Internet of Things information technology customer service method and system based on emotion perception

Abstract

The invention provides an Internet of Things information technology customer service method and system based on emotion perception, and relates to the technical field of information technology service, and the method comprises the following steps: obtaining real-time message data of a user and a customer service, and transmitting the real-time message data to a cloud server in real time through an Internet of Things platform; performing text analysis on the obtained real-time message data, and analyzing the current emotional state of the user through an emotional analysis method; analyzing the intention and purpose of the user according to the real-time message data, and generating reply content through a reply database according to the current emotional state of the user; and continuously monitoring interaction data between the user and the customer service, and dynamically adjusting reply content. Through the emotion perception technology, the emotional state of the user can be analyzed, so that the demand and the emotional state of the user are better understood, the reply content is adjusted according to the emotional state of the user, the psychological demand of the user can be closer, and more personalized, warm and intimate services are provided.

Description

Technical Field

[0001] The present invention relates to the field of information technology service technology, and in particular, to an Internet of Things information technology customer service method and system based on emotion perception. Background Art

[0002] As a core component of the IT field, the information technology service industry is committed to providing customers with comprehensive technical support and services including software development, hardware installation, and system maintenance. These services not only focus on solving existing technical problems, but also cover a wider range of dimensions such as consulting, management, operation and maintenance, and security, aiming to help users achieve their business goals and maximize value by improving the overall performance and efficiency of information technology systems. The core value of the information technology service industry lies in its ability to provide customized solutions for enterprises and organizations, enabling them to make full use of existing resources and the latest technological trends, so as to stand out in the market competition. With the rapid development of emerging technologies such as cloud computing, big data, artificial intelligence, and the Internet of Things, the information technology service industry is facing unprecedented opportunities and challenges. These technologies have not only changed the business operation model, but also greatly increased the demand for information technology services by enterprises. To this end, information technology service providers need to continuously update their service content and technical capabilities to adapt to the pace of technological progress and changes in market demand.

[0003] Customer service is a crucial area that aims to improve customer satisfaction and loyalty through effective communication and problem solving. These services are not limited to solving problems and questions encountered by customers, but also include providing product information, helping customers make purchasing decisions, and processing orders and returns. In the existing customer service methods, although technological advances have enabled customer service to exist and be provided to users in various forms, there are still some obvious defects: traditional customer service methods are usually difficult to capture and analyze the user's emotional state in real time, and there is no effective mechanism to identify the user's emotional changes during the interaction process, resulting in the service response may not match the user's actual emotional state, affecting the user experience.

[0004] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0005] In view of this, the present invention provides an Internet of Things information technology customer service method and system based on emotion perception to solve the above-mentioned problem of no effective mechanism to identify the emotional changes of users during the interaction process.

[0006] In order to solve the above problems, the specific technical solutions adopted by the present invention are as follows: According to one aspect of the present invention, an Internet of Things information technology customer service method based on emotion perception is provided, and the Internet of Things information technology customer service method based on emotion perception comprises the following steps: S1. Use IoT devices to obtain real-time message data between users and customer service, and transmit it to the cloud server in real time through the IoT platform. The real-time message data is the text information exchanged between users and customer service; S2. Perform text analysis on the acquired real-time message data, and analyze the user's current emotional state through sentiment analysis methods; S3, analyzing the user's intention and purpose according to the real-time message data, and generating reply content through the reply database according to the user's current emotional state; S4. Use the IoT platform to push the generated reply content back to the user's IoT device, continuously monitor the interaction data between the user and customer service, and dynamically adjust the reply content.

[0007] Preferably, performing text analysis on the acquired real-time message data and analyzing the user's current emotional state by an emotional analysis method comprises the following steps: S21, preprocessing the acquired real-time message data, the preprocessing including text cleaning, removal of stop words and word segmentation; S22, constructing an emotional topic model to analyze the preprocessed real-time message data, extracting the emotional topic distribution in the real-time message data, and establishing a text-emotion topic matrix; S23, based on the established text-emotional theme matrix, using the neighborhood rough set reduction algorithm to analyze the importance of the emotional theme, and simplifying the emotional theme according to the importance analysis result; S24. Analyze the user's current emotional state based on the simplified emotional themes using the emotional dictionary algorithm.

[0008] Preferably, constructing an emotional topic model to analyze the preprocessed real-time message data, extracting the emotional topic distribution in the real-time message data, and establishing a text-emotional topic matrix includes the following steps: S221, collecting historical sample message data, and analyzing the historical sample message data using a syntactic analysis method, and extracting emotional feature words and emotional expression words; S222, clustering the emotion feature words and emotion opinion words using a similarity measurement method, and filtering the clustering results using a mutual information method to obtain an emotion feature word set and an emotion opinion word set with semantic duplication eliminated; S223, constructing an emotion corpus according to the filtered emotion feature word set and emotion expression word set, and training a preset LDA topic model according to the emotion corpus to obtain an emotion topic model; S224. Analyze the preprocessed real-time message data according to the obtained emotional topic model, extract the emotional topic distribution, and establish a text-emotional topic matrix.

