Data processing method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to an artificial intelligence-based data processing method, which comprises the following steps of: obtaining behavior data of a target user; performing data cleaning and sorting processing on the behavior data to obtain a user behavior sequence; performing time sequence anomaly detection on the user behavior sequence based on a time sequence anomaly detection model to obtain a time sequence anomaly detection result; performing path conversion analysis on the user behavior sequence to obtain a path conversion analysis result; generating breakpoint detection data based on the time sequence anomaly detection result and the path conversion analysis result; performing suggestion generation processing on the breakpoint detection data based on a preset target agent to obtain an optimization suggestion; and carrying out output processing on the breakpoint detection data and the optimization suggestions. The invention further provides a data processing device based on artificial intelligence, computer equipment and a storage medium. The method can be applied to a user behavior analysis scene in the field of financial science and technology, effectively improves the processing efficiency of breakpoint recognition, and improves the generation efficiency of optimization suggestions.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and can be applied to the financial technology field, particularly to data processing methods, devices, computer equipment and storage media based on artificial intelligence. Background Technology

[0002] In the financial and insurance sector, the digital transformation of customer service and product marketing has significantly improved service reach efficiency, with companies building a comprehensive service network through multiple channels such as official websites, mobile apps, call centers, and offline branches. However, existing data collection and process optimization technologies still face key bottlenecks: current customer behavior monitoring and breakpoint identification mainly rely on preset fixed rule engines, such as triggering alerts by setting single conditions like "page dwell timeout" or "form filling interruption." This approach has significant drawbacks: firstly, breakpoint identification depends on manually configured rules, making it difficult to dynamically adapt to complex scenarios (such as differences in claims processes across different insurance types), resulting in a high false alarm rate; secondly, the optimization suggestion generation mechanism is lagging, typically requiring manual analysis of breakpoint data to develop improvement plans, a process that can take weeks and cannot respond to customer needs in real time.

[0003] Taking claims consultation in the financial sector as an example: When a customer initiates a critical illness insurance claim through an insurance company's app, the system can only record the interruption in filling out the "Disease Type Selection" form, but it cannot link it to the customer's historical data of concealing their diabetes history in the "Health Disclosure" section, nor can it identify potential risk signals from the customer's subsequent follow-up inquiries about "claim rejection conditions" through the call center. This isolated data collection method makes it impossible for companies to accurately locate service breakpoints (such as the real reason for incomplete material submission being inaccurate health disclosure), and it is also difficult to generate targeted optimization suggestions, ultimately leading to a chain of problems such as increased customer complaint rates and decreased repurchase rates.

[0004] Therefore, there is an urgent need for a customer service optimization method based on intelligent analysis to automatically identify breakpoints and generate real-time optimization suggestions, thereby solving the core pain points of inefficient breakpoint identification and lagging optimization suggestions in existing technologies. Summary of the Invention

[0005] The purpose of this application is to propose a data processing method, apparatus, computer equipment, and storage medium based on artificial intelligence, in order to solve the technical problems of existing customer behavior monitoring and breakpoint identification mainly relying on preset fixed rule engines, which result in inefficient breakpoint identification and lagging optimization suggestions.

[0006] Firstly, an artificial intelligence-based data processing method is provided, including: Obtain behavioral data of target users from pre-defined channels; The behavioral data is cleaned and sorted to obtain the corresponding user behavior sequence; The user behavior sequence is subjected to time-series anomaly detection based on a preset time-series anomaly detection model to obtain the corresponding time-series anomaly detection results. Perform path transformation analysis on the user behavior sequence to obtain the corresponding path transformation analysis results; Based on the temporal anomaly detection results and the path transformation analysis results, corresponding breakpoint detection data is generated; Based on a preset target intelligent agent, the breakpoint detection data is processed to generate suggestions, resulting in corresponding optimization suggestions. The breakpoint detection data and the optimization suggestions are then processed and output.

[0007] Secondly, an artificial intelligence-based data processing device is provided, comprising: The acquisition module is used to acquire behavioral data of target users from preset channels; The processing module is used to perform data cleaning and sorting on the behavioral data to obtain the corresponding user behavior sequence; The detection module is used to perform time-series anomaly detection on the user behavior sequence based on a preset time-series anomaly detection model, and obtain the corresponding time-series anomaly detection results; The analysis module is used to perform path transformation analysis on the user behavior sequence and obtain the corresponding path transformation analysis results; The first generation module is used to generate corresponding breakpoint detection data based on the timing anomaly detection results and the path transformation analysis results. The second generation module is used to perform suggestion generation processing on the breakpoint detection data based on a preset target intelligent agent to obtain corresponding optimization suggestions. The output module is used to process the breakpoint detection data and the optimization suggestions.

[0008] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described artificial intelligence-based data processing method.

[0009] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned artificial intelligence-based data processing method.

[0010] In the aforementioned scheme implemented by the data processing method, apparatus, computer equipment, and storage medium based on artificial intelligence, the following steps are taken: First, behavioral data of the target user is obtained from a preset channel; then, the behavioral data is cleaned and sorted to obtain a corresponding user behavior sequence; next, a preset temporal anomaly detection model is used to detect temporal anomalies in the user behavior sequence to obtain a corresponding temporal anomaly detection result; then, path transformation analysis is performed on the user behavior sequence to obtain a corresponding path transformation analysis result; subsequently, corresponding breakpoint detection data is generated based on the temporal anomaly detection result and the path transformation analysis result; subsequently, suggestion generation processing is performed on the breakpoint detection data based on a preset target intelligent agent to obtain corresponding optimization suggestions; finally, the breakpoint detection data and the optimization suggestions are output. Based on the above automated processing flow, this application obtains user behavior sequences by cleaning and sorting behavioral data. Then, it uses a temporal anomaly detection model to detect temporal anomalies in the user behavior sequences, obtaining detection results. Next, it performs path transformation analysis on the user behavior sequences, yielding analysis results. Finally, based on the temporal anomaly detection results and path transformation analysis results, it automatically generates breakpoint detection data. Subsequently, based on the use of a target intelligent agent, it automatically generates optimization suggestions from the breakpoint detection data. Thus, through the collaborative processing of the temporal anomaly detection model, path transformation analysis, and the target intelligent agent, this application can automatically and accurately achieve full-process behavior monitoring, breakpoint identification, and optimization suggestion generation, effectively improving the processing efficiency of breakpoint identification and the efficiency of optimization suggestion generation, while ensuring the accuracy of the generated breakpoint detection data and optimization suggestions. Attached Figure Description

[0011] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of an embodiment of the artificial intelligence-based data processing method according to this application; Figure 3 This is a schematic diagram of a structure of an embodiment of the artificial intelligence-based data processing apparatus according to this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0017] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0018] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0019] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0020] It should be noted that the artificial intelligence-based data processing method provided in the embodiments of this application is generally executed by a server / terminal device, and correspondingly, the artificial intelligence-based data processing device is generally set in the server / terminal device.

