An ai-driven intelligent analysis system and method for user session data
By using an AI-driven intelligent analysis system for user conversation data, combined with particle swarm optimization and ant colony optimization algorithms to optimize customer service resources, the system solves the problems of data fragmentation and single scheduling in conversation analysis of chain brand stores, and achieves efficient service response and resource utilization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG LETEN TECH DEV CO LTD
- Filing Date
- 2026-04-15
- Publication Date
- 2026-07-14
Smart Images

Figure CN122390792A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent analysis of user session data, and in particular to an AI-driven intelligent analysis system and method for user session data. Background Technology
[0002] To gain deeper insights into customer needs and communication efficiency, and to improve communication with missed customers in the store service process, it is necessary to conduct intelligent analysis and tracking management of customer conversations for chain brands. Through functions such as real-time monitoring, intelligent analysis, and data visualization, stores can be helped to improve customer service efficiency, optimize customer experience, and assist in store operational decisions. Summary of the Invention
[0003] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes an AI-driven intelligent analysis system and method for user session data, which can improve the automation level of session analysis, service response efficiency, and resource utilization efficiency in interconnected scenarios.
[0004] One embodiment of the present invention provides an AI-driven intelligent analysis system for user conversation data, comprising: a real-time conversation acquisition module for real-time acquisition of customer conversation data from multiple stores of a chain brand, wherein the customer conversation data includes at least text content, timestamps, store IDs, and customer group tags, and constructs a conversation dataset; a conversation feature analysis module for extracting semantic features and temporal behavioral features from customer conversations to obtain conversation feature vectors for each store and each customer group, and constructing a global baseline feature vector using the mean of the feature vectors of each subject; and a difference significance discrimination module for calculating the degree of difference between the conversation feature vector of a single store or group and the global baseline vector using a difference significance discrimination model; wherein the difference significance discrimination model is based on cosine similarity and F-test statistics to determine the significance of differences. The system is designed to identify stores or groups of people with significant specificity when the cosine similarity is below a threshold and the F-value is above a critical value. A multi-factor attribution localization module calculates the feature contribution of specific target groups, obtains the attribution weights of each influencing factor, and identifies the Top-N core causes. An intelligent strategy generation module, including a swarm intelligence optimization unit, constructs a fitness function based on particle swarm optimization, aiming to minimize the non-response rate, minimize manpower costs, and maximize service response rate. It iteratively optimizes the particle swarm algorithm to output the optimal customer service seat and service resource allocation scheme. Based on ant colony optimization, it constructs a reception priority path for non-response messages, iteratively solves for the optimal reception ranking and service process optimization path using pheromone update rules, and outputs the globally optimal service scheduling strategy and differentiated adjustment scheme.
[0005] According to some embodiments of the present invention, the system further includes an unanswered message statistics module, which is used to traverse the session data according to a preset time interval, classify and statistically analyze unanswered messages that exceed the preset reply time limit according to the store dimension and the timeout duration dimension, and synchronize the statistical results to the intelligent strategy generation module.
[0006] According to some embodiments of the present invention, the conversation feature analysis module includes a conversation semantic analysis unit, used to perform intent recognition and keyword extraction on customer consultation text through a pre-trained AI semantic model. The consultation content includes at least appointment time consultation and service type consultation. Based on the extraction results, high-frequency demand clustering is performed to identify the hot topic sets for each store and the whole. The conversation feature analysis module is also used to generate service process optimization feature data based on the hot topic set and synchronize it to the intelligent strategy generation module. The intelligent strategy generation module generates differentiated service adjustment measures that match the top-N core causes for target stores or target groups with significant specificity. The pre-trained AI semantic model is a BERT-like semantic model adjusted for appointment and service type consultation scenarios.
