Intelligent customer service reply method, system and device and medium

Abstract

The invention is applicable to the field of intelligent customer service, and discloses an intelligent customer service reply method, which comprises the following steps: collecting user data; analyzing the emotional state of the user based on the user data to obtain an analysis result; generating a response reply according to the analysis result; sending the questionnaire to the user; acquiring behavior data of the user in a questionnaire processing process; and optimizing the questionnaire based on the behavior data. The method aims at comprehensively collecting and analyzing the user data, accurately analyzing the emotional state of the user, and generating personalized response reply according to the emotional state. On the basis, the system sends a questionnaire to the user to deeply understand user preferences and demands, and in the investigation process, the system captures and records behavior data of the user in real time for evaluating rationality of questionnaire design and user participation degree. By deeply mining and analyzing the behavior data, the system can accurately identify problems existing in the questionnaire and automatically implement optimization measures so as to comprehensively improve the user experience and data quality of the questionnaire.

Description

Technical Field

[0001] The present invention relates to the field of intelligent customer service, and in particular to a reply method, system, device and medium for intelligent customer service. Background Art

[0002] With the rapid development of information technology, intelligent customer service systems have become an important bridge of communication between enterprises and customers; traditional customer service response methods are often based on fixed scripts or templates, which make it difficult to effectively respond to users' diverse needs and emotional changes, resulting in poor user experience; in order to improve the intelligence level of customer service responses and user experience, the industry has begun to explore intelligent customer service response methods based on user data analysis.

[0003] However, most existing intelligent customer service response methods only focus on understanding and analyzing the user's text content, and lack in-depth insights into the user's emotional state. In addition, although some intelligent customer service systems also send questionnaires to users to collect user feedback, they often lack comprehensive collection and analysis of user behavior data during the questionnaire processing process, making it difficult to continuously optimize the questionnaire design and affecting the accuracy of the survey results. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a reply method, system, device and medium for intelligent customer service.

[0005] In the first aspect, the present application proposes a response method for intelligent customer service, including: collecting user data, analyzing the user's emotional state based on the user data, obtaining analysis results, generating a response reply based on the analysis results, sending a questionnaire to the user, and obtaining the user's behavior data in the process of processing the questionnaire; optimizing the questionnaire based on the behavior data.

[0006] More specifically, in the above technical solution, obtaining the behavior data of the user in the process of processing the questionnaire, and optimizing the questionnaire based on the behavior data, includes:

[0007] Determine whether the behavior data contains abnormal behavior according to a preset threshold, and if so, mark the behavior data of the user as abnormal data;

[0008] Performing cluster analysis on the abnormal data to obtain clustering results, and determining the types of questions present in the questionnaire according to the clustering results;

[0009] According to the type of problem, formulate corresponding optimization measures;

[0010] Apply the optimization measures to subsequent questionnaires;

[0011] The behavior data includes answering time, skipping questions and submission interruption information.

[0012] More specifically, in the above technical solution, the types of problems include:

[0013] The first type of problem: the question of the questionnaire is not clearly stated;

[0014] The second type of problem: the questionnaire contains too many questions;

[0015] The third type of problem: the options of the questionnaire are not set reasonably;

[0016] The fourth problem type: the questionnaire page loads slowly;

[0017] The optimization measures include: simplifying the question expression, controlling the questions of the questionnaire, increasing the mutual exclusivity and exhaustiveness of the options in the questionnaire, and optimizing the page loading speed.

[0018] More specifically, in the above technical solution, collecting user data includes:

[0019] Acquire interaction data with the user, the interaction data including unstructured data of text, voice, image and video, and convert the unstructured data into a structured feature vector;

[0020] Constructing a corresponding machine learning model according to the structured feature vector, wherein the machine learning model includes a text classification model, a sentiment analysis model, a speech recognition model, an image classification model, and a target detection model;

[0021] The machine learning model is integrated into the intelligent customer service. When new user interaction data is received, it is converted into a structured feature vector and input into the corresponding machine learning model for analysis.

[0022] More specifically, in the above technical solution, after collecting user data, the user data is encrypted and stored, which specifically includes:

[0023] The pre-processed user data is encrypted using an AES symmetric encryption algorithm;

[0024] The encrypted user data is transmitted to the data storage server via a secure channel configured with SSL / TLS protocol;

[0025] The data storage server uses MySQL database storage and implements access control and authority management of the user data through the Apache Shiro framework.