[0021] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0022] Continue to refer to Figure 2 This document illustrates a flowchart of an embodiment of the AI-based data processing method according to this application. The order of steps in the flowchart can be changed, and some steps can be omitted, depending on different needs. The AI-based data processing method provided in this application can be applied to any scenario requiring data processing, and therefore can be applied to products in these scenarios, such as data processing products in the financial and insurance fields. The AI-based data processing method includes the following steps: Step S201: Obtain the target user's behavioral data from a preset channel.

[0023] In this embodiment, the data processing method based on artificial intelligence runs on an electronic device (e.g., Figure 1The server / terminal device shown can acquire target user behavior data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future-developed wireless connection methods. The executing entity of this application is specifically a data processing system, which can be simply referred to as the system. The aforementioned preset channel specifically refers to the multi-source data collection agent layer pre-built in the system. This multi-source data collection agent layer deploys multiple distributed agents on the official website, App, customer service system, and offline terminals to automatically track customer interaction behavior. The collection technologies employed by the multi-source data collection agent layer may include: using an event stream tracing mechanism to capture events such as clicks, dwell time, conversion paths, and voice commands; applying ASR and NLP parsing to voice and text interactions to extract semantic tags (such as "rejection of insurance" and "consultation and claims"); and streaming data to the central platform via Kafka or MQ to form a full-channel behavior log. In this multi-source data acquisition agent layer, each agent has lightweight edge learning capabilities, enabling local pre-aggregation of metrics (such as bounce rate and conversion funnel) to reduce central bandwidth pressure. Furthermore, all agents share unified identity mapping rules to ensure cross-channel tracking consistency.

[0024] Additionally, various behavioral data related to the target user's business journey can be extracted from the aforementioned omnichannel behavior logs based on the user ID (such as the user name). For example, page browsing records can be obtained from website server logs, operation steps and interaction times can be extracted from business operation record tables, and conversion event data can be collected from order management systems or business conversion tracking systems.

[0025] This application can be applied to user behavior analysis scenarios in the financial insurance field. For example, in a claims consultation scenario within the financial insurance field, such as a user initiating a claims application consultation through an insurance company's APP after being hospitalized due to illness, seeking information on the required materials and procedures, the collected user behavior data types can include: 1) Operational behavior: Entry path: Customers enter through "Homepage → Claims Entry → Online Consultation" or directly search for the keyword "claims". Page dwell time: Spending 2 minutes and 30 seconds on the "Claims Materials List" page, repeatedly reviewing the descriptions of "Medical Invoices" and "Diagnosis Certificates". Click behavior: Clicking on "Frequently Asked Questions" for "How long is the claims processing time?" and "How to reimburse medical expenses in other locations?". Form filling: Filling in "Disease Type" as "Acute Myocardial Infarction" in the consultation form, but not filling in "Number of Days of Hospitalization". 2) Interactive behavior: Online customer service dialogue: Customer asks: "How much compensation can I get in this situation?" (implying concern about the insured amount). After the customer service replies, the customer asks: "Do I need to provide the original medical records or a copy?" (a need to confirm the details of the materials). File Upload: Attempt to upload the "Hospitalization Summary" failed (incorrect file format), but was successfully uploaded again. 3) Emotion and Preference: Dialogue Emotion Analysis: NLP was used to detect customer anxiety in tone (e.g., "I need money urgently," "Please handle as soon as possible"). Service Channel Preference: After consultation, the customer chose "Telephone Follow-up" for further communication (instead of continuing online chat).

[0026] Step S202: Perform data cleaning and sorting on the behavioral data to obtain the corresponding user behavior sequence.

[0027] In this embodiment, the data cleaning and sorting process includes data cleaning and data sorting. Specifically, the data cleaning process includes: formulating data cleaning rules and checking the collected behavioral data one by one. For duplicate records caused by system failures, data missing key fields, and abnormal data caused by human error (such as extremely short interaction times, operation sequences that do not conform to business logic, etc.), these are marked and deleted or corrected. For example, for records missing customer IDs, if they cannot be supplemented through other means, they are removed from the dataset. Data cleaning removes noisy data and outliers, ensuring the accuracy and completeness of the data. Clean data can avoid misleading results in subsequent analysis and improve the accuracy of breakpoint detection.

[0028] The data sorting process includes: grouping each user's behavioral data according to the target user's identifier (such as user ID), and then sorting the behavioral data within each user group according to the interaction time, forming a user behavior sequence by user. For example, customer A's behavior sequence might be: browsing the homepage -> searching for insurance products -> viewing insurance product details -> adding to cart -> submitting an insurance order. Constructing the user behavior sequence involves organizing the scattered behavioral data in an orderly manner according to time and customer dimensions, providing a data structure that meets the model's requirements for subsequent temporal anomaly detection, enabling the model to better capture the temporal characteristics of user behavior.

[0029] Step S203: Perform time-series anomaly detection on the user behavior sequence based on a preset time-series anomaly detection model to obtain the corresponding time-series anomaly detection results.

[0030] In this embodiment, the specific implementation process of performing time-series anomaly detection on the user behavior sequence based on the preset time-series anomaly detection model to obtain the corresponding time-series anomaly detection result will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0031] Step S204: Perform path transformation analysis on the user behavior sequence to obtain the corresponding path transformation analysis results.

[0032] In this embodiment, the specific implementation process of performing path transformation analysis on the user behavior sequence to obtain the corresponding path transformation analysis results will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0033] Step S205: Generate corresponding breakpoint detection data based on the timing anomaly detection results and the path transformation analysis results.

[0034] In this embodiment, the specific implementation process of generating corresponding breakpoint detection data based on the timing anomaly detection results and the path transformation analysis results will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0035] Step S206: Based on the preset target intelligent agent, the breakpoint detection data is processed to generate suggestions, and corresponding optimization suggestions are obtained.

[0036] In this embodiment, the specific implementation process of generating suggestions based on the preset target intelligent agent on the breakpoint detection data to obtain corresponding optimization suggestions will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0037] Step S207: Output the breakpoint detection data and the optimization suggestions.

[0038] In this embodiment, the generated breakpoint detection data and optimization suggestions can be transmitted to relevant business personnel via email, text message, or interface display to help them quickly identify the problem links that need to be addressed first in complex business processes, thereby improving the efficiency of resolving potential breakpoint issues.