[0007] According to some embodiments of the present invention, the conversation feature analysis module includes a customer behavior time series analysis unit, which is used to perform time series statistics on the frequency of conversation interactions and the distribution of message interaction time for each store and each customer group, and extract customer active time period features and behavioral preference features; the intelligent strategy generation module is also used to generate online customer service manpower scheduling strategy based on customer active time period features, and generate targeted push strategy for preferential information based on customer behavioral preference features and low-activity time period distribution.
[0008] According to some embodiments of the present invention, in the difference significance discrimination module, the formula for calculating the F-test statistic is as follows: ;in The sum of variances between groups, Let be the sum of variances within groups, k be the number of stores or groups of people, and N be the total number of samples.
[0009] According to some embodiments of the present invention, the multi-factor attribution localization module uses the SHAP interpretability AI algorithm to calculate the feature contribution, and uses the SHAP value of each influencing factor as the attribution weight.
[0010] According to some embodiments of the present invention, the fitness function of the particle swarm optimization algorithm is a weighted normalization function:
[0011] ;
[0012] Where ω1, ω2, and ω3 are preset weighting coefficients, U is the non-response rate, Umax is the maximum allowable non-response rate, C is the manpower cost, Cmax is the maximum allowable manpower cost, R is the service response rate, and Rmax is the theoretical maximum response rate.
[0013] According to some embodiments of the present invention, the pheromone update rule of the ant colony algorithm is as follows: Where ρ is the pheromone evaporation coefficient, Let be the path pheromone concentration at time t. This represents the pheromone increment released by the k-th ant on the corresponding path.
[0014] According to some embodiments of the present invention, the customer demographic tags include at least one or more of the following: age dimension tags, spending power tags, customer origin tags, and historical service preference tags.
[0015] Another aspect of this invention provides an AI-driven intelligent analysis method for user conversation data, used in the system described in any one of claims 1 to 9, comprising the following steps: S100, real-time collection of customer conversation data from multiple stores of a chain brand, wherein the customer conversation data includes at least text content, timestamps, store IDs, and customer group tags, and constructs a conversation dataset; S200, extraction of semantic features and temporal behavioral features from customer conversations to obtain conversation feature vectors for each store and each customer group, and construction of a global baseline feature vector using the mean of each subject feature vector; S300, calculation of the degree of difference between the conversation feature vector of a single store or group and the global baseline vector using a difference significance discrimination model; wherein the difference significance discrimination model is based on cosine similarity. The F-test statistic is used to determine the significance of differences. When the cosine similarity is below the threshold and the F-value is above the critical value, the store or group is marked as a target subject with significant specificity. S400: The feature contribution of the specific target subject is calculated to obtain the attribution weight of each influencing factor and locate the top-N core causes. S500: Based on the particle swarm optimization algorithm, a fitness function is constructed with multiple objectives of minimizing the non-response rate, minimizing manpower costs, and maximizing the service response rate. The optimal customer service seat and service resource allocation scheme is output through particle iteration optimization. Based on the ant colony algorithm, a reception priority path is constructed for non-response messages. The optimal reception ranking and service process optimization path are solved iteratively using the pheromone update rule. The globally optimal service scheduling strategy and differentiated adjustment scheme are output.
[0016] The embodiments of this invention achieve at least the following beneficial effects: By performing unified feature extraction and global baseline vector construction on multi-store chain conversation data, the embodiments of this invention accurately identify the true specificity of stores and groups using a dual discriminant model combining cosine similarity and F-test. Then, through multi-factor attribution, the core causes are automatically located, and by relying on a hybrid particle swarm intelligence algorithm and ant colony intelligence algorithm, multi-objective optimization of customer service resources, optimal sorting of unanswered messages, and intelligent scheduling of service processes are achieved. This effectively solves problems such as fragmented multi-store data, misjudgment of differences, subjective human attribution, and single scheduling rules, significantly improving the accuracy of feature identification and cause location, realizing the output of globally optimal differentiated service strategies, and significantly improving the automation level of conversation analysis, service response efficiency, and resource utilization efficiency in chain scenarios. The system has stronger scalability and industrial applicability.