[0026] More specifically, in the above technical solution, after collecting user data, cleaning the user data and processing missing values ​​and abnormal values ​​specifically include:

[0027] Get the proportion and distribution of missing values ​​for each attribute in the data set;

[0028] If the proportion of missing values ​​is lower than a preset threshold, the records containing the missing values ​​are deleted; if the proportion of missing values ​​is higher than the preset threshold, the missing values ​​are estimated by using the mean interpolation or median interpolation method;

[0029] For continuous attributes, calculate the mean and standard deviation of the attribute value to determine whether there are abnormal values ​​outside the normal range;

[0030] For discrete attributes, outliers with low occurrence frequency are identified based on the frequency distribution of attribute values.

[0031] More specifically, in the above technical solution, analyzing the user's emotional state based on the user data and obtaining the analysis result specifically includes:

[0032] Based on the user data, a word segmentation tool is used to perform Chinese word segmentation, and a stop word list is used to filter words;

[0033] Extracting keyword features and sentiment word features representing sentiment tendencies from the user data, converting the keyword features and sentiment word features into word vectors through a Word2Vec word embedding model, and constructing a sentiment feature vector;

[0034] Classifying the emotional feature vectors to determine the emotional category to which each user data belongs;

[0035] According to the emotion categories, various types of emotional state visualization charts are generated through the data visualization library.

[0036] In a second aspect, the present application also proposes an intelligent customer service response system, including:

[0037] The first processing module: collect user data;

[0038] The second processing module is used to analyze the user's emotional state based on the user data and obtain the analysis result;

[0039] The third processing module generates a response reply according to the analysis result;

[0040] Fourth processing module: sending questionnaires to users;

[0041] The fifth processing module: obtaining the user's behavior data in the process of processing the questionnaire;

[0042] The sixth processing module: optimizing the questionnaire based on the behavioral data.

[0043] In a third aspect, the present application further proposes a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods described above when executing the computer program.

[0044] In a fourth aspect, the present application further proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method as described in any one of the above items is implemented.

[0045] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0046] The system sends questionnaires to users to gain an in-depth understanding of their preferences and needs, and evaluates the rationality of the questionnaire design and user engagement by capturing and recording user behavior data in real time (such as answering time, skipped questions, and submission interruption information). Optimizing the questionnaire design can ensure that the questionnaire content is closer to the user's reality and improve user engagement and feedback quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. 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 paying creative work.

[0048] Figure 1 It is a flowchart of a reply method of an intelligent customer service provided by an embodiment of the present invention;

[0049] Figure 2 It is a structural diagram of an intelligent customer service reply system provided by an embodiment of the present invention;

[0050] Figure 3 It is a schematic diagram of the structure of the terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0052] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0053] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0054] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0055] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0056] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0057] See also Figure 1 The present application proposes a response method of intelligent customer service, including: S101 collecting user data, S102 analyzing the user's emotional state based on the user data and obtaining the analysis result, S103 generating a response reply according to the analysis result, S104 sending a questionnaire to the user, S105 obtaining the user's behavior data in the process of processing the questionnaire; S106 optimizing the questionnaire based on the behavior data.

[0058] Acquire multimodal data of speech and facial expressions when the user interacts with the intelligent customer service system, and the multimodal data can be collected and obtained through existing equipment; pre-process the multimodal data, extract speech features and facial features, and obtain a user feature vector; input the user feature vector into a pre-trained support vector machine emotion classification model, judge the user's current emotional state through model reasoning, and obtain a label representing the emotion category; according to the emotion category label, retrieve candidate reply texts related to the emotional state from the knowledge base; calculate the similarity between the candidate reply text and the current dialogue context, use the TF-IDF algorithm to calculate the text similarity score, and select the reply with the highest similarity score as the best matching reply; perform content filtering on the best matching reply, and filter replies containing sensitive words and typos through regular expression matching; use natural language processing technology to perform grammatical error correction and semantic optimization on the filtered reply text to obtain an optimized reply text; convert the optimized reply text into speech through speech synthesis technology, or transmit it to the user in text form to complete an intelligent customer service reply; record each conversation in the background of the customer service system Log data, including user questions, system responses, and user emotional state information; regularly analyzing the collected session log data, using natural language processing technology to extract keywords from user feedback, counting the number and proportion of various types of feedback, and identifying problems in the customer service response process; optimizing and expanding the response content in the knowledge base for the identified problems to improve the coverage and accuracy of the responses; automatically generating questionnaires, the content of which is dynamically adjusted according to the user's basic attributes and interests, and sent to users via SMS or APP push or online responses, recording the user's behavioral data on filling out the questionnaire, including answering speed, number of modifications, dwell time, etc., using clustering algorithms to group users, and identifying the behavioral characteristics of different user groups when filling out questionnaires; comparing and analyzing user feedback on different questionnaire versions, using association rule mining algorithms to find out the key factors that affect the user's filling experience, and optimizing the content, format, and interaction method of the questionnaire accordingly; sending the optimized questionnaire to the user again, or sending the optimized questionnaire to other users for filling out when replying to other users, so that it can better meet business needs.