[0009] Preferably, clustering the emotion feature words and emotion opinion words by using a similarity measurement method, and filtering the clustering results by using a mutual information method to obtain an emotion feature word set and an emotion opinion word set with semantic duplication eliminated comprises the following steps: S2221, calculating the similarity between the emotion feature words and the emotion opinion words based on the semantic similarity and opinion similarity; S2222. Based on the similarity measurement result, cluster the emotion feature words and the emotion opinion words respectively using a clustering algorithm, and count the occurrence frequency of the words in each cluster; S2223, using the mutual information method to calculate the correlation between the words and the sentiment expression in each cluster, and screening out the clusters that meet the preset correlation threshold; S2224. Merge and optimize the filtered clustering results, eliminate semantically repeated emotion feature words and emotion opinion words, and obtain an emotion feature word set and an emotion opinion word set without semantic repetition.

[0010] Preferably, based on the established text-emotional theme matrix, using the neighborhood rough set reduction algorithm to analyze the importance of the emotional theme, and simplifying the emotional theme according to the importance analysis result includes the following steps: S231, converting the established text-emotion theme matrix into a theme decision engine, wherein the emotion theme in the theme decision engine is used as a condition attribute, each text is used as a decision rule, and the category to which each text belongs is used as a decision attribute; S232, using the neighborhood rough set method to calculate the positive domain of each emotional theme, and calculating the importance of each emotional theme according to the calculation result of the positive domain; S233, setting an importance threshold, and comparing the importance of each emotional topic with the set importance threshold, retaining emotional topics that meet the importance threshold to achieve simplification of the emotional topics.

[0011] Preferably, analyzing the user's current emotional state based on the simplified emotional themes using the emotional dictionary algorithm includes the following steps: S241. Construct a sentiment dictionary using the basic sentiment dictionary, degree adverb dictionary, negation dictionary, interjection dictionary, rhetorical interrogative dictionary and relational conjunction dictionary; S242, importing the simplified emotional theme into an emotional dictionary, identifying emotional vocabulary in the simplified emotional theme, where the emotional vocabulary includes basic emotional words, degree adverbs, negative words, interjections, rhetorical questions, and relational conjunctions; S243, assigning weights to the identified sentiment words in combination with semantic rules, and calculating the sentiment value of the sentiment topic; S244: Determine the user's emotional state based on the calculated emotion value.

[0012] Preferably, the calculation formula for calculating the sentiment value of the sentiment theme is: ; In the formula, V represents the sentiment value of the emotional topic; n represents the number of basic sentiment words; m represents the number of interjections; r represents the number of negative words; W i Represents the weight of the i-th basic sentiment word; N i represents the weight of the i-th interjection; D represents the weight of the degree adverb; X represents the negative word gating, if a negative word appears, the value is 1, otherwise the value is 0; Y represents the rhetorical question gate, if a rhetorical question appears, the value is 1, otherwise the value is 0; H represents the influence coefficient.

[0013] Preferably, analyzing the user's intention and purpose according to the real-time message data, and generating the reply content through the reply database according to the user's current emotional state includes the following steps: S31, collecting message text data with user intention annotations, and performing feature extraction on the message text data; S32, training the extracted features based on the intent recognition model of the RoBERTa pre-trained model, and identifying the user's intent purpose using the trained intent recognition model according to the extracted features; S33, selecting a reply template matching the user's intention from a reply database according to the user's intention; S34. Adjust the tone and content of the reply template according to the user's current emotional state to obtain the final reply content.

[0014] Preferably, according to the user's intended purpose, selecting a reply template matching the user's intended purpose from the reply database comprises the following steps: S331. Collect the user's intended questions and customer service response content through historical manual customer service records; S332. Use natural language processing technology to perform text analysis and semantic understanding on the intended questions and customer service response content, and extract key response information and intent labels; S333: Based on the reply key information, a reply template is constructed for each intent tag, and the reply template and the intent tag are associated to construct a reply database; S334, performing similarity matching between the user's intention purpose and the intention tags in the reply database; S335. According to the similarity matching result, a reply template associated with the intent tag with the highest similarity is selected as a reply template that matches the user's intent purpose.

[0015] According to another aspect of the present invention, there is provided an Internet of Things information technology customer service system based on emotion perception, the Internet of Things information technology customer service system based on emotion perception comprises: a data acquisition module, an emotion analysis module, a reply content generation module and a reply content push module, and the data acquisition module, the emotion analysis module, the reply content generation module and the reply content push module are sequentially connected; The data acquisition module is used to use the IoT device to obtain the real-time message data between the user and the customer service, and transmit it to the cloud server in real time through the IoT platform. The real-time message data is the text information exchanged between the user and the customer service; The sentiment analysis module is used to perform text analysis on the acquired real-time message data and analyze the user's current emotional state through sentiment analysis methods; A reply content generation module is used to analyze the user's intention and purpose based on the real-time message data, and generate reply content through the reply database according to the user's current emotional state; The reply content push module is used to push the generated reply content back to the user's IoT device using the IoT platform, continuously monitor the interaction data between the user and customer service, and dynamically adjust the reply content.

