Multi-channel interactive customer relationship management system

Abstract

The invention, which relates to the technical field of customer relationship management, discloses a multi-channel interactive customer relationship management system comprising a dynamic routing decision module, a multi-modal data fusion module and an intelligent feedback optimization module. The dynamic routing decision module evaluates channel load through a deep neural network, dynamically allocates client requests to an optimal node by utilizing reinforcement learning, and realizes load balancing and service continuity; the multi-modal data fusion module integrates text, voice and image data, constructs a space-time correlation graph, identifies a cross-channel behavior mode, and ensures data consistency through multi-dimensional verification; the intelligent feedback optimization module combines customer satisfaction evaluation and multi-modal sentiment analysis, optimizes a service strategy by using a genetic algorithm, and synchronizes the service strategy to a cross-channel knowledge graph to realize adaptive iteration; according to the method, the problems of unreasonable multi-channel load distribution, insufficient data fusion and consistency verification and inaccurate service strategy optimization are effectively solved, and the customer service quality and experience are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of customer relationship management, and in particular to a multi-channel interactive customer relationship management system. Background Art

[0002] With the rapid development of the internet and mobile technology, the channels for interaction between customers and businesses are becoming increasingly diverse, encompassing websites, mobile applications, social media, call centers, and other forms. This multi-channel interaction model provides businesses with a wider range of customer touchpoints, but it also presents numerous challenges. On the one hand, the volume of customer requests across different channels fluctuates, which can easily lead to overloaded channels on some channels and relatively idle channels on others, impacting service efficiency and customer experience. On the other hand, the types of data generated by each channel are complex and diverse, including text, voice, and images, and these data may contain contradictions or inconsistencies, making it difficult to effectively integrate and analyze, making it impossible to accurately grasp customers' actual needs and behavior patterns. Furthermore, traditional customer relationship management systems (CRMs) have limitations in optimizing service strategies, making it difficult to adjust service processes in real time based on customer feedback and sentiment, and failing to meet customers' growing demand for personalized and high-quality services.

[0003] In order to address the above-mentioned issues, there are already some related technical attempts on the market. For example, some systems use simple load balancing algorithms to distribute customer requests to different servers, but these algorithms are usually relatively fixed and cannot be dynamically adjusted according to the real-time channel load and service capabilities, and do not consider the priority of customer requests and cross-channel service continuity. Some systems also attempt to integrate multi-channel data, but most of them only stay at the data storage level, lack in-depth analysis and correlation of data, cannot effectively identify cross-channel behavior patterns, and cannot resolve data contradictions. In terms of service optimization, although there is a simple feedback mechanism based on customer satisfaction evaluation, it fails to fully utilize multimodal data for sentiment analysis and cannot accurately capture changes in customer emotions, resulting in insufficient accuracy in service strategy optimization.

[0004] In order to solve the above-mentioned defects, a technical solution is now provided. Summary of the Invention

[0005] The purpose of the present invention is to solve the problems of unreasonable load distribution, insufficient data fusion and consistency verification, and inaccurate service strategy optimization in existing customer relationship management systems in multi-channel interaction, and to propose a multi-channel interactive customer relationship management system.

[0006] The purpose of the present invention can be achieved through the following technical solutions: A multi-channel interactive customer relationship management system, comprising: The dynamic routing decision module collects multi-channel customer requests in real time, evaluates channel load through deep neural networks, and uses reinforcement learning to dynamically allocate requests to the optimal node to achieve load balancing and service continuity; The multimodal data fusion module integrates text, voice, and image data, constructs a spatiotemporal correlation map to identify cross-channel behavior patterns, and triggers multi-dimensional verification of conflicting information to ensure data consistency. The intelligent feedback optimization module combines customer satisfaction evaluation and multimodal sentiment analysis, uses genetic algorithms to optimize service strategies, and synchronizes them to the cross-channel knowledge graph to achieve adaptive iteration.

