classification Response strategy User satisfaction Warning threshold Response result (Yes / No) 0-5 Neutral "Hello, can you help me?" 0.12 0.05 Query goods Provide information 0.85 0.80 No 5-10 Happy "Thank you for your help!" 0.08 0.90 Feedback Praise response 0.90 0.80 No 10-15 Anxious "I need help, what's going on?" 0.14 0.70 Emergency problem Solve the problem 0.78 0.75 Yes 15-20 Unhappy "How slow are you?" 0.18 0.85 Complaint Apology response 0.60 0.75 Yes 20-25 Calm "Okay, thank you" 0.10 0.02 Greeting Simple response 0.95 0.80 No

[0086] Table 1

[0087] Experimental analysis:

[0088] By dynamically adjusting the optimized model of intent classification, the system can capture the small changes in user emotions in real time, and can more accurately determine the response strategy. When the user's emotion is anxious, the system preferentially selects the problem-solving strategy, improving the timeliness and relevance of the response;

[0089] The warning threshold of the system can effectively determine whether the user is in an unsatisfied emotional state through historical analysis of multiple rounds of dialogue and emotional intensity evaluation, and make response adjustments in advance;

[0090] The experimental results show that in the case of high user emotional intensity, the system can respond in time and mobilize appropriate strategies for processing, effectively improving user satisfaction.

[0091] The above only describes some exemplary embodiments of the present application in a descriptive manner, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without deviating from the spirit and scope of the present application. Therefore, the above figures and descriptions are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

Claims

1. An artificial intelligence customer service system based on natural language processing, characterized in that: It includes an input processing module, an intent coordination module, an intent optimization module, and an intent dynamic decision-making module; The input processing module accepts and preprocesses the user's voice and text input, and calculates the contextual entropy value based on the feature distribution of N rounds of historical dialogue and time decay weights. Specifically, it includes: It receives voice and text input from the user, converts the voice input into text through automatic speech recognition, and performs text preprocessing operations such as word segmentation and stop word removal to obtain the processed text sequence. The state of each round of dialogue is represented as a vector. Based on the feature distribution of N rounds of historical dialogue, the complexity of the current dialogue is calculated to obtain the contextual entropy of the current dialogue. The specific formula for calculating the contextual entropy is as follows: ,in, For the rounds of historical dialogue The probability distribution, The context entropy is represented by N, where N is the turn in the historical dialogue and t is the text representation of the current dialogue. The intent coordination module outputs an emotion intensity signal through a multimodal emotion recognition model based on the input information, and generates an intent correction vector by combining it with a preset intent-emotion mapping matrix. The intent optimization module constructs and optimizes a collaborative loss function, including an original intent loss term, a contextual entropy weighted penalty term, and a sentiment distribution alignment term. Based on the optimized model parameters, it classifies the user's input intent and dynamically adjusts it according to contextual entropy and sentiment analysis. The original intent loss term uses cross-entropy loss to measure the difference between the model's predicted intent and the true intent, representing the user's input intent label as follows: The probability distribution predicted by the model is expressed as The original intention loss term is obtained as follows: Where Q is the number of intent categories, The true label for the q-th type of intent. It is the probability of the q-th type of intent predicted by the model; The contextual entropy weighted penalty term is used to measure the entropy value of the model in different contexts. The weighted entropy penalty term constrains the output distribution of the model, and the specific formula is as follows: ,in, It is the probability distribution of intention q given the context. These are weighting coefficients; The sentiment distribution alignment term is used to ensure that the model's intent classification results are consistent with the sentiment distribution. The Kullback-Leibler divergence is used to measure the predicted sentiment distribution. Compared with the true distribution of emotions The difference between them is expressed by the following formula: ,in, and These are the actual distribution and the predicted distribution of emotions, respectively, where K is the number of emotion categories; The intent dynamic decision-making module is based on the output of the input processing, intent coordination and optimization modules, and adjusts and decides the system's response strategy in real time according to the context of the current dialogue and the user's emotions.