[0016] The beneficial effects of the present invention are: 1. The present invention uses emotion perception technology to analyze the user's emotional state, so as to better understand the user's needs and emotional state, and adjust the reply content according to the user's emotional state, which can be closer to the user's psychological needs and provide more personalized, warm and considerate services, thereby improving customer satisfaction. The Internet of Things devices and cloud servers are used to realize the automation of customer service, and the user's emotions can be perceived and service responses can be made without human intervention. This can save labor costs, improve service efficiency, and realize the automation and intelligence of customer service.

[0017] 2. The present invention can accurately identify the user's emotional state and intention by performing text analysis and sentiment analysis on real-time message data, which helps to better understand the user's needs and emotional state and provide an accurate basis for subsequent reply content generation. By establishing a text-emotion theme matrix and simplifying the emotional theme, the user's emotional theme can be grasped more accurately, so that targeted reply content can be generated based on the user's current emotional state using the sentiment dictionary algorithm. By analyzing the importance of emotional themes and calculating sentiment values, topics and emotional tendencies that users are more concerned about can be discovered, which helps to continuously improve its own intelligence level and service quality, and better meet user needs and expectations.

[0018] 3. The present invention can establish reply templates containing various intentions by analyzing historical manual customer service records and intent questions, and use natural language processing technology to perform text analysis and semantic understanding, and can extract key reply information and intent labels, thereby constructing a reply template that matches the user's intent, which helps to accurately select replies that meet the user's intent from a large number of reply templates. By selecting a matching reply template based on the user's intent and purpose, personalized reply content can be generated to meet the personalized requirements of users with different needs, which helps to improve user experience and enhance user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 is a flow chart of an Internet of Things information technology customer service method based on emotion perception according to an embodiment of the present invention; Figure 2 It is a principle block diagram of an Internet of Things information technology customer service system based on emotion perception according to an embodiment of the present invention.

[0020] In the figure: 1. Data acquisition module; 2. Sentiment analysis module; 3. Reply content generation module; 4. Reply content push module. DETAILED DESCRIPTION

[0021] To enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this application.

[0022] According to an embodiment of the present invention, there is provided an Internet of Things information technology customer service method and system based on emotion perception.

[0023] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. As Figure 1 shown, according to an embodiment of the present invention, there is provided an Internet of Things information technology customer service method based on emotion perception. The Internet of Things information technology customer service method based on emotion perception includes the following steps: S1. Use Internet of Things devices to obtain real-time message data between users and customer service, and transmit it to the cloud server in real time through the Internet of Things platform. The real-time message data is the text information exchanged between users and customer service. Specifically, the Internet of Things devices include smartphones, smart computers, etc.; when a user communicates with customer service, the Internet of Things devices can capture real-time message data and transmit it to the cloud server in real time through the Internet of Things platform. Among them, the Internet of Things platform provides a data transmission channel and protocol to ensure that data can be safely and stably transmitted to the specified cloud server; usually, the Internet of Things platform will provide various connection options, such as Wi-Fi, cellular network, LoRaWAN, etc., to meet the needs of different devices and scenarios.

[0024] S2. Perform text analysis on the obtained real-time message data, and analyze the current emotional state of the user through an emotion analysis method. As a preferred embodiment, performing text analysis on the obtained real-time message data and analyzing the current emotional state of the user through an emotion analysis method includes the following steps: S21. Preprocess the obtained real-time message data. The preprocessing includes text cleaning, stop word removal, and word segmentation. It should be noted that the real-time message data may contain various noises and unnecessary contents, such as HTML tags, special characters, punctuation marks, etc. In the text cleaning stage, these irrelevant contents need to be removed, and only the text information is retained. Stop words refer to words that frequently appear in the text but do not make a substantial contribution to text analysis, such as "of", "is", "in", etc.; in text analysis, these stop words are usually removed from the text to reduce the data volume and improve the analysis efficiency and accuracy.

[0025] Word segmentation is the process of dividing text data into words or phrases with practical meaning. Word segmentation divides text data into a sequence of words or phrases, providing a basis for subsequent analysis and other tasks.

[0026] S22, constructing an emotional topic model to analyze the preprocessed real-time message data, extracting the emotional topic distribution in the real-time message data, and establishing a text-emotion topic matrix; As a preferred implementation, constructing an emotional theme model to analyze the preprocessed real-time message data, and extracting the emotional theme distribution in the real-time message data, and establishing a text-emotion theme matrix includes the following steps: S221, collecting historical sample message data, and analyzing the historical sample message data using a syntactic analysis method, and extracting emotional feature words and emotional expression words; Specifically, using syntactic analysis to analyze historical sample message data and extracting emotional feature words and emotional expression words includes the following steps: Segment the historical sample message data into a sequence of individual sentences using a sentence segmenter; Split each sentence into words or sequences of words, and mark each word with its part of speech, that is, the grammatical role of the word in the sentence. Part of speech tagging can help identify the subject, predicate, object and other components in the sentence; Use dependency syntactic analysis to analyze the dependency relationship between words in a sentence, that is, the subordinate relationship between words, and conduct semantic analysis based on syntactic analysis, that is, understand the meaning and implications of the sentence; Based on semantic analysis, the emotional feature words and emotional expression words in the text are identified. Emotional feature words are usually keywords that express feelings and emotions, such as happy, angry, sad, etc. Emotional expression words are words used to express feelings and emotions, such as like, angry, sad, etc.