[0039] This application first obtains target user behavior data from preset channels; then cleans and sorts the behavior data to obtain corresponding user behavior sequences; next, it performs time-series anomaly detection on the user behavior sequences based on a preset time-series anomaly detection model to obtain corresponding time-series anomaly detection results; and then performs path transformation analysis on the user behavior sequences to obtain corresponding path transformation analysis results; subsequently, it generates corresponding breakpoint detection data based on the time-series anomaly detection results and the path transformation analysis results; subsequently, it performs suggestion generation processing on the breakpoint detection data based on a preset target intelligent agent to obtain corresponding optimization suggestions; finally, it outputs the breakpoint detection data and the optimization suggestions. Based on the above automated processing flow, this application obtains user behavior sequences by cleaning and sorting behavior data, then performs time-series anomaly detection on the user behavior sequences based on a time-series anomaly detection model to obtain time-series anomaly detection results, performs path transformation analysis on the user behavior sequences to obtain path transformation analysis results, then automatically generates breakpoint detection data based on the time-series anomaly detection results and the path transformation analysis results, and subsequently performs suggestion generation processing on the breakpoint detection data based on the target intelligent agent to automatically obtain optimization suggestions. Thus, this application, through a time-series anomaly detection model, path transformation analysis, and collaborative processing of the target intelligent agent, can automatically and accurately achieve full-process behavior monitoring, breakpoint identification, and optimization suggestion generation, effectively improving the processing efficiency of breakpoint identification and the efficiency of optimization suggestion generation, while ensuring the accuracy of the generated breakpoint detection data and optimization suggestions.

[0040] In some alternative implementations, step S203 includes the following steps: Call the pre-trained time series anomaly detection model.

[0041] In this embodiment, the model construction process of the aforementioned temporal anomaly detection model includes: selecting a suitable deep learning framework (such as TensorFlow or PyTorch) and building an LSTM-AE model (Long Short-Term Memory Autoencoder). The LSTM layer is used to capture long-term dependencies in user behavior sequences. The encoder compresses the input sequence into a low-dimensional latent representation, and the decoder reconstructs the original sequence from the latent representation. Specifically, the LSTM-AE model can be trained using a large amount of normal user behavior sequence data, and the model parameters can be adjusted so that the model can learn the feature representation of normal behavior patterns, thereby obtaining a well-trained temporal anomaly detection model. The LSTM-AE model combines the temporal modeling capability of LSTM with the feature learning and reconstruction capability of the autoencoder, effectively modeling user behavior sequences and providing a foundation for subsequent reconstruction error calculation and breakpoint judgment.

[0042] The user behavior sequence is reconstructed based on the aforementioned time-series anomaly detection model to obtain the corresponding reconstructed sequence.

[0043] In this embodiment, the user behavior sequence can be input into a trained temporal anomaly detection model, and the decoder in the temporal anomaly detection model can reconstruct the user behavior sequence to obtain the corresponding reconstructed sequence.

[0044] The reconstruction error between the user behavior sequence and the reconstructed sequence is calculated based on a preset distance metric strategy.

[0045] In this embodiment, the selection of the aforementioned distance metric strategy is not specifically limited and can be determined according to actual business needs. For example, Euclidean distance or Manhattan distance can be used. Furthermore, the difference between the user behavior sequence and the reconstructed sequence can be calculated based on the selected distance metric strategy to obtain the reconstruction error. Specifically, for example, for user behavior sequences and its reconstructed sequence ,calculate The reconstruction error reflects the degree of difference between the actual behavior sequence and the normal behavior pattern learned by the model. A larger reconstruction error means that the behavior sequence deviates significantly from the normal pattern, which may indicate anomalies and provides a quantitative basis for breakpoint determination.

[0046] The reconstruction error is numerically compared with a preset error threshold to obtain the corresponding first numerical comparison result.

[0047] In this embodiment, by analyzing the reconstruction error distribution of normal user behavior sequences in historical data, and combining business experience with the sensitivity requirements for breakpoint detection, a reasonable reconstruction error threshold (i.e., error threshold) is set. For example, by statistically analyzing the mean and standard deviation of reconstruction errors for normal behavior sequences, the threshold is set as the mean plus a certain multiple of the standard deviation. When the reconstruction error of a user behavior sequence exceeds this threshold, the corresponding segment of the sequence is determined to be a breakpoint. The first numerical comparison result mentioned above includes: the reconstruction error is greater than the error threshold, or the reconstruction error is not greater than the error threshold. Furthermore, setting the error threshold is a crucial step in breakpoint judgment; a reasonable threshold can balance the false positive rate and the false negative rate. By combining historical data and business experience, a threshold that can effectively identify abnormal behavior while avoiding excessive false alarms can be found, improving the accuracy of breakpoint detection. Based on the first numerical comparison result, a temporal anomaly detection result corresponding to the user behavior sequence is generated.

[0048] In this embodiment, if the first numerical comparison result indicates that the reconstruction error is greater than the error threshold, a first temporal anomaly detection result is generated, with the segment corresponding to the user behavior sequence being a breakpoint. Conversely, if the first numerical comparison result indicates that the reconstruction error is not greater than the error threshold, a second temporal anomaly detection result is generated, with the segment corresponding to the user behavior sequence not being a breakpoint.

[0049] This application utilizes a pre-trained temporal anomaly detection model to reconstruct user behavior sequences, obtaining corresponding reconstructed sequences. Then, it calculates the reconstruction error between the user behavior sequences and the reconstructed sequences based on a preset distance metric strategy. The reconstruction error is then numerically compared with a preset error threshold to obtain a first numerical comparison result. Subsequently, a temporal anomaly detection result corresponding to the user behavior sequence is generated based on this first numerical comparison result. Based on this processing flow, this application, through the combined use of a temporal anomaly detection model, a distance metric strategy, and an error threshold, can efficiently and accurately complete the temporal anomaly detection processing of user behavior sequences, ensuring the accuracy of the temporal anomaly detection results and improving the accuracy of breakpoint detection.

[0050] In some optional implementations of this embodiment, step S204 includes the following steps: The user behavior sequence is processed to construct a transformation graph, resulting in the corresponding target transformation graph.

[0051] In this embodiment, the conversion graph construction process includes: further analyzing the user behavior sequence to identify different behavior nodes, such as "viewing the policy," "submitting a claim," and "registering an account." Then, based on the user's actual behavior path in the business journey, the conversion relationships between the behavior nodes are constructed to form the target conversion graph. ,in, For node combination, The set of edges is represented by the conversion graph, which visually illustrates the customer's behavioral path and conversion relationships throughout the business journey. This provides a clear structural framework for subsequent conversion rate calculations and breakpoint determination, helping business personnel understand the flow patterns of user behavior.

[0052] The path conversion rate is calculated on the target conversion graph to obtain the corresponding path conversion rate.

[0053] In this embodiment, the specific implementation process of calculating the path conversion rate of the target conversion graph to obtain the corresponding path conversion rate will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0054] The path conversion rate is compared with a preset conversion rate threshold to obtain a corresponding second numerical comparison result.