[0017] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description
[0018] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:
[0019] Figure 1 This is a schematic block diagram of the system modules according to an embodiment of the present invention;
[0020] Figure 2 This is a flowchart illustrating the method according to an embodiment of the present invention.
[0021] Figure label:
[0022] The system includes a real-time conversation acquisition module 100, a conversation feature analysis module 200, a difference significance discrimination module 300, a multi-factor attribution localization module 400, and an intelligent strategy generation module 500. Detailed Implementation
[0023] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0024] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.
[0025] Reference Figure 1 This invention proposes an AI-driven intelligent analysis system for user session data, comprising:
[0026] The real-time conversation acquisition module 100 is used to collect customer conversation data from multiple stores of a chain brand in real time. The customer conversation data includes at least text content, timestamps, store IDs, and customer group tags, and is constructed into a conversation dataset.
[0027] The conversation feature analysis module 200 is used to extract semantic features and temporal behavioral features from customer conversations, obtain conversation feature vectors for each store and each customer group, and construct a global benchmark feature vector based on the mean of the feature vectors of each subject.
[0028] The difference significance discrimination module 300 is used to calculate the degree of difference between the conversation feature vector of a single store or group and the global baseline vector through the difference significance discrimination model. The difference significance discrimination model is based on cosine similarity and F test statistic to determine the difference significance. When the cosine similarity is lower than the threshold and the F value is greater than the critical value, the store or group is marked as a target subject with significant specificity.
[0029] The multi-factor attribution localization module 400 is used to calculate the feature contribution of specific target subjects, obtain the attribution weight of each influencing factor, and locate the top-N core causes.
[0030] The intelligent strategy generation module 500 includes a swarm intelligence optimization unit. Based on the particle swarm optimization algorithm, it constructs a fitness function with multiple objectives, including minimizing the non-response rate, minimizing manpower costs, and maximizing the service response rate. It then outputs the optimal customer service agent and service resource allocation scheme through particle iterative optimization. Based on the ant colony algorithm, it constructs a reception priority path for non-response messages and uses pheromone update rules to iteratively solve for the optimal reception order and service process optimization path. Finally, it outputs the globally optimal service scheduling strategy and differentiated adjustment scheme.
[0031] In some embodiments, the system of the present invention further includes an unanswered message statistics module, which is used to traverse the session data at preset time intervals, classify and statistically analyze unanswered messages that exceed the preset reply time limit by store dimension and timeout duration dimension, and synchronize the statistical results to the intelligent strategy generation module.
[0032] In some embodiments, the conversation feature analysis module 200 includes a conversation semantic analysis unit, which is used to perform intent recognition and keyword extraction on customer consultation text through a pre-trained AI semantic model. The consultation content includes at least appointment time consultation and service type consultation. Based on the extraction results, high-frequency demand clustering is performed to identify the hot topic sets for each store and the whole. The conversation feature analysis module 200 is also used to generate service process optimization feature data based on the hot topic set and synchronize it to the intelligent strategy generation module 500.
[0033] In some embodiments, the pre-trained AI semantic model is a BERT-like semantic model adapted for appointment and service type consultation scenarios.
[0034] In some embodiments, the intelligent strategy generation module 500 generates differentiated service adjustment measures that match the core causes of Top-N for target stores or target groups with significant specificity.
[0035] In some embodiments, the conversation feature analysis module 200 includes a customer behavior time series analysis unit, which is used to perform time series statistics on the frequency of conversation interactions and the distribution of message interaction time for each store and each customer group, and extract customer active time period features and behavioral preference features.
[0036] In some embodiments, the intelligent strategy generation module 500 is also used to generate an online customer service manpower dispatch strategy based on the characteristics of customer active time periods, and to generate a targeted push strategy for preferential information based on customer behavior preference characteristics and the distribution of low-activity time periods.