[0027] S222, clustering the emotion feature words and emotion opinion words using a similarity measurement method, and filtering the clustering results using a mutual information method to obtain an emotion feature word set and an emotion opinion word set with semantic duplication eliminated; As a preferred implementation, the emotion feature words and emotion opinion words are clustered using a similarity measurement method, and the clustering results are filtered using a mutual information method to obtain an emotion feature word set and an emotion opinion word set that eliminate semantic duplication, including the following steps: S2221, calculating the similarity between the emotion feature words and the emotion opinion words based on the semantic similarity and opinion similarity; It should be noted that by using knowledge bases such as Hownet, the semantic relationship and path distance between emotion feature words and emotion opinion words are calculated to obtain the semantic similarity between them; When calculating the similarity between emotion feature words and emotion opinion words, in addition to considering semantic similarity, it is also necessary to consider the opinion similarity between them. The opinion similarity can be calculated based on the similarity of words in emotion polarity, that is, whether the polarity of emotion words is similar. For example, if both words express positive emotions or both express negative emotions, their opinion similarity is high; if one word expresses positive emotions and the other word expresses negative emotions, their opinion similarity is low.

[0028] Finally, by comprehensively considering the semantic similarity and opinion similarity, a weighted summation of the semantic similarity and opinion similarity is performed to obtain the final similarity value.

[0029] S2222. Based on the similarity measurement result, cluster the emotion feature words and the emotion opinion words respectively using a clustering algorithm, and count the occurrence frequency of the words in each cluster; It should be noted that, by constructing a similarity matrix based on the similarity measurement results between the calculated emotion feature words and emotion opinion words, and then clustering the emotion feature words and emotion opinion words using the hierarchical clustering method, the emotion feature words and emotion opinion words can be assigned to different clusters, and by traversing each cluster and calculating the number of occurrences of each word therein as the frequency of occurrence, finally the degree of correlation between the words in each cluster and the emotional expression is determined through correlation analysis technology.

[0030] S2223, using the mutual information method to calculate the correlation between the words and the sentiment expression in each cluster, and screening out the clusters that meet the preset correlation threshold; It should be noted that mutual information is an indicator used to measure the correlation between two random variables. It reflects the amount of information contained in one random variable about another random variable.

[0031] For each word in each cluster, the mutual information between it and the emotional expression is calculated, and then the words are sorted according to the mutual information value with the emotional expression. According to the preset correlation threshold, the cluster containing the words whose mutual information value is greater than or equal to the threshold is filtered out.

[0032] S2224. Merge and optimize the filtered clustering results, eliminate semantically repeated emotion feature words and emotion opinion words, and obtain an emotion feature word set and an emotion opinion word set without semantic repetition.

[0033] It should be noted that, for the merged clusters, it is detected whether there are repeated emotional feature words and emotional opinion words. If there are repeated ones, only one of them is retained to eliminate the semantic duplication.

[0034] S223, constructing an emotion corpus according to the filtered emotion feature word set and emotion expression word set, and training a preset LDA topic model according to the emotion corpus to obtain an emotion topic model; It should be noted that the emotion corpus is constructed by taking the filtered emotion feature word set and emotion expression word set as the vocabulary, and then each vocabulary sample is represented as a bag-of-words model, and the bag-of-words model is used as the input of the LDA model to realize the training of the LDA topic model, and finally the emotion topic model is obtained.

[0035] S224. Analyze the preprocessed real-time message data according to the obtained emotional topic model, extract the emotional topic distribution, and establish a text-emotional topic matrix.

[0036] Specifically, the preprocessed real-time message data is converted into a bag-of-words model and loaded into the sentiment topic model, which can capture different sentiment topics and extract the sentiment topic distribution of each message. The matrix is ​​initialized according to the number of topics in the sentiment topic model. The rows of the matrix represent different texts (real-time messages) and the columns represent different sentiment topics. The matrix is ​​filled according to the sentiment topic distribution of each message.