[0055] In this embodiment, a conversion rate threshold is pre-set based on the business's tolerance for customer churn and the stringency of breakpoint detection. When the conversion rate of a certain conversion path exceeds this threshold, a breakpoint alarm is triggered, indicating that the path has a high risk of customer churn and is a low-conversion stage. The second numerical comparison result mentioned above includes either a path conversion rate greater than the conversion rate threshold or a path conversion rate less than the conversion rate threshold. Furthermore, by setting the conversion rate threshold, breakpoint determination filters out paths with low conversion rates, clearly identifying the stages in the business that require focused attention and optimization, providing a clear target direction for generating subsequent optimization suggestions.

[0056] Based on the second numerical comparison result, a path transformation analysis result corresponding to the user behavior sequence is generated.

[0057] In this embodiment, if the second numerical comparison result indicates that the path conversion rate is greater than the conversion rate threshold, then a first path conversion analysis result is generated, where the path corresponding to the path conversion rate is a potential breakpoint. Conversely, if the second numerical comparison result indicates that the path conversion rate is not greater than the conversion rate threshold, then a first path conversion analysis result is generated, where the path corresponding to the path conversion rate is not a potential breakpoint.

[0058] This application constructs a conversion graph from a user behavior sequence to obtain a corresponding target conversion graph. Then, it calculates the path conversion rate from the target conversion graph to obtain a corresponding path conversion rate. Next, it compares the path conversion rate with a preset conversion rate threshold to obtain a second numerical comparison result. Finally, it generates a path conversion analysis result corresponding to the user behavior sequence based on this second numerical comparison result. Based on this processing flow, this application constructs a target conversion graph from a user behavior sequence, then compares the path conversion rate obtained from the target conversion graph with a preset conversion rate threshold, and generates a path conversion analysis result corresponding to the user behavior sequence based on the obtained second numerical comparison result. This allows for automatic and accurate path conversion analysis of user behavior sequences, improving processing efficiency, ensuring the accuracy of the obtained path conversion analysis results, and providing a clear target direction for subsequent optimization suggestions.

[0059] In some optional implementations, the calculation of the path conversion rate of the target conversion graph to obtain the corresponding path conversion rate includes the following steps: For each transformation path in the target transformation graph, count the number of positive transformation edges corresponding to the transformation path and the total number of edges.

[0060] In this embodiment, each of the above transformation paths refers to the transformation path from each behavior node to the next behavior node in the target transformation graph. The number of positive transformation edges is counted for each transformation path. (i.e., the number of times the transformation was successfully completed) and the total number of edges (That is, the total number of conversion attempts).

[0061] Obtain the preset conversion rate calculation formula.

[0062] In this embodiment, the conversion rate calculation formula is as follows: .in, For conversion rate, This represents the number of positive transformation edges. This represents the total number of edges.

[0063] The number of positive conversion edges and the total number of edges are calculated based on the conversion rate calculation formula to obtain the corresponding calculation results.

[0064] In this embodiment, the number of positive transformation edges and the total number of edges can be substituted into the above-mentioned transformation rate calculation formula for calculation, and the calculation result can be used as the path transformation rate of the transformation path.

[0065] The calculation result is used as the path conversion rate of the conversion path.

[0066] This application calculates the number of positive conversion edges and the total number of edges corresponding to each conversion path in the target conversion graph. Then, it obtains a preset conversion rate calculation formula. Next, it calculates the number of positive conversion edges and the total number of edges based on the conversion rate calculation formula, obtaining the corresponding calculation result. This result is then used as the path conversion rate of the conversion path. Based on this processing flow, this application efficiently and accurately calculates the path conversion rate of the target conversion graph for each conversion path in the target conversion graph, ensuring the accuracy of the obtained path conversion rate. This allows for the rapid and accurate identification of paths with low conversion rates, i.e., potential breakpoints, through subsequent analysis of the path conversion rate.

[0067] In some alternative implementations, step S205 includes the following steps: Based on a preset report generation strategy, the timing anomaly detection results and the path transformation analysis results are processed to generate reports, resulting in corresponding breakpoint reports.

[0068] In this embodiment, the report generation process includes integrating the time-series anomaly detection results with the path conversion analysis results to generate a detailed breakpoint report in a unified format. The breakpoint report includes information such as the breakpoint location (specific behavioral stage or conversion path), breakpoint type (time-series anomaly or conversion rate anomaly), the time range of the breakpoint occurrence, and the characteristics of the affected customer group (e.g., age, region, consumption habits). Furthermore, some key information can be displayed in chart form using data visualization tools (such as Tableau or Power BI) to enhance the report's readability. In addition, the breakpoint report is a crucial output of the breakpoint detection phase, providing business personnel with comprehensive and detailed breakpoint information, helping them quickly understand existing problems in their business and providing a basis for subsequent optimization decisions.

[0069] Obtain the breakpoint location information from the breakpoint report.

[0070] In this embodiment, the required breakpoint location information can be extracted and filtered from the breakpoint report.

[0071] Based on the breakpoint location information, the corresponding breakpoint context data is extracted from the behavior data.

[0072] In this embodiment, relevant contextual data at the time of the breakpoint can be extracted from the original behavioral data based on the breakpoint location information in the breakpoint report. This includes basic user information (such as user ID, age, gender, etc.), behavioral sequence details (such as specific operation steps, interaction time, etc.), and system environment information (such as device type, operating system version, etc.). This data is organized into a structured format for use in subsequent optimization strategy generation. The breakpoint contextual data sample provides rich details for generating optimization suggestions, facilitating a deeper understanding of the background and reasons for the breakpoint occurrence, thereby generating more targeted and effective optimization suggestions.

[0073] A corresponding risk heatmap is constructed based on the breakpoint location information.

[0074] In this embodiment, by selecting appropriate visualization tools, and using the various behavioral stages or conversion paths contained in the breakpoint location information as coordinate axes, different colors or intensities are used to represent the risk level of the breakpoints. For example, red represents high-risk breakpoints, yellow represents medium-risk breakpoints, and green represents low-risk breakpoints. Through the risk heatmap, business personnel can intuitively see the distribution of breakpoints in the customer journey and quickly locate key problem areas. The risk heatmap presents breakpoint information in an intuitive and visual way, quickly attracting the attention of business personnel and helping them quickly identify problem stages that need to be prioritized in complex business processes, thus improving problem-solving efficiency.

[0075] The breakpoint report, the breakpoint context data, and the risk heatmap are integrated and processed to obtain the corresponding breakpoint detection data.

[0076] In this embodiment, the generated breakpoint report, breakpoint context data, and risk heatmap can be integrated and processed, and the resulting integrated data can be used as the corresponding breakpoint detection data.