[0007] Furthermore, the execution steps of the dynamic routing decision module are as follows: Collect customer interaction requests from various channels such as websites, mobile applications, social media, and call centers in real time, and extract request features including interaction channel type, request content semantic features, customer identification code, and request initiation timestamp; A channel load assessment model is built based on a deep neural network. The model inputs the current processing queue length, historical average response time, and current number of online customer service personnel of each service node, and outputs a real-time service capability score for each channel. A dynamic routing strategy is established using a reinforcement learning algorithm. The routing weight coefficient is calculated based on the matching degree between customer request characteristics and channel service capability scores. When the request backlog of a channel exceeds the threshold, a cross-channel load balancing mechanism is automatically triggered. When generating dynamic routing instructions, high-priority customer requests are prioritized and assigned to the channel with the highest service score. At the same time, a cross-channel session status tracking identifier is established to ensure service continuity.

[0008] Furthermore, the specific operation steps for triggering the cross-channel load balancing mechanism in the dynamic routing decision module are as follows: Set dynamic thresholds based on historical data; Real-time monitoring of request backlogs and service capability scores across channels; When the backlog of a channel exceeds the threshold and the service capability score is lower than the average, the load balancing mechanism is triggered; Filter candidate channels from currently available channels with a service capability score greater than 0.7 and a load queue length less than 30; If multiple channels meet the conditions, the channel with the highest matching degree with the customer request type will be selected first; The routing weight coefficient generated by the Q-Learning algorithm is used to calculate the comprehensive score of the candidate channel: comprehensive score = service capability score × 0.6 + routing weight coefficient × 0.4; Select the channel with the highest comprehensive score as the target allocation node; Migrate non-urgent requests in the backlog channel to the target channel in proportion, and the migration ratio is dynamically adjusted according to the backlog level; High-priority requests, including VIP customer and complaint requests, remain in the original channel but are assigned to a customer service node with a score higher than the preset standard; Generate a unique session identifier for each cross-channel migration request and synchronize interaction records in real time through Redis; After the target channel customer service takes over, they can quickly obtain context through the session ID to avoid customers repeating the problem description; Monitor the queue length and response time of each channel in real time after migration. If the load on the target channel does not drop as expected within 5 minutes, trigger a second migration. The load balancing decision result is fed back to the Q-Learning model as training data to update the routing strategy weights.

[0009] Furthermore, the multimodal data fusion module performs the following steps: Establish a standardized data cleaning pipeline to perform feature extraction on heterogeneous interaction data such as text, voice, and image. The BERT model is used to extract semantic vectors from text data. Voice data is processed through voiceprint recognition and voice-to-text conversion. Convolutional neural networks are used to extract visual features from image data. Build a spatiotemporal correlation graph to correlate the interaction events generated by the same customer across different channels along a timeline, identify cross-channel behavior patterns, and activate the data credibility verification submodule when contradictory information is detected in multi-channel interactions; When implementing real-time data fusion, the attention mechanism is used to dynamically adjust the weight of each channel's data. If the same demand is generated from three consecutive channels, the weight of the feature is automatically increased to 1.5 times; When generating a unified customer profile, a forgetting factor algorithm is introduced to automatically perform weight decay processing on historical interaction data that exceeds the set time limit.