[0059] Specifically, the intelligent customer service system collects the user's facial expression and voice data through a high-definition camera and microphone array, with a sampling rate of 41kHz and a bit depth of 16bit. The collected voice data is preprocessed, including denoising, silence detection and endpoint detection, and 39-dimensional MFCC features and fundamental frequency features are extracted. Face detection and alignment are performed on the facial image, 68 facial key point coordinates and local binary pattern histogram features are extracted, and the voice features and facial features are spliced ​​to form a 128-dimensional user feature vector. The SVM multi-classification model trained based on the LibSVM library is used to identify user emotions, and the classification accuracy rate reaches more than 85%. Emotion categories include happiness, surprise, anger, disgust, fear, sadness and neutrality. According to the identified emotional tags, relevant candidate replies are retrieved in the knowledge base, which contains more than 100,000 question-answer pairs, covering 20 subject areas. The improved TF-IDF algorithm is used to calculate the similarity between the candidate replies and the current dialogue context, and the part-of-speech weight and word order information are introduced to improve the accuracy of short text matching. The best matching reply is filtered for sensitive words. The sensitive word library includes categories such as politics, violence, and pornography, with a total of 10,000 words. A rule-based and statistical method is used to detect and correct grammatical errors in the reply text, and a pre-trained Seq2Seq model is used for semantic optimization and sentence rewriting. The optimized text is converted into speech. The WaveNet acoustic model and Tacotron2 speech synthesis model are used. The synthesized speech is natural and fluent, and the similarity with real-person speech reaches more than 5 points. The customer service system automatically records log data such as user questions, system responses, and emotional states for each session, generating an average of more than 50GB of data per day. The Word2Vec and LDA topic models are used to extract keywords and cluster analysis of user feedback, identify problems such as insufficient reply coverage and inaccurate content, and supplement and optimize the knowledge base in a targeted manner, increasing the reply coverage by more than 20%. Based on the user's basic attributes such as age, gender, occupation and interest tags, a decision tree algorithm is used to automatically generate a personalized questionnaire. The questionnaire content is dynamically selected from the question bank. Each user receives a different questionnaire. The questionnaire is sent to the target user in the form of a link through push functions such as mobile phone text messages, APP messages or online replies. The average response time is controlled within 1 minute. When users fill out the questionnaire, the system will record the user's behavior data in real time in the background, including an average answering time of 8 seconds, 5 revisions, and a stay time of 2 minutes. The K-means clustering algorithm is used to group users and identify 3 typical user groups: quick filling type, serious thinking type and casual filling type. Different groups show different behavioral characteristics when filling out the questionnaire.By comparing and analyzing user feedback from different questionnaire versions and using the Apriori association rule mining algorithm, we found that the top three key factors affecting the user filling experience are questionnaire length, question difficulty, and interface design. Based on this, we optimized the questionnaire, keeping the number of questions within 20, with moderate difficulty and a beautiful and simple interface. The optimized questionnaire was sent to users again for filling out, or other similar users were invited to fill out. After an average of three rounds, the completion rate of the questionnaire increased from 60% to 95%, and the data quality was significantly improved, providing strong support for subsequent business analysis.

[0060] By collecting and analyzing users' behavioral data when processing questionnaires, the intelligent customer service system can accurately identify problems in questionnaire design, such as unclear question wording and unreasonable option settings, so as to make timely optimization and improve users' filling experience and satisfaction.

[0061] The collection and analysis of behavioral data helps the intelligent customer service system evaluate the rationality of questionnaire design and user engagement. By optimizing the questionnaire design, it can reduce the generation of invalid and abnormal data and improve the accuracy and reliability of the survey results.

[0062] Optimization measures based on behavioral data, such as simplifying question wording, controlling the number of questionnaire questions, increasing the mutual exclusivity and exhaustiveness of options, etc., can further enhance the intelligence level of the intelligent customer service system and make it more in line with user needs and preferences.