[0077] This application generates breakpoint reports by processing the results of temporal anomaly detection and path transformation analysis based on a preset report generation strategy. Then, it obtains the breakpoint location information from the breakpoint reports. Next, it extracts the corresponding breakpoint context data from the behavioral data based on the breakpoint location information and constructs a corresponding risk heatmap based on the breakpoint location information. Finally, it integrates the breakpoint reports, breakpoint context data, and risk heatmaps to obtain the corresponding breakpoint detection data. Based on this processing flow, this application generates breakpoint reports from the results of temporal anomaly detection and path transformation analysis using a report generation strategy. It intelligently extracts breakpoint context data from the behavioral data and constructs a risk heatmap based on the breakpoint location information in the breakpoint reports. Furthermore, it integrates the breakpoint reports, breakpoint context data, and risk heatmaps to obtain multi-dimensional breakpoint detection data, improving the richness and accuracy of the generated breakpoint detection data.

[0078] In some optional implementations of this embodiment, step S206 includes the following steps: The target agent invokes a pre-defined knowledge base, a large language model, and a policy attribution network.

[0079] In this embodiment, the aforementioned target intelligent agent can also be referred to as the optimization strategy generation agent. The construction process of the knowledge base includes: organizing business experts, data analysts, and other relevant personnel to collect historical successful optimization cases, industry best practices, business rules, and other knowledge. This knowledge is then categorized and stored according to multiple aspects such as page optimization, wording adjustment, and recommendation strategies to establish a structured knowledge base. A database management system (such as MySQL or MongoDB) can be used to store and manage the knowledge base, ensuring efficient knowledge retrieval and updates. The knowledge base is a crucial source of knowledge for generating optimization suggestions; a rich knowledge reserve can provide diverse reference solutions for different types of breakpoints, improving the comprehensiveness and feasibility of optimization suggestions.

[0080] Based on the breakpoint report, the knowledge base is searched to obtain corresponding similar historical cases.

[0081] In this embodiment, the retrieval process includes: determining retrieval conditions based on information in the breakpoint report, such as breakpoint type, user group characteristics, and business scenario; and retrieving historical cases similar to the current breakpoint from the knowledge base using methods such as keyword matching and semantic similarity calculation to obtain corresponding similar historical cases. For example, if the current breakpoint is a timing anomaly in the form filling process, and the affected customer group is young users, then the knowledge base will be searched for similar form filling optimization cases targeting young users.

[0082] Among them, the similar case retrieval can quickly find historical experiences similar to the current breakpoint, providing a direct reference for the generation of optimization suggestions, avoiding the need to generate suggestions from scratch, and improving the efficiency and quality of optimization suggestion generation.

[0083] Based on the large language model, the breakpoint context data and similar historical cases are used for reasoning to obtain the corresponding first optimization suggestion.

[0084] In this embodiment, the selection of the aforementioned large language model is not specifically limited and can be determined according to actual business needs. For example, models such as GPT4 can be used. The reasoning processing based on the large language model includes: organizing the breakpoint context data and retrieved similar historical cases, and encapsulating them according to the input format required by the large language model. Simultaneously, the problem to be solved is clearly defined, such as generating page optimization suggestions and script optimization suggestions, and the problem description is clearly added to the input. For example, the input may include "The current breakpoint is a timing anomaly in the form filling stage, affecting young users. Below are breakpoint context data samples and similar historical cases. Please generate targeted page optimization suggestions." Then, the prepared input is provided to the large language model, which analyzes and reasons based on the input information, combining knowledge from the knowledge base to generate customized first optimization suggestions. For example, for a breakpoint in the form filling stage, the large language model may generate suggestions such as "For young users, improve the form field layout, adopting a simpler and more stylish design; distinguish and label required and non-required fields using different colors or icons; provide clear prompts and examples when errors are made."

[0085] Among them, the large language model possesses powerful language understanding and generation capabilities, enabling it to generate logical and targeted first optimization suggestions based on input information. Customized suggestion generation can meet individual needs under different breakpoint conditions, improving the practicality and effectiveness of optimization suggestions.

[0086] The breakpoint detection data is processed by the attribution network described above to obtain the corresponding causal inference results.

[0087] In this embodiment, the above-mentioned causal inference processing includes: using a policy attribution network to calculate... .in, For the transformation result, This involves analyzing the causal impact of behavioral characteristics on conversion results and generating corresponding causal inferences. For example, regarding page load speed (a behavioral characteristic) and conversion rate (a conversion result), the difference between page load speed and conversion rate is calculated using experimental design or data analysis methods. This involves estimating (i.e., the conversion rate when page load speed is actively changed) and (i.e., the observed relationship between page load speed and conversion rate), determining whether a significant causal relationship exists between them. Causal inference can distinguish between "system problems" and "customer preference differences," clarifying the root cause of breakpoints. Only by finding the root cause of the problem can effective optimization strategies be developed, avoiding resource waste caused by blind optimization.

[0088] A second optimization suggestion is generated based on the causal inference results.

[0089] In this embodiment, based on the causal inference results, different optimization strategies are adopted for different types of breakpoint causes, and corresponding second optimization suggestions are generated. Specifically, for system problems, such as slow page loading speed, the technical team is promptly notified for repair and optimization, such as optimizing code and upgrading server configuration; for differences in customer preferences, such as different customer groups having different levels of acceptance of product recommendation methods, the optimization suggestions are adjusted to better meet the needs of different customer groups, for example, recommending more innovative and fashionable products to younger customers, and more practical and stable products to older customers. Targeted optimization ensures the effectiveness and accuracy of the optimization strategy, taking corresponding measures according to different breakpoint causes, improving the efficiency and effectiveness of problem solving, and maximizing customer conversion rates and experience.

[0090] The first optimization suggestion and the second optimization suggestion are integrated to obtain the corresponding optimization suggestion.

[0091] In this embodiment, the specific implementation process of integrating the first optimization suggestion and the second optimization suggestion to obtain the corresponding optimization suggestion will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0092] Innovatively, the optimization strategy generation agent not only generates optimization solutions but also automatically estimates the ROI potential and risk indicators for each solution. For example, for a page optimization suggestion, by analyzing the impact of similar optimization measures on conversion rates in historical data, the costs required to implement the optimization measures (such as development costs, time costs, etc.), and combining industry benchmarks and model prediction results, it estimates its "expected conversion rate increase of +2.1%, risk <3%". The ROI potential and risk indicators are provided to business decision-makers along with the optimization content, providing a quantitative reference for decision-making. The ROI potential and risk indicator estimation provides a scientific basis for business decisions, helping decision-makers weigh and select among multiple optimization solutions, prioritizing those with higher ROI potential and lower risk, thereby improving the efficiency of business resource utilization.

[0093] This application utilizes a pre-defined knowledge base, a large language model, and a policy attribution network based on the target intelligent agent. Then, it retrieves similar historical cases from the knowledge base based on breakpoint reports. Next, it infers from the breakpoint context data and similar historical cases using the large language model to obtain a first optimization suggestion. Following this, it performs causal inference on the breakpoint detection data using the policy attribution network to obtain a causal inference result. Finally, it generates a second optimization suggestion based on the causal inference result. The first and second optimization suggestions are then integrated to obtain the final optimization suggestion. Based on this process, this application, through the combined use of a knowledge base, a large language model, and a policy attribution network, can intelligently and accurately generate logically sound, targeted, and content-rich optimization suggestions from breakpoint detection data. This improves the intelligence of optimization suggestion generation, better meets personalized needs under different breakpoint conditions, and ultimately enhances the practicality and effectiveness of the optimization suggestions.