[0010] Furthermore, the specific operation steps of detecting contradictory information in multi-channel interaction and starting the data credibility verification submodule in the dynamic routing decision module are as follows: Trigger conditions: Real-time analysis of spatiotemporal correlation graphs reveals inconsistent behavior logic or data for the same customer across multiple channels; Conflict determination rules: timestamp conflicts, behavioral logic conflicts, and data value conflicts; The verification process is as follows: Check the legitimacy of the data source: verify whether the channel has passed enterprise certification; Anonymous or temporary conversation channels are marked as low credibility and continue to be verified; Time window analysis: Set a time difference threshold. If the timestamps of cross-channel events exceed the threshold and the content is inconsistent, an alert will be triggered. Behavioral logic verification: Call the business process rule engine to check whether the operation complies with the preset logic; Third-party interface call, logistics status: call the logistics company API to obtain the actual package receipt time and signatory information; Payment voucher: connect to Alipay or WeChat payment interface to verify whether the order amount and status are consistent; Physical evidence comparison includes image or video verification: using hash algorithms or object detection models to compare the credentials uploaded by the customer with those in the system archive; History retrieval: Analyzes customer interaction records from the past 30 days to detect any unusual patterns. Behavioral profile matching: Compare current behavior with normal behavior. If the deviation exceeds the threshold, it is marked as high risk. Data integrity checks include key field verification: checking whether the order number, customer ID, and contact information are complete. If missing, the automatic completion process is triggered; Data tampering detection: Verify whether the original data has been tampered with by comparing the hash values; The verification result processing and decision-making process is as follows: Automatic adjudication: If external data returns a clear result, it automatically corrects the conflicting data and updates the spatiotemporal map; Manual intervention: When the external interface is unavailable or the verification result is ambiguous, a manual review ticket is generated and assigned to the customer service team to contact the customer for verification; Risk management: After confirming false information, mark the customer account risk level and restrict operation permissions; Update the knowledge graph to record conflicting events and handling results.

[0011] Furthermore, the specific operation steps of the intelligent feedback optimization module are as follows: Satisfaction evaluation data is collected after each customer service closed loop, and the sentiment analysis model is used to automatically analyze the emotional change curve during customer interaction; Build a dual-channel feedback analysis model to compare and verify manual evaluation data with machine sentiment analysis results. When the difference between the two exceeds 30%, the manual review mechanism is triggered; Utilizing an adaptive strategy optimization engine and a genetic algorithm to dynamically adjust the response priority weights of each channel, automatically eliminating parameter combinations that have achieved less than 5% improvement over three consecutive optimization cycles. Establish a cross-channel knowledge graph update mechanism to synchronize optimized service strategies to the decision databases of all access channels to ensure the real-time and consistency of strategy updates.

[0012] Furthermore, the specific operation steps of the intelligent feedback optimization module to automatically analyze the emotion change curve during the customer interaction process through the sentiment analysis model are as follows: Collect omnichannel customer interaction data, including text, voice, and images; Noise removal, word segmentation, and stop word filtering; Convert recordings to text using ASR and add timestamps; Convert to lowercase and handle abbreviations; Use the pre-trained VADER model or BERT fine-tuned model to output a sentiment polarity score between -1 and 1 for each sentence; Arrange sentiment scores in chronological order of interaction to generate time series data; Smooth the data through interpolation, draw a graph of sentiment changes, and mark key nodes; Generate an emotional curve for the entire interaction process, annotate events that trigger emotional fluctuations, associate the emotional score with the original interaction data, and store it in MongoDB for subsequent analysis; Then, by integrating multimodal data such as text, voice, and images, the accuracy of sentiment analysis and the ability to understand context can be improved.