2. The artificial intelligence customer service system based on natural language processing according to claim 1, characterized in that: The intent coordination module outputs an emotion intensity signal based on the input information through a multimodal emotion recognition model, and generates an intent correction vector by combining it with a preset intent-emotion mapping matrix. Specifically, it includes: A multimodal emotion recognition model is constructed, which extracts features from the input speech and text, including acoustic features and emotional features. The features from different modalities are then fused by weighted average to obtain a global emotion feature vector. The fused feature vector is then used for classification to obtain the probability distribution of emotion categories. The input information is analyzed using a multimodal emotion recognition model to extract the emotion intensity signal, which represents the intensity of the user's current emotion. The emotion intensity signal vector is represented as follows: ,in, Indicates the first The intensity value of the emotion category. It is the number of emotion categories; Set up an intention-emotion mapping matrix Matrix multiplication is used to calculate the emotion intensity signal and the intention-emotion mapping matrix, generating an intention correction vector. ,in, This indicates the influence of emotion u on intention v. This represents the correction value for each intention.

3. The artificial intelligence customer service system based on natural language processing according to claim 2, characterized in that: The construction of the multimodal emotion recognition model involves extracting features from the input speech and text, including acoustic and emotional features. The features from different modalities are then weighted and averaged to obtain a global emotion feature vector. This fused feature vector is then used for classification to obtain the probability distribution of emotion categories. Further steps include: The acoustic feature vector extracted from speech is represented as The sentiment feature vector extracted from the text is represented as The fused emotion feature vector is obtained as follows ,in, It is a feature of fusion. and These are weighting coefficients; The fused emotion feature vector is input into the emotion classifier using the softmax function, and an emotion category probability distribution is output: ,in, , and , These are the classifier's weights and biases, respectively. It is the predicted probability of the k-th emotion category, where K is the total number of emotion categories.

4. The artificial intelligence customer service system based on natural language processing according to claim 1, characterized in that: The intent optimization module constructs and optimizes the collaborative loss function, including the original intent loss term, the contextual entropy weighted penalty term, and the sentiment distribution alignment term. Based on the optimized model parameters, it classifies the user's input intent and dynamically adjusts it according to contextual entropy and sentiment analysis. Specifically, it includes: Construct the collaborative loss function, the specific formula is as follows: ,in, The loss term represents the original intent classification. This represents the context entropy weighted penalty term. Indicates the alignment of sentiment distribution. and It is a hyperparameter that controls the weights of various losses; By optimizing the collaborative loss function To obtain the optimal parameters of the model. Based on the optimized model parameters, the user's input is classified into intents. The intents are then dynamically adjusted according to the results of contextual entropy and sentiment analysis. The specific formula is as follows: ,in, It is the probability of the intent category predicted by the model. Let f represent context entropy, which is a function based on context entropy and sentiment analysis. It is a dynamically adjusted prediction result. This is the result of sentiment analysis.

5. The artificial intelligence customer service system based on natural language processing according to claim 1, characterized in that: The intent dynamic decision-making module is based on the output of the input processing, intent coordination, and optimization modules, and adjusts and decides the system's response strategy in real time according to the context of the current dialogue and the user's emotions. Specifically, it includes: Based on the dynamically adjusted intent prediction results, and combined with the current dialogue context and user sentiment, a response strategy is determined, specifically using the following formula: ,in, It is the first One response option, It is the degree of matching between intent category, response options, and sentiment analysis. These are the weights of the response options; the selected response strategy A is the response that maximizes the matching degree. The model adjusts itself based on user feedback satisfaction. When the feedback satisfaction is negative, the model is further optimized by adjusting the loss function and updating the weights to adapt to different dialogue contexts and user emotions.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the functional modules of the artificial intelligence customer service system based on natural language processing as described in any one of claims 1 to 5.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the functional modules of the artificial intelligence customer service system based on natural language processing as described in any one of claims 1 to 5.

Images