[0037] In some embodiments, the formula for calculating the F-test statistic in the difference significance discrimination module 300 is:
[0038] ;
[0039] in The sum of variances between groups, Let be the sum of variances within groups, k be the number of stores or groups of people, and N be the total number of samples.
[0040] In some embodiments, the multi-factor attribution localization module 400 uses the SHAP interpretability AI algorithm to calculate the feature contribution, and uses the SHAP value of each influencing factor as the attribution weight.
[0041] In some embodiments, the fitness function of the particle swarm optimization algorithm is a weighted normalization function:
[0042] ;
[0043] Where ω1, ω2, and ω3 are preset weighting coefficients, U is the non-response rate, Umax is the maximum allowable non-response rate, C is the manpower cost, Cmax is the maximum allowable manpower cost, R is the service response rate, and Rmax is the theoretical maximum response rate.
[0044] In some embodiments, the pheromone update rule of the ant colony algorithm is as follows:
[0045] ;
[0046] Where ρ is the pheromone evaporation coefficient. Let be the path pheromone concentration at time t. This represents the pheromone increment released by the k-th ant on the corresponding path.
[0047] In some embodiments, customer demographics may include at least one or more of the following: age, spending power, customer origin, and historical service preference.
[0048] Reference Figure 2 This invention proposes an AI-driven intelligent analysis method for user session data, comprising the following steps:
[0049] S100: Real-time collection of customer conversation data from multiple stores of a chain brand. The customer conversation data includes at least text content, timestamps, store IDs, and customer demographic tags, and is constructed into a conversation dataset.
[0050] S200. Extract semantic features and temporal behavioral features from customer conversations to obtain conversation feature vectors for each store and each customer group, and construct a global benchmark feature vector based on the mean of each subject feature vector.
[0051] S300. Calculate the degree of difference between the conversation feature vector of a single store or group and the global baseline vector using the difference significance discrimination model. The difference significance discrimination model is based on cosine similarity and F test statistic to determine the difference significance. When the cosine similarity is lower than the threshold and the F value is greater than the critical value, the store or group is marked as a target subject with significant specificity.
[0052] S400. Calculate the feature contribution of the specific target subject, obtain the attribution weight of each influencing factor, and locate the Top-N core causes.
[0053] S500, based on the particle swarm optimization algorithm, constructs a fitness function with multiple objectives of minimizing the non-response rate, minimizing manpower costs, and maximizing the service response rate. It then outputs the optimal customer service seat and service resource allocation scheme through particle iteration optimization. Based on the ant colony optimization algorithm, it constructs a reception priority path for non-response messages and uses pheromone update rules to iteratively solve for the optimal reception order and service process optimization path. Finally, it outputs the globally optimal service scheduling strategy and differentiated adjustment scheme.
[0054] Although specific embodiments are described herein, those skilled in the art will recognize that many other modifications or alternative embodiments are also within the scope of this disclosure. For example, any of the functions and / or processing capabilities described in connection with a particular device or component can be performed by any other device or component. Furthermore, while various exemplary embodiments and architectures have been described according to embodiments of this disclosure, those skilled in the art will recognize that many other modifications to the exemplary embodiments and architectures described herein are also within the scope of this disclosure.
[0055] The foregoing description, with reference to block diagrams and flowcharts of systems, methods, systems, and / or computer program products according to exemplary embodiments, has described certain aspects of this disclosure. It should be understood that one or more blocks in the block diagrams and flowcharts, as well as combinations of blocks in the block diagrams and flowcharts, can be implemented by executing computer-executable program instructions, respectively. Similarly, according to some embodiments, some blocks in the block diagrams and flowcharts may not need to be executed in the order shown, or may not all need to be executed. Furthermore, additional components and / or operations beyond those shown in the blocks in the block diagrams and flowcharts may exist in some embodiments.