[0063] By continuously optimizing the questionnaire, the intelligent customer service system can continuously accumulate user feedback and behavior data, provide strong support for the iterative upgrade of the system, and promote the system to develop in a more efficient and intelligent direction.

[0064] The optimized questionnaire can more accurately reflect user needs and problems. The intelligent customer service system can make targeted improvements and optimizations based on these feedbacks, thereby improving customer service quality and user satisfaction. By optimizing the questionnaire design to make it more concise, clear and easy to understand, it can reduce the difficulty and resistance of users to fill in the questionnaire, thereby enhancing user participation and the representativeness of the survey results.

[0065] In some embodiments, the customer service system can analyze the user's behavioral data during the survey process, such as the length of time to answer questions, skipped questions, submission interruptions, etc., to identify deficiencies in the questionnaire design and optimize subsequent questionnaires. Including:

[0066] Obtaining the behavior data of the user during the questionnaire survey recorded by the customer service system, wherein the behavior data includes the time taken to answer questions, skipped questions, and submission interruption information;

[0067] Preprocessing the behavior data, removing abnormal data, and normalizing the behavior data;

[0068] Determine whether the user has abnormal behavior in the behavior data in terms of answering time, skipping questions, and submission interruption according to a preset threshold, and if there is an abnormality, mark the user behavior data as abnormal data;

[0069] The K-means clustering algorithm is used to perform cluster analysis on the abnormal data, and the main types of problems existing in the questionnaire design are determined according to the data characteristics of each cluster in the clustering results;

[0070] For each type of question obtained through cluster analysis, analyze the specific performance of this type of question in the questionnaire design and summarize the common characteristics of this type of question;

[0071] According to the common characteristics of each type of questions, formulate corresponding questionnaire optimization measures;

[0072] Apply the optimized questionnaire design to subsequent surveys and continue to record user behavior data;

[0073] Compare and analyze the newly collected user behavior data with previous data to evaluate the effectiveness of questionnaire optimization measures;

[0074] Further adjust optimization measures based on the analysis results, and continuously iterate and optimize the questionnaire design.

[0075] Specifically, the customer service system recorded the behavior data of 10,000 users during the questionnaire survey, including answering time, skipping questions, and submission interruption information. The data was preprocessed to eliminate 5% of abnormal data, and the minimum-maximum normalization method was used to normalize the data to the [0,1] interval. According to the threshold set by experience, user behaviors with answering time exceeding 10 minutes, skipping more than 5 questions, or submitting more than 2 interruptions were marked as abnormal, accounting for 8%. The K-means clustering algorithm was used to cluster the abnormal data, and the optimal number of clusters was determined to be 4 by the elbow rule. The first type of problem is that the question description is unclear, which makes it difficult for users to understand. The second type is that there are too many questionnaire questions, which makes users lose patience. The third type is that the question options are set unreasonably. The fourth type is that the slow loading speed of the questionnaire page affects the user experience. For these four types of problems, corresponding optimization measures were formulated: simplify the question description, control the questionnaire length within 20 questions, increase the mutual exclusivity and exhaustiveness of options, and optimize the page loading speed to within 3 seconds. The optimized questionnaire was applied to the subsequent survey of 3,000 users, and the proportion of abnormal behavior was reduced to 5%. The chi-square test found that the abnormal proportions before and after optimization were significantly different, p < 0.01. Based on the analysis results, the questionnaire design specifications were further refined, and the A / B test was used to continuously optimize the questionnaire, improving data quality and user experience.

[0076] Simplifying the wording of questions may include: Reviewing the content of the questions: Re-examine the wording of each question to ensure that they are clear, direct, and easy to understand. Avoiding professional terms: If the questionnaire is aimed at ordinary users, try to avoid using industry terms or complex language, and replace them with easy-to-understand terms. Paragraphing and formatting: Long sentences can be split into short sentences, or bullet points and lists can be used to make information easier to digest. Example, original question: "Based on your experience, please evaluate the efficiency and accuracy of our intelligent customer service in solving complex problems." After optimization: "Do you think our intelligent customer service is fast in solving difficult problems? Is the solution accurate?".

[0077] Controlling the questions in the questionnaire may include: determining the core purpose of the questionnaire and only including directly relevant questions; if two or more questions are measuring similar concepts, try to merge them into a composite question; remove questions that do not contribute much to the core goal, are redundant, or do not provide new information; use open-ended questions at key points to obtain more detailed feedback, rather than multiple closed questions. Example:

[0078] Suppose there were originally 25 questions, and by deleting redundant questions (such as repeated satisfaction surveys) and merging similar questions (such as merging three satisfaction questions about product features into one), they were eventually reduced to 20 questions.