[0037] S23, based on the established text-emotional theme matrix, using the neighborhood rough set reduction algorithm to analyze the importance of the emotional theme, and simplifying the emotional theme according to the importance analysis result; As a preferred implementation, based on the established text-emotional theme matrix, the importance of the emotional theme is analyzed using a neighborhood rough set reduction algorithm, and the emotional theme is simplified according to the importance analysis result, including the following steps: S231, converting the established text-emotion theme matrix into a theme decision engine, wherein the emotion theme in the theme decision engine is used as a condition attribute, each text is used as a decision rule, and the category to which each text belongs is used as a decision attribute; It should be noted that the topic decision engine is also called the topic decision system. In the topic decision system, the topic is regarded as a conditional attribute, each text in the corpus is regarded as a decision rule, and the category to which each text belongs is regarded as a decision attribute. It can be seen that different topics have different importance for texts of different categories. The relative simplification of the knowledge of the topic decision system refers to deleting redundant topics while keeping the dependency relationship between text topics and text categories in the original topic decision table unchanged.

[0038] Sentiment topics refer to the main emotions or sentiment categories identified in sentiment analysis, while text refers to the original message data from users or documents. Sentiment topics can be defined according to the needs of sentiment analysis, such as positive, negative, neutral, etc.

[0039] S232, using the neighborhood rough set method to calculate the positive domain of each emotional theme, and calculating the importance of each emotional theme according to the calculation result of the positive domain; It should be noted that in the neighborhood rough set theory, the positive domain refers to the set of objects that can be accurately classified. For each emotional theme, its positive domain is calculated, that is, the set of texts that can be clearly classified into specific emotions is found. The importance of an emotional theme refers to the degree of role of the theme in the classification or decision-making process. By analyzing the impact of each emotional theme on text classification, the importance of each emotional theme can be calculated.

[0040] Specifically, the calculation formula for importance is: ; In the formula, G represents the importance of each emotional topic; Indicates that the emotional theme TC K The size of the positive domain in the emotional theme set RED; It represents the size of the positive domain under the initial emotional theme set RED; D represents the decision attribute; S233, setting an importance threshold, and comparing the importance of each emotional topic with the set importance threshold, retaining emotional topics that meet the importance threshold to achieve simplification of the emotional topics.

[0041] S24. Analyze the user's current emotional state based on the simplified emotional themes using the emotional dictionary algorithm.

[0042] As a preferred implementation, according to the simplified emotional theme, using the emotional dictionary algorithm, analyzing the user's current emotional state includes the following steps: S241. Construct a sentiment dictionary using the basic sentiment dictionary, degree adverb dictionary, negation dictionary, interjection dictionary, rhetorical interrogative dictionary and relational conjunction dictionary; S242, importing the simplified emotional theme into an emotional dictionary, identifying emotional vocabulary in the simplified emotional theme, where the emotional vocabulary includes basic emotional words, degree adverbs, negative words, interjections, rhetorical questions, and relational conjunctions; It should be noted that by traversing the words in the sentiment dictionary and matching them with the simplified sentiment topics, if a sentiment word appears in the sentiment topic, it is considered to be a sentiment word.

[0043] S243, assigning weights to the identified sentiment words in combination with semantic rules, and calculating the sentiment value of the sentiment topic; It should be noted that a set of semantic rules are formulated to determine the weight of each sentiment word. These rules include the sentiment polarity, sentiment intensity, and association with other sentiment words of the word. According to the defined semantic rules, each sentiment word is assigned a corresponding weight. These weights can be numerical values ​​representing the intensity or degree of the emotion. For example, positive sentiment words have positive weights, while negative sentiment words have negative weights.

[0044] Specifically, the calculation formula for calculating the sentiment value of the emotional topic is: ; In the formula, V represents the sentiment value of the emotional topic; n represents the number of basic sentiment words; m represents the number of interjections; r represents the number of negative words; W i Represents the weight of the i-th basic sentiment word; N i represents the weight of the i-th interjection; D represents the weight of the degree adverb; X represents the negative word gating, if a negative word appears, the value is 1, otherwise the value is 0; Y represents the rhetorical question gate, if a rhetorical question appears, the value is 1, otherwise the value is 0; H represents the influence coefficient.

[0045] S244: Determine the user's emotional state based on the calculated emotion value.

[0046] It should be noted that judging the user's emotional state according to the calculated emotional value includes the following steps: Divide the range of sentiment values ​​into several intervals, such as negative values ​​for negative sentiment, positive values ​​for positive sentiment, and values ​​close to zero for neutral sentiment; Set the threshold of the sentiment value according to the actual situation. For example, a value greater than 0.5 indicates positive sentiment, a value less than -0.5 indicates negative sentiment, and a value between -0.5 and 0.5 indicates neutral sentiment. S3, analyzing the user's intention and purpose according to the real-time message data, and generating reply content through the reply database according to the user's current emotional state; As a preferred implementation, analyzing the user's intention and purpose according to the real-time message data, and generating the reply content through the reply database according to the user's current emotional state includes the following steps: S31, collecting message text data with user intention annotations, and performing feature extraction on the message text data; Specifically, feature extraction is the process of converting text data into feature representations that can be used for model training and analysis.