[0013] Furthermore, the specific operation steps of the intelligent feedback optimization module by fusing text, voice, and image multimodal data are as follows: Use the Librosa library to extract the fundamental frequency, speech rate, and energy of the speech signal, train a CNN-based speech emotion model, and output classification labels for anger, calmness, and anxiety. Use the ResNet model to classify the emojis or screenshots sent by customers, extract the text in the images using Tesseract, and incorporate it into the text analysis process; Map text sentiment scores, speech sentiment labels, and image classification results into a unified dimension; If the voice anger label is consistent with the text negativity score, the weight is increased to 1.2 times; If the emoji and text are inconsistent, a conflict detection is triggered and the manual review weight is increased by 50%; Extract keywords from each round of conversation using TF-IDF or BERT and correlate them with sentiment fluctuations; Input multimodal emotion sequences and predict subsequent emotion trends.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention, through a dynamic routing decision module, can collect multi-channel customer requests in real time and evaluate the load of each channel. It uses a reinforcement learning algorithm to dynamically allocate requests to the optimal node, achieving load balancing and service continuity. This mechanism effectively solves the problem of some channels being overloaded while other channels are idle in traditional systems, significantly improving resource utilization and service efficiency. At the same time, through the cross-channel load balancing mechanism, it ensures that high-priority customer requests are processed first, further improving the customer experience; (2) In this invention, the multimodal data fusion module integrates multiple data types such as text, voice, and images, constructs a spatiotemporal correlation map to identify cross-channel behavior patterns, and initiates multi-dimensional verification when contradictory information is detected to ensure data consistency. This deep fusion and verification mechanism not only improves the accuracy and reliability of data, but also enables a more comprehensive understanding of customer behavior and needs. In addition, the introduction of a forgetting factor algorithm to perform weight decay processing on historical interaction data further improves the timeliness and accuracy of customer portraits, providing strong support for precise service. (3) In this invention, the intelligent feedback optimization module combines customer satisfaction evaluation with multimodal sentiment analysis, uses genetic algorithms to dynamically optimize service strategies, and synchronizes the optimization results to the cross-channel knowledge graph to achieve adaptive iteration. This module can capture changes in customer sentiment in real time and accurately adjust service strategies to ensure the relevance and effectiveness of services. At the same time, through the dual-channel feedback analysis model and automatic elimination mechanism, the accuracy and efficiency of service strategy optimization are further improved, providing customers with a more personalized and high-quality service experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is the overall system block diagram of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0017] It should be understood that the terms “include” and “comprising” used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0018] It should also be understood that the terminology used in this disclosure is for the purpose of describing specific embodiments only and is not intended to limit the disclosure. As used in this disclosure and the claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should be further understood that the term "and / or" as used in this disclosure and the claims refers to any and all possible combinations of one or more of the associated listed items, including and including these combinations.

[0019] like Figure 1 As shown, a multi-channel interactive customer relationship management system includes a dynamic routing decision module, a multimodal data fusion module and an intelligent feedback optimization module; The dynamic routing decision module collects multi-channel customer requests in real time, evaluates channel load through deep neural networks, and uses reinforcement learning to dynamically allocate requests to the optimal node to achieve load balancing and service continuity; Collect customer interaction requests from various channels such as websites, mobile applications, social media, and call centers in real time, and extract request features including interaction channel type, request content semantic features, customer identification code, and request initiation timestamp; A channel load assessment model is built based on a deep neural network. The model inputs the current processing queue length, historical average response time, and current number of online customer service personnel of each service node, and outputs a real-time service capability score for each channel. A dynamic routing strategy is established using a reinforcement learning algorithm. The routing weight coefficient is calculated based on the matching degree between customer request characteristics and channel service capability scores. When the request backlog of a channel exceeds the threshold, the cross-channel load balancing mechanism is automatically triggered. The specific process is as follows: Set dynamic thresholds based on historical data, such as a channel's real-time request queue length > 50 or an average response time > 5 minutes. Monitor the request backlog (queue length) and service capability score of each channel in real time. Trigger the load balancing mechanism when a channel's backlog exceeds the threshold and its service capability score falls below the system average (e.g., score < 0.6). Select candidate channels from the currently available channels with high service capability scores (e.g., >0.7) and low loads (queue length <30). If multiple channels meet these criteria, prioritize the channel that best matches the customer request type (e.g., voice requests are prioritized for the call center). Using the routing weight coefficient generated by the Q-Learning algorithm (result of step 3), calculate the comprehensive score of the candidate channels: comprehensive score = service capability score × 0.6 + routing weight coefficient × 0.4; select the channel with the highest comprehensive score as the target allocation node; Non-urgent requests (such as general inquiries) in the backlogged channel will be migrated to the target channel in proportion. The migration ratio will be dynamically adjusted according to the backlog level (for example, 30% of requests will be migrated when the backlog exceeds the threshold of 20%). High-priority requests (such as VIP customer and complaint requests) will remain in the original channel but will be assigned to a dedicated high-scoring customer service node. A unique session identifier (UUID) is generated for each cross-channel migration request, and interaction records (such as information provided by the customer and processing progress) are synchronized in real time through Redis. After the target channel customer service takes over, they can quickly obtain context through the session ID, avoiding the customer's repeated description of the problem. Monitor the queue length and response time of each channel in real time after migration. If the target channel load does not drop significantly within 5 minutes (for example, the queue length only decreases by 10%), trigger a second migration. The decision results of this load balancing (such as the migration channel and response time changes) are fed back to the Q-Learning model as training data to update the routing policy weights.