[0079] Increase the mutual exclusivity and exhaustiveness of options include: Mutual exclusivity: Make sure each option represents a unique answer and there is no overlap between answers. Exhaustiveness: List all possible answer options and try to avoid open options such as "other" unless they are really necessary. Use scales: For some types of questions (such as satisfaction, frequency, etc.), using scales (such as Likert scales) can provide more refined answer choices.

[0080] For example, the original option: "Are you satisfied with our product? A. Very satisfied B. Satisfied C. Average D. Dissatisfied E. Very dissatisfied". It already has good mutual exclusivity and exhaustiveness, but if you find that the "average" option is often selected and no specific feedback can be provided, you can consider adding a more detailed description option.

[0081] Optimizing page loading speed includes: compressing image size, using appropriate image formats (such as WebP) to reduce loading time; merging CSS and JavaScript files, and using CDN (content distribution network) to speed up resource loading; optimizing database queries, using caching mechanisms, and reducing server response time; using asynchronous loading (such as Lazy Load) to reduce the amount of data required when the page is first loaded; using tools (such as Google PageSpeed ​​Insights) to regularly monitor and optimize page performance. For example, a code review of the questionnaire page revealed a large number of uncompressed images and redundant HTTP requests. By compressing images, merging resource files, and using CDN, the page loading time was successfully reduced from 6 seconds to within 3 seconds.

[0082] It is understandable that the analysis of answering time: analyze the time users spend on each question, identify which questions cause users to spend too much time, which may be unclear or the options are too complicated. Skip question analysis: count the proportion of questions skipped by users. If a question is skipped a lot, it may be necessary to redesign the question or adjust the questionnaire structure to make it more attractive for users to answer. Submission interruption information analysis: analyze the reasons and frequency of user submission interruptions. If the interruption rate is high, it may be necessary to optimize the loading speed of the questionnaire, simplify the questionnaire process, or provide a function to save progress.

[0083] It should be noted that behavioral data collection is a well-known technology in the field and will not be elaborated in this application. For example, the following methods can be used: Timestamp recording: Set a timestamp record for each question or each page of the questionnaire. When the user browses or answers a question, the system records the current timestamp. By calculating the difference between adjacent timestamps, the time it takes the user to answer the question can be obtained. User interaction log: records all interactive behaviors of the user on the questionnaire page, including clicks, scrolling, input, etc. These logs can reflect whether the user has skipped certain questions or returned to modify the answer multiple times. Submission interruption information: When a user is suddenly interrupted while filling out a questionnaire (such as closing the page, browser crashing, etc.), the system can record the questionnaire progress and the user's operation status at the time of the interruption as submission interruption information.

[0084] Optional, subsequent optimization strategies include: Adjust the order of questions: Based on the time users spend answering questions and the data on skipped questions, rearrange the order of questions in the questionnaire and put the questions that users are more interested in or easier to answer in front. Simplify question statements: For questions that take users too long to answer, redesign the question statements to make them more concise and clear. Optimize option design: For complex multiple-choice questions or scale questions, simplify the number of options or provide clear option explanations to reduce the difficulty of user selection. Provide progress feedback: Display a progress bar or remaining time in the questionnaire to let users know the completion status of the questionnaire and reduce the possibility of interruption.

[0085] Added save function: For longer questionnaires, provide the function of saving progress, allowing users to continue filling out the questionnaire after interruption.

[0086] In some embodiments, the intelligent customer service system collects data when communicating with users.

[0087] include:

[0088] Acquire user interaction data collected in the intelligent customer service system, wherein the user interaction data includes unstructured data of a preset number of types such as text, voice, image, and video;

[0089] Preprocessing the user interaction data, converting the unstructured data into structured feature vectors through data cleaning and feature extraction;

[0090] According to the data type of the structured feature vector, a corresponding machine learning model is constructed, wherein the machine learning model includes a text classification model, a sentiment analysis model, a speech recognition model, an image classification model, and a target detection model;

[0091] Using the labeled data to train the machine learning model to obtain a trained machine learning model;

[0092] Integrate the trained machine learning model into the intelligent customer service system. When new user interaction data is received, convert it into a structured feature vector and input it into the corresponding machine learning model for analysis to obtain user intention and emotional tendency.