[0047] S32, training the extracted features based on the intent recognition model of the RoBERTa pre-trained model, and identifying the user's intent purpose using the trained intent recognition model according to the extracted features; It should be noted that by using the aforementioned extracted features as training set input, using the deep learning framework to load the RoBERTa pre-trained model, and building an intent recognition model on top of the RoBERTa model, it usually includes adding additional fully connected layers or other appropriate structures to adapt to specific intent recognition tasks. The output of this layer is usually a score for classifying user intent. The intent recognition model is then trained using the prepared training set data. During the training process, appropriate loss functions (such as cross entropy loss) and optimizers (such as Adam) are used to optimize the model parameters. Finally, the user's real-time message data is input into the trained intent recognition model to obtain the corresponding intent label or probability distribution, thereby realizing the recognition of the user's intent purpose.

[0048] S33, selecting a reply template matching the user's intention from a reply database according to the user's intention; As a preferred embodiment, according to the user's intended purpose, selecting a reply template matching the user's intended purpose from a reply database comprises the following steps: S331. Collect the user's intended questions and customer service response content through historical manual customer service records; Specifically, obtain historical manual customer service records, and then use artificial intelligence assistants or customer service personnel to classify and label the intent of each conversation, and classify the questions raised by users and the customer service replies into different intent categories.

[0049] S332. Use natural language processing technology to perform text analysis and semantic understanding on the intended questions and customer service response content, and extract key response information and intent labels; Specifically, using natural language processing technology to perform text analysis and semantic understanding on the intended questions and customer service response content, and extracting key response information and intent labels includes the following steps: Segment the intended questions and customer service responses into sequences of words or phrases, mark the part of speech of each word, and analyze the grammatical structure of the sentence; Identify named entities in text and the relationships between entities in text; Classify intent questions using a naive Bayes classification model to determine the intent category each question belongs to; Perform semantic analysis on customer service replies to understand whether the customer service replies match the user's questions; Identify keywords and phrases in the text and extract key information related to the intent. Keywords and phrases can reflect the theme or content of the text. Key information includes service name, service requirements, etc. to help answer questions or perform follow-up operations.

[0050] S333: Based on the reply key information, a reply template is constructed for each intent tag, and the reply template and the intent tag are associated to construct a reply database; It should be noted that based on the reply key information, a corresponding reply template is designed for each intent tag. The reply template contains general information that can answer specific intents. Each reply template is associated with the corresponding intent tag so that an appropriate reply can be selected according to the user's intention. All associated reply templates and intent tags are stored in a database to obtain a reply database.

[0051] S334, performing similarity matching between the user's intention purpose and the intention tags in the reply database; Specifically, the cosine similarity calculation method is used to compare the similarity between the intent purpose proposed by the user and the intent label in each reply database, and a similarity score list is generated, which shows the degree of similarity between each intent label and the user's intent.

[0052] S335. According to the similarity matching result, a reply template associated with the intent tag with the highest similarity is selected as a reply template that matches the user's intent purpose.

[0053] S34. Adjust the tone and content of the reply template according to the user's current emotional state to obtain the final reply content.

[0054] Specifically, the tone and content of the reply template are adjusted according to the user's current emotional state, and the final reply content includes: Fine-tune the tone of your reply based on the user’s emotional state. For example, for an angry user, the tone of your reply might need to be gentler and more understanding; for a happy user, the reply could be more lively and positive.

[0055] For users who display positive emotions, reinforce those positive emotions by responding, such as with words of praise or encouragement.

[0056] For users who display negative emotions, adopt strategies to alleviate such emotions, such as providing specific help options, expressing sympathy, or offering discounts.

[0057] S4. Use the IoT platform to push the generated reply content back to the user's IoT device, continuously monitor the interaction data between the user and customer service, and dynamically adjust the reply content.

[0058] Specifically, using the IoT platform to push the generated reply content back to the user's IoT device, and continuously monitoring the interaction data between the user and customer service, dynamically adjusting the reply content includes the following steps: Continuously collect interaction data between users and customer service, and use real-time monitoring of user emotional changes to analyze user reactions to replies; Based on the collected interaction data and user feedback, analyze the effect of the reply content and determine whether adjustments are needed; Use machine learning and automation rules to dynamically adjust the tone and content of reply templates based on user feedback and emotional changes. Through continuous monitoring and adjustment, continuously optimize the reply content to improve user satisfaction and interaction efficiency.

[0059] like Figure 2 As shown, according to another embodiment of the present invention, an Internet of Things information technology customer service system based on emotion perception is provided, and the Internet of Things information technology customer service system based on emotion perception includes: a data acquisition module 1, an emotion analysis module 2, a reply content generation module 3 and a reply content push module 4, and the data acquisition module 1, the emotion analysis module 2, the reply content generation module 3 and the reply content push module 4 are connected in sequence; Data acquisition module 1 is used to use IoT devices to acquire real-time message data between users and customer service, and transmit it to the cloud server in real time through the IoT platform. The real-time message data is the text information exchanged between users and customer service; Emotion analysis module 2, used to perform text analysis on the acquired real-time message data and analyze the user's current emotional state through the emotion analysis method; The reply content generation module 3 is used to analyze the user's intention and purpose according to the real-time message data, and generate reply content through the reply database according to the user's current emotional state; The reply content push module 4 is used to push the generated reply content back to the user's IoT device using the IoT platform, and continuously monitor the interaction data between the user and the customer service to dynamically adjust the reply content.