[0094] In some optional implementations of this embodiment, the process of integrating the first optimization suggestion and the second optimization suggestion to obtain the corresponding optimization suggestion includes the following steps: The first optimization suggestion and the second optimization suggestion are integrated to obtain the integrated target optimization suggestion.

[0095] In this embodiment, the first optimization suggestion and the second optimization suggestion can be integrated to obtain the target optimization suggestion after content integration.

[0096] Call the preset knowledge graph.

[0097] In this embodiment, the aforementioned knowledge graph can specifically be a pre-constructed insurance business knowledge graph (Insurance-KG) related to insurance business. Specifically, the construction process of the insurance business knowledge graph includes: Step 1: 1) Node and relationship definition. Entity node classification: Product terms: Extract key terms from insurance contracts (such as "waiting period for critical illness insurance" and "deductible"), and label the term type (such as "restrictive terms" and "scope of coverage"). Customer type: Define tags based on customer profiles (such as "high-risk users" and "patients with a history of diabetes"), and associate them with characteristics (such as age and health status). Regulatory rules: Extract rules from regulatory documents (such as the "Insurance Law" and the "Advertising Law") (such as "prohibition of promising guaranteed returns" and "prohibition of vague claims conditions"), and label the rule category (such as "marketing compliance" and "product design compliance"). Marketing scripts: Collect historical marketing texts (such as telephone scripts and APP push copy), and label the script type (such as "promotional" and "risk warning"). 2) Relationship definition: Restrictive relationship: such as "Term A → Restriction → Script B" (indicating that a certain term prohibits the use of a specific script). Applicability: e.g., "Clause C → Applies to → Customer Type D" (indicating that a clause only applies to a specific customer). Conflict: e.g., "Rule E → Conflict → Script F" (indicating that a script violates regulatory requirements). Dependency: e.g., "Script G → Dependency → Clause H" (indicating that a script requires citing specific clause content).

[0098] Step 2: Graph Construction. 1) Data Collection: Structured Data: Extract clause text and rule entries from insurance product databases and regulatory document libraries. Semi-structured Data: Parse product manuals and compliance manuals in PDF / Word format, extracting titles, paragraphs, and keywords. Unstructured Data: Mine the relationship between sales pitches and clauses through historical dialogue records (such as customer service chat logs and marketing emails). 2) Automated Processing: Entity Recognition: Use NLP tools (such as named entity recognition models) to label entities in the text (e.g., "waiting period," "high-risk user"). Relationship Extraction: Identify relationships between entities through pattern matching or supervised learning models (e.g., "Clause X prohibits the use of sales pitch Y"). Knowledge Fusion: Merge the same entity from different sources (e.g., normalize "critical illness insurance waiting period" and "major illness insurance waiting period"). Manual Verification: Sampling check the accuracy of relationships at key nodes (e.g., core clauses, high-frequency sales pitches). Correct logical errors through expert review (e.g., "Clause A restricts sales pitch B" but its use is actually allowed).

[0099] Step 3: Graph Storage and Query. 1) Storage Method: Use a graph database (such as Neo4j or JanusGraph) to store nodes and relationships, supporting efficient querying. Add attributes to each node (such as clause effective time and language style). 2) Query Interface: Design a Cypher query template (such as "MATCH (r:Rule)-[:Restriction]->(t:Talk) RETURN r, t") for subsequent compliance verification.

[0100] Among them, the knowledge graph integrates scattered regulatory rules, product terms, and marketing scripts into a semantic network, supporting rapid verification of policy compliance through logical reasoning. For example, when a new script is input, it can automatically traverse all related terms and rules to determine whether there are any conflicts.

[0101] Based on the knowledge graph, the target optimization suggestion is validated for compliance to determine whether it meets the corresponding constraints.

[0102] In this embodiment, the compliance verification includes: Step 1: Suggestion Pre-Check. 1) Suggestion Analysis: After generating optimization suggestions, the system extracts key elements: Recommended content: e.g., "Recommend critical illness insurance product A to users." Applicable conditions: e.g., "Target customers are healthy individuals aged 30-45." Accompanying wording: e.g., "Purchasing product A entitles you to double the coverage amount." The suggestion elements are mapped to entities in a knowledge graph (e.g., "Product A," "30-45 year olds," "Double coverage wording"). 2) Contextual Supplementation: Implicit information is supplemented based on the suggestion's context (e.g., "Recommending critical illness insurance" implies a need to check the user's health declaration).

[0103] Step 2: Graph Query and Verification. 1) Query Rules: For each element in the optimization suggestion, retrieve relevant constraints in the graph: Script Compliance: Check if the script is prohibited by any regulatory rules (e.g., does "doubling the sum assured" trigger the "prohibition of exaggerating returns" rule). Terms Applicability: Check if the product terms match the target customer type (e.g., does "diabetic users" apply to "standard critical illness insurance terms"). Conflict Detection: Verify whether multiple element combinations are contradictory (e.g., does "recommended product A" conflict with "users are high-risk individuals"). 2) Logical Verification: Perform path reasoning for complex suggestions (e.g., multi-term combination recommendations): Example path: (User type) → [Applicable] → (Terms) → [Restrictions] → (Script). If there are "prohibited" or "conflicting" relationships in the path, mark the suggestion as non-compliant.

[0104] Step 3: Violation Handling. 1) Suggestion Interception: If the verification result is Verify(a_t,KG)=True, the optimization suggestion is deemed to have passed compliance verification, meaning it meets the constraints. 2) Suggestion Interception: If the verification result is Verify(a_t,KG)=False, the system refuses to execute the suggestion and returns the specific reason for the violation (e.g., "The wording violates Article X of the Advertising Law"). 2) Audit Log: Records the complete context of the violating suggestion (including Agent ID, timestamp, and associated nodes). Categorizes and labels the violation type (e.g., "Misleading Marketing," "Terms Abuse") for subsequent analysis.

[0105] If the target optimization suggestion satisfies the constraints, then the target optimization suggestion shall be adopted as the optimization suggestion.

[0106] In this embodiment, if the target optimization suggestion is found to meet the constraints, it will be used as the final optimization suggestion. If the target optimization suggestion is found to not meet the constraints, the system will refuse to execute the suggestion, return the specific reason for the violation, and may further trigger a manual review process for high-risk violations (such as those involving legal disputes) for secondary confirmation by compliance experts.