[0020] When generating dynamic routing instructions, high-priority customer requests are prioritized and assigned to the channel with the highest service score. At the same time, a cross-channel session status tracking identifier is established to ensure service continuity.

[0021] The multimodal data fusion module integrates text, voice, and image data, constructs a spatiotemporal correlation map to identify cross-channel behavior patterns, and triggers multi-dimensional verification (external interface calls and logical verification) for conflicting information to ensure data consistency. Establish a standardized data cleaning pipeline to perform feature extraction on heterogeneous interactive data such as text, voice, and images. The BERT model is used to extract semantic vectors from text data. Voice data is processed through voiceprint recognition and voice-to-text conversion. Convolutional neural networks are used to extract visual features from image data. Build a spatiotemporal correlation graph to correlate the interaction events generated by the same customer across different channels along a timeline, identify cross-channel behavior patterns, and activate the data credibility verification submodule when contradictory information is detected in multi-channel interactions. Specifically: Trigger conditions: Real-time analysis of spatiotemporal correlation graphs reveals inconsistent behavioral logic or data for the same customer across multiple channels (for example, the app displays "order completed," but a phone complaint states "goods not received"). Inconsistency determination rules include: timestamp conflicts (e.g., channel A records a time of 14:00, while channel B records a time of 14:05, but the content is inconsistent); behavioral logic conflicts (e.g., requesting a refund without submitting a return application); and data value conflicts (e.g., inconsistent payment amounts). The verification process includes: Check the legitimacy of the data source: Verify whether the channel has passed enterprise authentication (for example, whether the social media account is officially bound to the company); Anonymous or temporary conversation channels are marked as "low credibility" and require further verification; Time window analysis: Set a reasonable time difference threshold (for example, 5 minutes). If the timestamp of cross-channel events exceeds the threshold and the content is inconsistent, trigger an alarm; Behavioral logic verification: Call the business process rule engine to check whether the operation complies with the preset logic (for example, returns require an application to be submitted first); Third-party interface call, logistics status: Call the logistics company API (such as SF Express) to obtain the actual package receipt time and signatory information; Payment voucher: Connect to the Alipay / WeChat payment interface to verify the consistency of order amount and status; Physical evidence comparison: Image / video verification: Use hash algorithms or target detection models (such as YOLO) to compare the credentials uploaded by the customer with the system archive for consistency; Historical record retrieval: Analyze the customer's interaction records over the past 30 days to detect whether there are abnormal patterns (such as frequent conflicting complaints, high-risk tags); Behavioral profile matching: Compare current behavior with regular behavior (such as active time periods, commonly used channels). If the deviation exceeds the threshold (such as complaints from unfamiliar channels in the early morning), it is marked as "high risk"; Data integrity check, key field verification: Check whether the order number, customer ID, contact information, etc. are complete. If missing, trigger the automatic completion process (such as SMS requesting customers to supplement information); Data tampering detection: Verify whether the original data has been tampered with through hash value comparison (such as SHA-256); The verification result processing and decision-making process are as follows: Automatic adjudication: If external data (such as logistics APIs) returns a clear result, the system automatically corrects the conflicting data and updates the spatiotemporal graph; Manual intervention: When the external interface is unavailable or the verification result is ambiguous, a manual review work order is generated and assigned to the customer service team to contact the customer for verification; Risk management: After confirming false information, the customer account risk level is marked and operation permissions are restricted (such as prohibiting high-frequency work order submissions); the knowledge graph is updated to record conflicting events and processing results for subsequent AI model training.

[0022] When implementing real-time data fusion, the attention mechanism is used to dynamically adjust the data weight of each channel. If the same demand is generated in three consecutive channels, the feature weight will be automatically increased to 1.5 times. When generating a unified customer portrait, the forgetting factor algorithm is introduced to automatically perform weight decay processing on historical interaction data that exceeds the set time limit.