[0056] Therefore, blocks in block diagrams and flowcharts support combinations of means for performing a specified function, combinations of elements or steps for performing a specified function, and program instruction means for performing a specified function. It should also be understood that each block in a block diagram and flowchart, and combinations of blocks in block diagrams and flowcharts, can be implemented by a dedicated hardware computer system or a combination of dedicated hardware and computer instructions that performs a specific function, element, or step.
[0057] The program modules, applications, etc., described herein may include one or more software components, including, for example, software objects, methods, data structures, etc. Each such software component may include computer-executable instructions that, in response to execution, cause at least a portion of the functionality described herein (e.g., one or more operations of the exemplary methods described herein) to be performed.
[0058] Software components can be coded using any of a variety of programming languages. An exemplary programming language could be a low-level programming language, such as assembly language associated with a specific hardware architecture and / or operating system platform. Software components including assembly language instructions may need to be converted into executable machine code by an assembler before being executed by the hardware architecture and / or platform. Another exemplary programming language could be a higher-level programming language that is portable across multiple architectures. Software components including higher-level programming languages may need to be converted into an intermediate representation by an interpreter or compiler before execution. Other examples of programming languages include, but are not limited to, macro languages, shell or command languages, job control languages, scripting languages, database query or search languages, or report writing languages. In one or more exemplary embodiments, a software component containing instructions from one of the above-described programming language examples can be executed directly by the operating system or other software components without first being converted into another form.
[0059] Software components can be stored as files or other data storage structures. Software components of similar type or related function can be stored together in a specific directory, folder, or library. Software components can be static (e.g., pre-defined or fixed) or dynamic (e.g., created or modified at runtime).
[0060] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. An AI-driven intelligent analysis system for user conversation data, characterized in that, include: The real-time conversation acquisition module is used to collect customer conversation data from multiple stores of a chain brand in real time. The customer conversation data includes at least text content, timestamps, store IDs, and customer group tags, and is constructed into a conversation dataset. The conversation feature analysis module is used to extract semantic features and temporal behavioral features from customer conversations, obtain conversation feature vectors for each store and each customer group, and construct a global baseline feature vector based on the mean of the feature vectors of each subject. The difference significance discrimination module is used to calculate the degree of difference between the conversation feature vector of a single store or group and the global baseline vector through the difference significance discrimination model. The difference significance discrimination model is based on cosine similarity and F test statistic to determine the difference significance. When the cosine similarity is lower than the threshold and the F value is greater than the critical value, the store or group is marked as a target subject with significant specificity. The multi-factor attribution localization module is used to calculate the feature contribution of specific target subjects, obtain the attribution weight of each influencing factor, and locate the top-N core causes. The intelligent strategy generation module includes a swarm intelligence optimization unit. Based on the particle swarm optimization algorithm, it constructs a fitness function with multiple objectives: minimizing the non-response rate, minimizing manpower costs, and maximizing the service response rate. Through particle iteration optimization, it outputs the optimal customer service agent and service resource allocation scheme. Based on the ant colony optimization algorithm, it constructs a reception priority path for non-response messages and uses pheromone update rules to iteratively solve for the optimal reception order and service process optimization path. Finally, it outputs the globally optimal service scheduling strategy and differentiated adjustment scheme.
2. The AI-driven intelligent analysis system for user session data according to claim 1, characterized in that, The system also includes an unanswered message statistics module, which is used to traverse the session data at preset time intervals, classify and statistically analyze unanswered messages that exceed the preset reply time limit by store dimension and timeout duration dimension, and synchronize the statistical results to the intelligent strategy generation module.