[0060] To sum up, with the help of the above-mentioned technical scheme of the present invention, the present invention can analyze the emotional state of the user through emotion perception technology, so as to better understand the user's needs and emotional state, and adjust the reply content according to the user's emotional state, which can be closer to the user's psychological needs, and provide more personalized, warm and considerate services, thereby improving customer satisfaction. The Internet of Things devices and cloud servers are used to realize the automation of customer service, and the user's emotions can be perceived and service responses can be performed without human intervention. This can save labor costs, improve service efficiency, and realize the automation and intelligence of customer service; the present invention can accurately identify the user's emotional state and intention purpose by performing text analysis and emotion analysis on real-time message data, which helps to better understand the user's needs and emotional state, and provide an accurate basis for subsequent reply content generation. After establishing a text-emotion theme matrix and simplifying the emotion theme, it can be more accurate. Accurately grasp the user's emotional theme, so that according to the user's current emotional state, targeted reply content can be generated using the emotional dictionary algorithm. Through the importance analysis of emotional themes and the calculation of emotional values, topics and emotional tendencies that users are more concerned about can be discovered, which is helpful to continuously improve its own intelligence level and service quality, and better meet the needs and expectations of users; the present invention can establish a reply template containing various intentions by analyzing historical manual customer service records and intention questions, and use natural language processing technology to perform text analysis and semantic understanding, and can extract reply key information and intent tags, so as to construct a reply template that matches the user's intention, which is helpful to accurately select replies that meet the user's intention from a large number of reply templates. By selecting a matching reply template according to the user's intention and purpose, personalized reply content can be generated to meet the personalized requirements of users with different needs, which is helpful to improve user experience and enhance user satisfaction.

[0061] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program code.

[0062] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An Internet of Things information technology customer service method based on emotion perception, characterized in that: The Internet of Things information technology customer service method based on emotion perception includes the following steps: S1. Use IoT devices to obtain real-time message data between users and customer service, and transmit it to the cloud server in real time through the IoT platform. The real-time message data is text information exchanged between users and customer service; S2. Perform text analysis on the acquired real-time message data, and analyze the user's current emotional state through sentiment analysis methods; S3, analyzing the user's intention and purpose according to the real-time message data, and generating reply content through the reply database according to the user's current emotional state; S4. Use the IoT platform to push the generated reply content back to the user's IoT device, continuously monitor the interaction data between the user and customer service, and dynamically adjust the reply content.

2. According to claim 1, the method for Internet of Things information technology customer service based on emotion perception is characterized in that: The step of performing text analysis on the acquired real-time message data and analyzing the user's current emotional state by using an emotional analysis method comprises the following steps: S21, preprocessing the acquired real-time message data, wherein the preprocessing includes text cleaning, stop word removal and word segmentation processing; S22, constructing an emotional topic model to analyze the preprocessed real-time message data, extracting the emotional topic distribution in the real-time message data, and establishing a text-emotion topic matrix; S23, based on the established text-emotional theme matrix, using the neighborhood rough set reduction algorithm to analyze the importance of the emotional theme, and simplifying the emotional theme according to the importance analysis result; S24. Analyze the user's current emotional state based on the simplified emotional themes using the emotional dictionary algorithm.

3. The method for Internet of Things information technology customer service based on emotion perception according to claim 2 is characterized in that: The construction of the emotional topic model analyzes the preprocessed real-time message data, extracts the emotional topic distribution in the real-time message data, and establishes the text-emotion topic matrix, including the following steps: S221, collecting historical sample message data, and analyzing the historical sample message data using a syntactic analysis method, and extracting emotional feature words and emotional expression words; S222, clustering the emotion feature words and emotion opinion words using a similarity measurement method, and filtering the clustering results using a mutual information method to obtain an emotion feature word set and an emotion opinion word set with semantic duplication eliminated; S223, constructing an emotion corpus according to the filtered emotion feature word set and emotion expression word set, and training a preset LDA topic model according to the emotion corpus to obtain an emotion topic model; S224. Analyze the preprocessed real-time message data according to the obtained emotional topic model, extract the emotional topic distribution, and establish a text-emotional topic matrix.

4. The method for Internet of Things information technology customer service based on emotion perception according to claim 3 is characterized in that: The method of clustering the emotion feature words and emotion opinion words by using the similarity measurement method, and filtering the clustering results by using the mutual information method to obtain the emotion feature word set and emotion opinion word set with semantic duplication eliminated includes the following steps: S2221, calculating the similarity between the emotion feature words and the emotion opinion words based on the semantic similarity and opinion similarity; S2222. Based on the similarity measurement result, cluster the emotion feature words and the emotion opinion words respectively using a clustering algorithm, and count the occurrence frequency of the words in each cluster; S2223, using the mutual information method to calculate the correlation between the words and the sentiment expression in each cluster, and screening out the clusters that meet the preset correlation threshold; S2224. Merge and optimize the filtered clustering results, eliminate semantically repeated emotion feature words and emotion opinion words, and obtain an emotion feature word set and an emotion opinion word set without semantic repetition.