[0023] The intelligent feedback optimization module combines customer satisfaction evaluation with multimodal sentiment analysis (text, voice, and image), uses genetic algorithms to optimize service strategies, and synchronizes them to the cross-channel knowledge graph for adaptive iteration. Satisfaction evaluation data is collected after each customer service closed loop. At the same time, the sentiment analysis model is used to automatically analyze the emotional change curve during the customer interaction process. The specific process is as follows: Collect omnichannel customer interaction data, including text (online chat, email), voice (converting call recordings to text), and images (such as screenshots of emojis in customer service notes). Remove noise (such as advertising links and special symbols), perform word segmentation (using Jieba for Chinese), and filter out stop words. Use ASR (such as Alibaba Cloud Speech Recognition) to convert recordings to text and annotate with timestamps. Convert recordings to lowercase and handle abbreviations (such as "pls" → "please"). Use a pre-trained VADER model (for short social media text) or a fine-tuned BERT model (for specialized domains such as e-commerce customer service) to output a sentiment polarity score (-1 to 1) for each sentence. For example, "Terrible service!" → -0.8 (negative); "Issue solved, thank you!" → 0.7 (positive). Arrange sentiment scores in chronological order of interaction to generate time series data. Smooth the data using interpolation methods (e.g., linear interpolation) to plot sentiment change curves, marking key nodes (e.g., lows and peaks). Generate a sentiment curve for the entire interaction, annotating events that trigger sentiment fluctuations (e.g., long wait times, problem resolution). Associate the sentiment scores with the original interaction data and store them in MongoDB for subsequent analysis. Then, by integrating multimodal data such as text, speech, and images, we can improve the accuracy of sentiment analysis and the ability to understand context. The process is as follows: Use the Librosa library to extract the pitch, speech rate, and energy of speech signals, train a CNN-based speech emotion model (such as pre-trained on the RAVDESS dataset), and output classification labels such as anger, calmness, and anxiety. Use the ResNet model to classify emojis and screenshots sent by customers, and use Tesseract to extract text from images (such as handwritten customer feedback) and incorporate it into the text analysis process. Map text sentiment scores, voice sentiment labels, and image classification results to a unified dimension (e.g., a scale of -1 to 1). If the voice anger label matches the text negativity score, the weight is increased by 1.2 times. If the emoticon contradicts the text (e.g., text "satisfied" but an unsatisfied emoticon is sent), conflict detection is triggered, and the manual review weight is increased by 50%. Use TF-IDF or BERT to extract keywords from each round of conversation (e.g., "refund," "logistics") and correlate emotional fluctuations. Input multimodal emotional sequences to predict subsequent emotional trends (e.g., the customer may become angry again in 10 minutes).

[0024] Build a dual-channel feedback analysis model to compare and verify manual evaluation data with machine sentiment analysis results. When the difference between the two exceeds 30%, the manual review mechanism is triggered; Utilizing an adaptive strategy optimization engine and a genetic algorithm to dynamically adjust the response priority weights of each channel, automatically eliminating parameter combinations that have achieved less than 5% improvement over three consecutive optimization cycles. Establish a cross-channel knowledge graph update mechanism to synchronize optimized service strategies to the decision databases of all access channels to ensure the real-time and consistency of strategy updates.

[0025] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-channel interactive customer relationship management system, characterized in that: include: A dynamic routing decision module collects multi-channel customer requests in real time, evaluates channel loads through deep neural networks, and dynamically allocates requests to the optimal node using reinforcement learning to achieve load balancing and service continuity. A multimodal data fusion module integrates text, voice, and image data, constructs spatiotemporal correlation maps to identify cross-channel behavior patterns, and triggers multi-dimensional verification of conflicting information to ensure data consistency. The intelligent feedback optimization module is used to combine customer satisfaction evaluation with multimodal sentiment analysis, use genetic algorithms to optimize service strategies, and synchronize them to the cross-channel knowledge graph to achieve adaptive iteration.