3. The AI-driven intelligent analysis system for user session data according to claim 1, characterized in that, The conversation feature analysis module includes a conversation semantic analysis unit, which is used to identify intent and extract keywords from customer consultation texts through a pre-trained AI semantic model. The consultation content includes at least appointment time consultation and service type consultation. Based on the extraction results, high-frequency demand clustering is performed to identify the hot topic sets for each store and the whole. The conversation feature analysis module is also used to generate service process optimization feature data based on the hot topic set and synchronize it to the intelligent strategy generation module. The intelligent strategy generation module generates differentiated service adjustment measures that match the core causes of the Top-N targets for stores or target groups with significant specificity. The pre-trained AI semantic model is a BERT-like semantic model adjusted for appointment and service type consultation scenarios.
4. The AI-driven intelligent analysis system for user session data according to claim 1, characterized in that, The conversation feature analysis module includes a customer behavior time series analysis unit, which is used to perform time series statistics on the frequency of conversation interactions and the distribution of message interaction time for each store and each customer group, and extract customer active time period characteristics and behavioral preference characteristics; the intelligent strategy generation module is also used to generate online customer service manpower scheduling strategies based on customer active time period characteristics, and generate targeted push strategies for preferential information based on customer behavioral preference characteristics and the distribution of low-activity time periods.
5. The AI-driven intelligent analysis system for user session data according to claim 1, characterized in that, In the difference significance discrimination module, the formula for calculating the F-test statistic is: ; in The sum of variances between groups, Let be the sum of variances within groups, k be the number of stores or groups of people, and N be the total number of samples.
6. The AI-driven intelligent analysis system for user session data according to claim 1, characterized in that, The multi-factor attribution localization module uses the SHAP interpretability AI algorithm to calculate the feature contribution, and uses the SHAP value of each influencing factor as the attribution weight.
7. The AI-driven intelligent analysis system for user session data according to claim 1, characterized in that, The fitness function of the particle swarm optimization algorithm is a weighted normalization function: ; Where ω1, ω2, and ω3 are preset weighting coefficients, U is the non-response rate, Umax is the maximum allowable non-response rate, C is the manpower cost, Cmax is the maximum allowable manpower cost, R is the service response rate, and Rmax is the theoretical maximum response rate.
8. The AI-driven intelligent analysis system for user session data according to claim 1, characterized in that, The pheromone update rule of the ant colony algorithm is as follows: ; Where ρ is the pheromone evaporation coefficient. Let be the path pheromone concentration at time t. This represents the pheromone increment released by the k-th ant on the corresponding path.
9. The AI-driven intelligent analysis system for user session data according to claim 1, characterized in that, The customer demographic tags include at least one or more of the following: age, spending power, customer origin, and historical service preference.
10. An AI-driven intelligent analysis method for user session data, used in the system as described in any one of claims 1 to 9, characterized in that, Includes the following steps: S100. Real-time collection of customer conversation data from multiple stores of a chain brand. The customer conversation data includes at least text content, timestamps, store IDs, and customer demographic tags, and is constructed into a conversation dataset. S200. Extract semantic features and temporal behavioral features from customer conversations to obtain conversation feature vectors for each store and each customer group, and construct a global benchmark feature vector based on the mean of each subject feature vector. S300. Calculate the degree of difference between the conversation feature vector of a single store or group and the global baseline vector using a difference significance discrimination model; the difference significance discrimination model is based on cosine similarity and F test statistic to determine the difference significance. When the cosine similarity is lower than the threshold and the F value is greater than the critical value, the store or group is marked as a target subject with significant specificity. S400. Calculate the feature contribution of the specific target subject, obtain the attribution weight of each influencing factor, and locate the Top-N core causes. S500, based on the particle swarm optimization algorithm, constructs a fitness function with multiple objectives of minimizing the non-response rate, minimizing manpower costs, and maximizing the service response rate. It then outputs the optimal customer service seat and service resource allocation scheme through particle iteration optimization. Based on the ant colony optimization algorithm, it constructs a reception priority path for non-response messages and uses pheromone update rules to iteratively solve for the optimal reception order and service process optimization path. Finally, it outputs the globally optimal service scheduling strategy and differentiated adjustment scheme.