[0107] This application integrates the first and second optimization suggestions to obtain an integrated target optimization suggestion. Then, it invokes a pre-defined knowledge graph. Subsequently, it performs compliance verification on the target optimization suggestion based on the knowledge graph, determining whether the target optimization suggestion meets the corresponding constraints. If the target optimization suggestion meets the constraints, it is adopted as the optimization suggestion. Based on the above processing flow, this application, by utilizing the semantic association capabilities of the knowledge graph, can achieve automated and interpretable review of the target optimization suggestion, effectively improving the efficiency of compliance verification and ensuring the compliance of the generated optimization suggestion.

[0108] In some optional implementations, the system also includes an effect evaluation and reinforcement learning module. This module's function is to automatically collect new behavioral data after the optimization suggestions are implemented, evaluate the effects in real time, and update the Agent's decision-making strategy. The specific implementation process includes: 1. Experimental Grouping and A / B Testing. Implementation Process: Traffic Splitting: The system evenly distributes user traffic to the control group (Ctrl) and the experimental group (Exp) based on random seeds such as user ID, device characteristics, or timestamps, ensuring consistent user characteristic distribution between the two groups. Key Metric Definition: Define core evaluation metrics (e.g., click-through rate (CTR), conversion rate (CVR)). The control group uses the original strategy, while the experimental group uses the new optimized strategy. Performance Calculation: Calculate the improvement ratio of the experimental group relative to the control group using the formula Δ=(CTR_exp-CTR_ctrl) / CTR_ctrl to determine whether the new strategy is significantly better than the baseline. Statistical Significance Test: Use a t-test or chi-square test to verify whether the results are statistically significant (e.g., p-value < 0.05). A / B testing quantifies the actual effect of the optimization strategy by comparing the performance of the control and experimental groups. The randomness of traffic splitting ensures that the results are not affected by user bias, and the Δ value directly reflects the magnitude of strategy improvement, providing initial feedback for subsequent reinforcement learning.

[0109] 2. Reinforcement learning update mechanism. Implementation process: Reward function design: Define a multi-objective reward function R_t=α conversion_gain β `churn_rate`, where: `conversion_gain`: the benefit of improved conversion rate (e.g., increased order volume); `churn_rate`: user churn rate or strategy risk cost (e.g., complaint rate); α, β: weighting coefficients, balancing benefit and risk. Strategy Update: Each Agent adjusts its strategy based on the current reward value: if `R_t` > threshold, the current strategy is added to the strategy library and its execution probability is increased; if `R_t` ≤ threshold, the strategy's weight is reduced or it is directly eliminated. Exploration-Exploitation Balance: Through ε-greedy or Thompson Sampling mechanisms, Agents are allowed to try low-return strategies with a certain probability to explore potential optimization space. Reinforcement learning transforms business objectives (e.g., conversion, risk control) into mathematical optimization problems through a reward function. The reward value drives Agent self-iteration, forming a closed loop of "trial and error-feedback-optimization," ensuring maximum long-term strategy benefit.

[0110] In addition, the system also includes a flywheel self-evolution mechanism and monitoring interface, specifically including: 1. Flywheel Self-Evolution Mechanism. Implementation Process: Loop Flow: New Measure Deployment: The Agent generates an optimized strategy and pushes it to the production environment; Data Collection: Real-time collection of user behavior data (e.g., clicks, conversions, interaction duration); Model Relearning: Updating the breakpoint detection model (identifying strategy failure points) and the strategy recommendation model (generating new strategies); Strategy Regeneration: Generating the next round of optimized strategies based on the learning results, forming a closed loop. Dynamic Parameter Adjustment: Automatically adjusting hyperparameters such as model learning rate and exploration intensity based on system performance (e.g., reward value fluctuations). The flywheel mechanism achieves self-evolution of "data → model → strategy" through continuous iteration, similar to online learning in machine learning. Each loop optimizes the strategy based on the latest data, ensuring the system adapts to market changes and user preference shifts.

[0111] 2. Visualized Monitoring. Implementation Process: Core Indicator Dashboard: Conversion Rate Increase: Real-time display of conversion rate comparison curves between the experimental and control groups; Customer Satisfaction Increase: Generation of trend charts based on NPS scores or survey data; Average Interaction Time Decrease: Monitoring user interaction efficiency with the system and optimizing the conciseness of communication messages. Decision Explanation Report: Generating a natural language report for each Agent decision, explaining the basis for strategy selection (e.g., "Choosing message X because of high historical conversion rate and compliance"); marking key decision points (e.g., "Rejecting recommendations to high-risk users due to risk threshold triggering"). Visualized monitoring makes the black-box optimization process transparent, helping operations personnel quickly locate problems (e.g., a strategy causing a decrease in satisfaction), while meeting regulatory requirements for algorithm explainability. The explanation report serves as a "decision audit log," supporting post-event review and compliance audit.

[0112] In some alternative implementations, the user information obtained is subject to user consent and complies with relevant laws and policies.

[0113] Furthermore, any software tools or components not belonging to our company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0114] Furthermore, this application utilizes an agent-based intelligent system to automatically collect customer data, intelligently identify breakpoints, and perform self-evolutionary optimization in financial and insurance scenarios, constructing a dynamic credit granting and service flywheel system centered on customer experience and fueled by data. This solution balances interpretability, automation, and compliance, possessing high practicality and innovative value. Specifically, the technical innovations of this application can be summarized as follows: 1. Agent-level data collection and optimization collaboration mechanism: Agents from different channels learn autonomously and share their status to achieve unified perception of cross-domain data.

[0115] 2. Intelligent detection algorithm for behavioral breakpoints: Combining time series reconstruction and conversion graph analysis to improve the accuracy of customer churn point identification.

[0116] 3. Causal-driven optimization generative model: Replace correlation analysis with causal inference and output explainable optimization suggestions.

[0117] 4. Reinforcement learning self-feedback flywheel: The optimization strategy is automatically evolved through a reward mechanism, avoiding manual parameter tuning.

[0118] 5. Auditable knowledge graph constraint layer: Ensures that all agent optimization results comply with financial compliance requirements and prevents illegal marketing or misleading promises.

[0119] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0120] It should be emphasized that, to further ensure the privacy and security of the above optimization suggestions, these suggestions can also be stored in a node of a blockchain.

[0121] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0122] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0123] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0124] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0125] Further reference Figure 3 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a data processing device based on artificial intelligence, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0126] like Figure 3 As shown, the artificial intelligence-based data processing device 300 described in this embodiment includes: an acquisition module 301, a processing module 302, a detection module 303, an analysis module 304, a first generation module 305, a second generation module 306, and an output module 307. Wherein: The acquisition module 301 is used to acquire the behavioral data of the target user from a preset channel; Processing module 302 is used to perform data cleaning and sorting on the behavioral data to obtain the corresponding user behavior sequence; The detection module 303 is used to perform time-series anomaly detection on the user behavior sequence based on a preset time-series anomaly detection model, and obtain the corresponding time-series anomaly detection result; Analysis module 304 is used to perform path transformation analysis on the user behavior sequence and obtain the corresponding path transformation analysis results; The first generation module 305 is used to generate corresponding breakpoint detection data based on the timing anomaly detection results and the path transformation analysis results. The second generation module 306 is used to perform suggestion generation processing on the breakpoint detection data based on a preset target intelligent agent to obtain corresponding optimization suggestions. The output module 307 is used to output the breakpoint detection data and the optimization suggestions.