2. A multi-channel interactive customer relationship management system according to claim 1, characterized in that: The execution steps of the dynamic routing decision module are as follows: Collect customer interaction requests from various channels such as websites, mobile applications, social media, and call centers in real time, and extract request features including interaction channel type, request content semantic features, customer identification code, and request initiation timestamp; A channel load assessment model is built based on a deep neural network. The model inputs the current processing queue length, historical average response time, and current number of online customer service personnel of each service node, and outputs a real-time service capability score for each channel. A dynamic routing strategy is established using a reinforcement learning algorithm. The routing weight coefficient is calculated based on the matching degree between customer request characteristics and channel service capability scores. When the request backlog of a channel exceeds the threshold, a cross-channel load balancing mechanism is automatically triggered. When generating dynamic routing instructions, high-priority customer requests are prioritized and assigned to the channel with the highest service score. At the same time, a cross-channel session status tracking identifier is established to ensure service continuity.

3. A multi-channel interactive customer relationship management system according to claim 2, characterized in that: The specific steps for triggering the cross-channel load balancing mechanism in the dynamic routing decision module are as follows: Set dynamic thresholds based on historical data; Real-time monitoring of request backlogs and service capability scores across channels; When the backlog of a channel exceeds the threshold and the service capability score is lower than the average, the load balancing mechanism is triggered; Filter candidate channels from currently available channels with a service capability score greater than 0.7 and a load queue length less than 30; If multiple channels meet the conditions, the channel with the highest matching degree with the customer request type will be selected first; The routing weight coefficient generated by the Q-Learning algorithm is used to calculate the comprehensive score of the candidate channel: comprehensive score = service capability score × 0.6 + routing weight coefficient × 0.4; Select the channel with the highest comprehensive score as the target allocation node; Migrate non-urgent requests in the backlog channel to the target channel in proportion, and the migration ratio is dynamically adjusted according to the backlog level; High-priority requests, including VIP customer and complaint requests, remain in the original channel but are assigned to a customer service node with a score higher than the preset standard; Generate a unique session identifier for each cross-channel migration request and synchronize interaction records in real time through Redis; After the target channel customer service takes over, they can quickly obtain context through the session ID to avoid customers repeating the problem description; Monitor the queue length and response time of each channel in real time after migration. If the load on the target channel does not drop as expected within 5 minutes, trigger a second migration. The load balancing decision result is fed back to the Q-Learning model as training data to update the routing strategy weights.

4. A multi-channel interactive customer relationship management system according to claim 1, characterized in that: The execution steps of the multimodal data fusion module are as follows: Establish a standardized data cleaning pipeline to perform feature extraction on heterogeneous interaction data such as text, voice, and image. The BERT model is used to extract semantic vectors from text data. Voice data is processed through voiceprint recognition and voice-to-text conversion. Convolutional neural networks are used to extract visual features from image data. Build a spatiotemporal correlation graph to correlate the interaction events generated by the same customer across different channels along a timeline, identify cross-channel behavior patterns, and activate the data credibility verification submodule when contradictory information is detected in multi-channel interactions; When implementing real-time data fusion, the attention mechanism is used to dynamically adjust the weight of each channel's data. If the same demand is generated from three consecutive channels, the weight of the feature is automatically increased to 1.5 times; When generating a unified customer profile, a forgetting factor algorithm is introduced to automatically perform weight decay on historical interaction data that exceeds the set time limit.