[0093] According to the user's intention and emotional tendency, combined with preset rules and strategies, matching reply templates and service items are retrieved from the knowledge base to generate personalized customer service response results;

[0094] Obtain user feedback data on the customer service response results, associate with the service log of the intelligent customer service system, and evaluate customer service quality and user satisfaction;

[0095] Based on the evaluation results, the knowledge base, response strategy and recommendation algorithm of the intelligent customer service system are optimized and iterated to improve the customer service level;

[0096] Collect key business indicators of the intelligent customer service system, generate visual data reports and dashboards, and provide data support for business decisions.

[0097] Specifically, the intelligent customer service system collects user interaction data, including 1,000 texts, 500 voices, 200 images, 100 videos and other unstructured data. These data are preprocessed, and the TF-IDF algorithm is used to extract text features, the MFCC algorithm is used to extract voice features, and the CNN is used to extract image features, and then they are converted into structured feature vectors. According to the type of feature vector, the corresponding machine learning model is constructed, such as using the BERT model for text classification, the BiLSTM model for sentiment analysis, the CNN+CTC model for speech recognition, the ResNet model for image classification, and the YOLO model for target detection. The model is trained using 5,000 annotated data, and the model parameters are optimized using the cross-validation method, and finally a model with an accuracy rate of more than 90% is obtained. The trained model is integrated into the intelligent customer service system. When the user asks a new question, the system automatically converts it into a structured feature and inputs it into the corresponding model for analysis. For example, when a user asks, "When will my order be shipped?", the system identifies the user's intention as "querying the order status" through the text classification model, and the sentiment analysis model determines the user's emotion as "anxiety". Then, the system retrieves relevant reply templates and service items from the knowledge base, such as "Hello, your order is being processed and is expected to be shipped within 3 working days. We will arrange it for you as soon as possible, please be patient." At the same time, the system will also recommend related products and services based on the user's historical interaction data and preferences. By collecting user feedback on customer service responses, such as satisfaction scores and the number of follow-up questions, associating service logs, using logistic regression models to evaluate customer service quality, and using collaborative filtering algorithms to optimize the knowledge base and recommendation strategies, the level of customer service is continuously improved. Finally, the system uses visualization tools such as Tableau to count key indicators of customer service, such as problem resolution rate, response time, user satisfaction, etc., to generate intuitive data reports and dashboards to provide data support for business decisions.

[0098] In some embodiments, the collected data is encrypted and stored to ensure data security. This includes:

[0099] According to data security requirements, the AES symmetric encryption algorithm is used to encrypt the preprocessed data. The encrypted data is transmitted to the data storage server through a secure channel configured with the SSL / TLS protocol. The data storage server uses a MySQL database to store the encrypted data, and implements access control and authority management of the data through the Apache Shiro framework.

[0100] Specifically, in the data preprocessing stage, the AES-256-bit encryption algorithm is used to encrypt the data using a randomly generated 256-bit key. During the encryption process, the data is divided into fixed-size blocks, each of which is 128 bits in size, and then each block is encrypted, and the encrypted data block size is 256 bits. The encrypted data is transmitted through a secure channel configured with the SSL / TLS protocol, in which the public and private key pairs generated by the RSA-2048-bit algorithm are used for authentication and key exchange, and a secure communication tunnel is established to ensure the confidentiality and integrity of the data transmission process. After being transmitted to the data storage server, the encrypted data is stored using the AES_ENCRYPT function of the MySQL database, and the key is protected by the hash value calculated by the SHA-256 algorithm and stored in a separate key management server. When accessing data, permission control is performed through the Apache Shiro framework, and the role-based access control model (RBAC) is adopted to dynamically generate database query statements based on the user's role and permissions, avoiding the risk of injection attacks through splicing SQL statements and ensuring the security of data access. At the same time, the session management and caching mechanism provided by Shiro reduces frequent database access and improves system performance.

[0101] In some embodiments, data is cleaned to handle missing values ​​and outliers. This includes:

[0102] Get the missing value ratio and distribution of each attribute in the data set. If the missing value ratio is lower than the preset threshold, delete the records containing missing values. If the missing value ratio is higher than the preset threshold, use the mean interpolation or median interpolation method to estimate the missing values. For continuous attributes, calculate the mean and standard deviation of the attribute value, and use the Z-score method to determine whether there are outliers outside the normal range. For discrete attributes, use the value_counts() function to identify outliers with extremely low frequency according to the frequency distribution of the attribute value. Use the train_test_split() function to divide the cleaned data set into training set, validation set, and test set according to the preset ratio for building and evaluating machine learning models.