5. The method of Internet of Things information technology customer service based on emotion perception according to claim 2 is characterized in that: The method of analyzing the importance of emotional topics based on the established text-emotional topic matrix by using a neighborhood rough set reduction algorithm and simplifying the emotional topics according to the importance analysis results includes the following steps: S231, converting the established text-emotion theme matrix into a theme decision engine, wherein the emotion theme is used as a conditional attribute, each text is used as a decision rule, and the category to which each text belongs is used as a decision attribute; S232, using the neighborhood rough set method to calculate the positive domain of each emotional theme, and calculating the importance of each emotional theme according to the calculation result of the positive domain; S233, setting an importance threshold, and comparing the importance of each emotional theme with the set importance threshold, retaining emotional themes that meet the importance threshold to achieve simplification of the emotional themes.

6. The method of Internet of Things information technology customer service based on emotion perception according to claim 2 is characterized in that: The method of analyzing the user's current emotional state according to the simplified emotional theme and using the emotional dictionary algorithm comprises the following steps: S241. Construct a sentiment dictionary using the basic sentiment dictionary, degree adverb dictionary, negation dictionary, interjection dictionary, rhetorical interrogative dictionary and relational conjunction dictionary; S242, importing the simplified emotional theme into an emotional dictionary, identifying emotional vocabulary in the simplified emotional theme, wherein the emotional vocabulary includes basic emotional words, degree adverbs, negative words, interjections, rhetorical questions and relational conjunctions; S243, assigning weights to the identified sentiment words in combination with semantic rules, and calculating the sentiment value of the sentiment topic; S244: Determine the user's emotional state based on the calculated emotion value.

7. The method for Internet of Things information technology customer service based on emotion perception according to claim 6 is characterized in that: The calculation formula for calculating the sentiment value of the sentiment theme is: ； In the formula, V represents the sentiment value of the emotional topic; n represents the number of basic sentiment words; m represents the number of interjections; r represents the number of negative words; W i Represents the weight of the i-th basic sentiment word; N i represents the weight of the i-th interjection; D represents the weight of the degree adverb; X represents the negative word gating, if a negative word appears, the value is 1, otherwise the value is 0; Y represents the rhetorical question gate, if a rhetorical question appears, the value is 1, otherwise the value is 0; H represents the influence coefficient.

8. The method of Internet of Things information technology customer service based on emotion perception according to claim 1 is characterized in that: The method of analyzing the user's intention and purpose according to the real-time message data and generating reply content through the reply database according to the user's current emotional state includes the following steps: S31, collecting message text data with user intention annotations, and performing feature extraction on the message text data; S32, training the extracted features based on the intent recognition model of the RoBERTa pre-trained model, and identifying the user's intent purpose using the trained intent recognition model according to the extracted features; S33, selecting a reply template matching the user's intention from a reply database according to the user's intention; S34. Adjust the tone and content of the reply template according to the user's current emotional state to obtain the final reply content.

9. The method for Internet of Things information technology customer service based on emotion perception according to claim 8 is characterized in that: The step of selecting a reply template matching the user's intention from a reply database according to the user's intention includes the following steps: S331. Collect the user's intended questions and customer service response content through historical manual customer service records; S332. Use natural language processing technology to perform text analysis and semantic understanding on the intended questions and customer service response content, and extract key response information and intent labels; S333: Based on the reply key information, a reply template is constructed for each intent tag, and the reply template and the intent tag are associated to construct a reply database; S334, performing similarity matching between the user's intention purpose and the intention tags in the reply database; S335. According to the similarity matching result, a reply template associated with the intent tag with the highest similarity is selected as a reply template that matches the user's intent purpose.

10. An Internet of Things information technology customer service system based on emotion perception, used to implement the Internet of Things information technology customer service method based on emotion perception as claimed in any one of claims 1 to 9, characterized in that: The Internet of Things information technology customer service system based on emotion perception includes: a data acquisition module, an emotion analysis module, a reply content generation module and a reply content push module, and the data acquisition module, the emotion analysis module, the reply content generation module and the reply content push module are connected in sequence; The data acquisition module is used to use the Internet of Things device to acquire the real-time message data between the user and the customer service, and transmit it to the cloud server in real time through the Internet of Things platform. The real-time message data is the text information exchanged between the user and the customer service; The emotion analysis module is used to perform text analysis on the acquired real-time message data and analyze the user's current emotional state through an emotion analysis method; The reply content generation module is used to analyze the user's intention and purpose according to the real-time message data, and generate reply content through the reply database according to the user's current emotional state; The reply content push module is used to push the generated reply content back to the user's IoT device using the IoT platform, and continuously monitor the interaction data between the user and the customer service to dynamically adjust the reply content.

Images