5. A multi-channel interactive customer relationship management system according to claim 4, characterized in that: The specific steps for detecting conflicting information in multi-channel interactions in the dynamic routing decision module and starting the data credibility verification submodule are as follows: Trigger conditions: Real-time analysis of spatiotemporal correlation graphs reveals inconsistent behavior logic or data for the same customer across multiple channels; Conflict determination rules: timestamp conflicts, behavioral logic conflicts, and data value conflicts; The verification process is as follows: Check the legitimacy of the data source: verify whether the channel has passed enterprise certification; Anonymous or temporary conversation channels are marked as low credibility and continue to be verified; Time window analysis: Set a time difference threshold. If the timestamps of cross-channel events exceed the threshold and the content is inconsistent, an alert will be triggered. Behavioral logic verification: Call the business process rule engine to check whether the operation complies with the preset logic; Third-party interface call, logistics status: call the logistics company API to obtain the actual package receipt time and signatory information; Payment voucher: connect to Alipay or WeChat payment interface to verify whether the order amount and status are consistent; Physical evidence comparison includes image or video verification: using hash algorithms or object detection models to compare the credentials uploaded by the customer with those in the system archive; History retrieval: Analyzes customer interaction records from the past 30 days to detect any unusual patterns. Behavioral profile matching: Compare current behavior with normal behavior. If the deviation exceeds the threshold, it is marked as high risk. Data integrity checks include key field verification: checking whether the order number, customer ID, and contact information are complete. If missing, the automatic completion process is triggered; Data tampering detection: Verify whether the original data has been tampered with by comparing the hash values; The verification result processing and decision-making process is as follows: Automatic adjudication: If external data returns a clear result, it automatically corrects the conflicting data and updates the spatiotemporal map; Manual intervention: When the external interface is unavailable or the verification result is ambiguous, a manual review ticket is generated and assigned to the customer service team to contact the customer for verification; Risk management: After confirming false information, mark the customer account risk level and restrict operation permissions; Update the knowledge graph to record conflicting events and handling results.

6. A multi-channel interactive customer relationship management system according to claim 1, characterized in that: The specific operation steps of the intelligent feedback optimization module are as follows: Satisfaction evaluation data is collected after each customer service closed loop, and the sentiment analysis model is used to automatically analyze the emotional change curve during customer interaction; Build a dual-channel feedback analysis model to compare and verify manual evaluation data with machine sentiment analysis results. When the difference between the two exceeds 30%, the manual review mechanism is triggered; Utilizing an adaptive strategy optimization engine and a genetic algorithm to dynamically adjust the response priority weights of each channel, automatically eliminating parameter combinations that have achieved less than 5% improvement over three consecutive optimization cycles. Establish a cross-channel knowledge graph update mechanism to synchronize optimized service strategies to the decision databases of all access channels to ensure the real-time and consistency of strategy updates.

7. A multi-channel interactive customer relationship management system according to claim 6, characterized in that: The specific operation steps of the intelligent feedback optimization module to automatically analyze the emotional change curve during customer interaction through the sentiment analysis model are as follows: Collect omnichannel customer interaction data, including text, voice, and images; Noise removal, word segmentation, and stop word filtering; Convert recordings to text using ASR and add timestamps; Convert to lowercase and handle abbreviations; Use the pre-trained VADER model or BERT fine-tuned model to output a sentiment polarity score between -1 and 1 for each sentence; Arrange sentiment scores in chronological order of interaction to generate time series data; Smooth the data through interpolation, draw a graph of sentiment changes, and mark key nodes; Generate an emotional curve for the entire interaction process, annotate events that trigger emotional fluctuations, associate the emotional score with the original interaction data, and store it in MongoDB for subsequent analysis; Then, by integrating multimodal data such as text, voice, and images, the accuracy of sentiment analysis and the ability to understand context can be improved.

8. A multi-channel interactive customer relationship management system according to claim 7, characterized in that: The specific operation steps of the intelligent feedback optimization module by fusing text, voice, and image multimodal data are as follows: Use the Librosa library to extract the fundamental frequency, speech rate, and energy of the speech signal, train a CNN-based speech emotion model, and output classification labels for anger, calmness, and anxiety. Use the ResNet model to classify the emojis or screenshots sent by customers, extract the text in the images using Tesseract, and incorporate it into the text analysis process; Map text sentiment scores, speech sentiment labels, and image classification results into a unified dimension; If the voice anger label is consistent with the text negativity score, the weight is increased to 1.2 times; If the emoji and text are inconsistent, a conflict detection is triggered and the manual review weight is increased by 50%; Extract keywords from each round of conversation using TF-IDF or BERT and correlate them with sentiment fluctuations; Input multimodal emotion sequences and predict subsequent emotion trends.

Images