[0103] Specifically, when preprocessing the data set, the missing value ratio of each attribute is first counted. For example, through statistics, it is found that the missing value ratio of the "age" attribute is 5%, which is lower than the preset 5% threshold, so the records containing missing values ​​of "age" are directly deleted; while the missing value ratio of the "income" attribute is 2%, which is higher than the preset threshold, and the median interpolation method is used to estimate the missing values, that is, all valid values ​​of the "income" attribute are arranged in ascending order, and the value in the middle position is used to replace the missing value. Next, the continuous attribute "monthly consumption amount" is detected for outliers, and its mean is calculated to be 2316 yuan and the standard deviation is 864 yuan. Then, the Z-score method is used to determine whether the "monthly consumption amount" of each record exceeds the normal range, and the records with an absolute value of Z-score greater than 3 are marked as outliers. For the discrete attribute "occupation", outlier identification is performed, and the value_counts() function is used to count the frequency of each occupation. It is found that the "other" category only appears twice, which is much lower than other categories, so it is marked as an outlier. Finally, use the train_test_split() function to divide the cleaned dataset into training, validation, and test sets in a ratio of 8:1:1, containing 9,600, 1,200, and 1,200 records respectively, in preparation for the subsequent construction of the machine learning model.

[0104] In some embodiments, the user's emotional state is analyzed based on the collected data. This includes:

[0105] The original text data of users is obtained, and the Jieba word segmentation tool is used to perform Chinese word segmentation. The stop word list is used to remove meaningless words, and the text data is converted into structured feature data. Keywords and sentiment word features that represent emotional tendencies are extracted, and the feature words are converted into word vectors using the Word2Vec word embedding model to construct sentiment analysis feature vectors. The naive Bayes algorithm is used to classify the sentiment feature vectors to determine the sentiment category to which each user text data belongs. Various types of emotional state visualization charts are generated through the Matplotlib and Seaborn data visualization libraries to intuitively present the results of user sentiment analysis.

[0106] Specifically, first, obtain the original text interaction data of the user in the intelligent customer service system, use the Jieba Chinese word segmentation tool to segment the text, set the word segmentation granularity to the precise mode, and load the common Chinese stop word list to filter out meaningless words such as "de" (的) and "le" (了). Then, extract the keywords and sentiment words indicating sentiment tendency in the text, such as "like" and "hate", and use the pre-trained Word2Vec word embedding model to convert these feature words into 300-dimensional word vectors. Next, splice the sentiment word vectors to construct a feature vector for sentiment analysis. Use the Naive Bayes algorithm to classify the sentiment feature vector, and set the category labels to three emotional states: positive, neutral, and negative. Through the cross-validation method, use 80% of the data for model training and 20% of the data for testing, and evaluate that the model accuracy reaches more than 85%. Finally, use the Matplotlib and Seaborn visualization libraries to generate a pie chart to show the proportion of each type of emotion, and generate a line chart to show the change trend of the user's emotions over a period of time, providing intuitive emotion analysis results for the customer service staff. Through the real-time monitoring and analysis of the user's emotions, the intelligent customer service system can adjust the service strategy and response method according to the user's emotional state, provide more personalized and user-friendly services, and improve the user satisfaction and experience quality.

[0107] Please refer to Figure 2 , this application also proposes a response system for an intelligent customer service, including:

[0108] The first processing module 201: collect user data;

[0109] The second processing module 202: analyze the emotional state of the user based on the user data and obtain the analysis result;

[0110] The third processing module 203: generate a response reply according to the analysis result;

[0111] The fourth processing module 204: send a questionnaire to the user;

[0112] The fifth processing module 205: obtain the behavior data of the user during the process of processing the questionnaire;

[0113] The sixth processing module 206: optimize the questionnaire based on the behavior data.

[0114] It can be understood that the content in the embodiment of the response method of the intelligent customer service as Figure 1 shown is applicable to the embodiment of this intelligent customer service response system. The functions specifically implemented by the embodiment of this intelligent customer service response system are the same as those in the embodiment of the response method of the intelligent customer service as Figure 1 shown, and the beneficial effects achieved are the same as those in the embodiment of the response method of the intelligent customer service as Figure 1The beneficial effects achieved by the embodiment of the intelligent customer service reply method shown are also the same.

[0115] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0116] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0117] See also Figure 3 The embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, the intelligent customer service reply method described in any one of the above methods is implemented.

[0118] The computer device 3 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will appreciate that Figure 3 It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0119] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0120] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 3. Further, the memory 302 may also include both an internal storage unit of the computer device 3 and an external storage device. The memory 302 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory 302 may also be used to temporarily store data that has been output or is to be output.

[0121] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the intelligent customer service reply method as described in any one of the above methods is implemented.

[0122] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0123] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A reply method of intelligent customer service, characterized in that: include: Collect user data; Analyze the user's emotional state based on the user data and obtain analysis results; generating a response reply according to the analysis result; Send questionnaires to users; Obtaining user behavior data during the process of processing the questionnaire; The questionnaire is optimized based on the behavioral data.

2. The intelligent customer service reply method according to claim 1, characterized in that: Obtaining user behavior data during processing of the questionnaire, and optimizing the questionnaire based on the behavior data, including: Determine whether the behavior data contains abnormal behavior according to a preset threshold, and if so, mark the behavior data of the user as abnormal data; Performing cluster analysis on the abnormal data to obtain clustering results, and determining the types of questions present in the questionnaire according to the clustering results; According to the type of problem, formulate corresponding optimization measures; Apply the optimization measures to subsequent questionnaires; The behavior data includes answering time, skipping questions and submission interruption information.

3. The intelligent customer service reply method according to claim 2, characterized in that: The types of questions include: The first type of problem: the question of the questionnaire is not clearly stated; The second type of problem: the questionnaire contains too many questions; The third type of problem: the options of the questionnaire are not set reasonably; The fourth problem type: the questionnaire page loads slowly; The optimization measures include: simplifying the question expression, controlling the questions of the questionnaire, increasing the mutual exclusivity and exhaustiveness of the options in the questionnaire, and optimizing the page loading speed.

4. The intelligent customer service reply method according to claim 1, characterized in that: User data collected includes: Acquire interaction data with the user, the interaction data including unstructured data of text, voice, image and video, and convert the unstructured data into a structured feature vector; Constructing a corresponding machine learning model according to the structured feature vector, wherein the machine learning model includes a text classification model, a sentiment analysis model, a speech recognition model, an image classification model, and a target detection model; The machine learning model is integrated into the intelligent customer service. When new user interaction data is received, it is converted into a structured feature vector and input into the corresponding machine learning model for analysis.

5. The intelligent customer service reply method according to claim 1, characterized in that: After collecting user data, the user data is encrypted and stored, specifically including: The pre-processed user data is encrypted using an AES symmetric encryption algorithm; The encrypted user data is transmitted to the data storage server via a secure channel configured with SSL / TLS protocol; The data storage server uses MySQL database storage and implements access control and authority management of the user data through the Apache Shiro framework.

6. The intelligent customer service reply method according to claim 1, characterized in that: After collecting user data, clean the user data and process missing values ​​and outliers, including: Get the proportion and distribution of missing values ​​for each attribute in the data set; If the proportion of missing values ​​is lower than a preset threshold, the records containing the missing values ​​are deleted; if the proportion of missing values ​​is higher than the preset threshold, the missing values ​​are estimated by using mean interpolation or median interpolation methods; For continuous attributes, calculate the mean and standard deviation of the attribute value to determine whether there are abnormal values ​​outside the normal range; For discrete attributes, outliers with low occurrence frequency are identified based on the frequency distribution of attribute values.

7. The intelligent customer service reply method according to claim 1, characterized in that: Analyzing the user's emotional state based on the user data and obtaining analysis results specifically includes: Based on the user data, a word segmentation tool is used to perform Chinese word segmentation, and a stop word list is used to filter words; Extracting keyword features and sentiment word features representing sentiment tendencies from the user data, converting the keyword features and sentiment word features into word vectors through a Word2Vec word embedding model, and constructing a sentiment feature vector; Classifying the emotional feature vectors to determine the emotional category to which each user data belongs; According to the emotion categories, various types of emotional state visualization charts are generated through the data visualization library.

8. An intelligent customer service response system, characterized in that: include: The first processing module: collect user data; The second processing module is used to analyze the user's emotional state based on the user data and obtain the analysis result; The third processing module generates a response reply according to the analysis result; Fourth processing module: sending questionnaires to users; The fifth processing module: obtaining the user's behavior data in the process of processing the questionnaire; The sixth processing module: optimizing the questionnaire based on the behavioral data.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images