[0127] In this embodiment, the operations performed by the above modules or units correspond one-to-one with the steps of the artificial intelligence-based data processing method in the aforementioned embodiments, and will not be repeated here.

[0128] In some optional implementations of this embodiment, the detection module 303 includes: The first calling submodule is used to call the pre-trained time anomaly detection model; The reconstruction submodule is used to reconstruct the user behavior sequence based on the time-series anomaly detection model to obtain the corresponding reconstruction sequence. The first calculation submodule is used to calculate the reconstruction error between the user behavior sequence and the reconstruction sequence based on a preset distance metric strategy; The first comparison submodule is used to compare the reconstruction error with a preset error threshold to obtain the corresponding first numerical comparison result. The first generation submodule is used to generate a time-series anomaly detection result corresponding to the user behavior sequence based on the first numerical comparison result.

[0129] In some optional implementations of this embodiment, the analysis module 304 includes: The first construction submodule is used to perform transformation graph construction processing on the user behavior sequence to obtain the corresponding target transformation graph; The second calculation submodule is used to calculate the path conversion rate of the target conversion graph to obtain the corresponding path conversion rate. The second comparison submodule is used to compare the path conversion rate with a preset conversion rate threshold to obtain the corresponding second numerical comparison result. The second generation submodule is used to generate path transformation analysis results corresponding to the user behavior sequence based on the second numerical comparison result.

[0130] In some optional implementations of this embodiment, the second calculation submodule includes: The statistics unit is used to count the number of positive transformation edges corresponding to each transformation path in the target transformation graph and the total number of edges. The acquisition unit is used to acquire the preset conversion rate calculation formula; The calculation unit is used to calculate the number of positive conversion edges and the total number of edges based on the conversion rate calculation formula, and obtain the corresponding calculation results; The first determining unit is used to use the calculation result as the path conversion rate of the conversion path.

[0131] In some optional implementations of this embodiment, the first generation module 305 includes: The third generation submodule is used to perform report generation processing on the time-series anomaly detection results and the path transformation analysis results based on a preset report generation strategy, so as to obtain the corresponding breakpoint report. The acquisition submodule is used to acquire the breakpoint location information in the breakpoint report; The extraction submodule is used to extract the corresponding breakpoint context data from the behavior data based on the breakpoint location information; The second construction submodule is used to construct a corresponding risk heat map based on the breakpoint location information; The first integration submodule is used to integrate the breakpoint report, the breakpoint context data, and the risk heatmap to obtain the corresponding breakpoint detection data.

[0132] In some optional implementations of this embodiment, the second generation module 306 includes: The second invocation submodule is used to invoke a preset knowledge base, a large language model, and a policy attribution network based on the target intelligent agent. The retrieval submodule is used to retrieve and process the knowledge base based on the breakpoint report to obtain the corresponding similar historical cases; The reasoning submodule is used to perform reasoning processing on the breakpoint context data and the similar historical cases based on the large language model to obtain the corresponding first optimization suggestion; The inference submodule is used to perform causal inference processing on the breakpoint detection data based on the policy attribution network to obtain the corresponding causal inference result. The fourth generation submodule is used to generate a corresponding second optimization suggestion based on the causal inference result; The second integration submodule is used to integrate the first optimization suggestion and the second optimization suggestion to obtain the corresponding optimization suggestion.

[0133] In some optional implementations of this embodiment, the second integration submodule includes: An integration unit is used to integrate the first optimization suggestion and the second optimization suggestion to obtain the integrated target optimization suggestion; The calling unit is used to invoke a pre-defined knowledge graph; The judgment unit is used to perform compliance verification on the target optimization suggestion based on the knowledge graph, and to determine whether the target optimization suggestion meets the corresponding constraints. The second determining unit is used to take the target optimization suggestion as the optimization suggestion if the target optimization suggestion satisfies the constraint conditions.

[0134] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0135] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0136] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0137] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for data processing methods based on artificial intelligence. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0138] In some embodiments, the processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions of the artificial intelligence-based data processing method.

[0139] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0140] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the artificial intelligence-based data processing method described above.

[0141] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0142] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A data processing method based on artificial intelligence, characterized in that, Includes the following steps: Obtain behavioral data of target users from pre-defined channels; The behavioral data is cleaned and sorted to obtain the corresponding user behavior sequence; The user behavior sequence is subjected to time-series anomaly detection based on a preset time-series anomaly detection model to obtain the corresponding time-series anomaly detection results. Perform path transformation analysis on the user behavior sequence to obtain the corresponding path transformation analysis results; Based on the temporal anomaly detection results and the path transformation analysis results, corresponding breakpoint detection data is generated; Based on a preset target intelligent agent, the breakpoint detection data is processed to generate suggestions, resulting in corresponding optimization suggestions. The breakpoint detection data and the optimization suggestions are then processed and output.

2. The data processing method based on artificial intelligence according to claim 1, characterized in that, The step of performing time-series anomaly detection on the user behavior sequence based on a preset time-series anomaly detection model to obtain the corresponding time-series anomaly detection results specifically includes: Call the pre-trained time series anomaly detection model; The user behavior sequence is reconstructed based on the aforementioned time-series anomaly detection model to obtain the corresponding reconstructed sequence. The reconstruction error between the user behavior sequence and the reconstructed sequence is calculated based on a preset distance metric strategy. The reconstruction error is numerically compared with a preset error threshold to obtain the corresponding first numerical comparison result; Based on the first numerical comparison result, a temporal anomaly detection result corresponding to the user behavior sequence is generated.

3. The data processing method based on artificial intelligence according to claim 1, characterized in that, The step of performing path transformation analysis on the user behavior sequence to obtain the corresponding path transformation analysis results specifically includes: The user behavior sequence is processed to construct a transformation graph to obtain the corresponding target transformation graph; The target conversion map is processed by calculating the path conversion rate to obtain the corresponding path conversion rate; The path conversion rate is numerically compared with a preset conversion rate threshold to obtain a corresponding second numerical comparison result; Based on the second numerical comparison result, a path transformation analysis result corresponding to the user behavior sequence is generated.

4. The data processing method based on artificial intelligence according to claim 3, characterized in that, The step of calculating the path conversion rate of the target conversion map to obtain the corresponding path conversion rate specifically includes: For each transformation path in the target transformation graph, count the number of positive transformation edges corresponding to the transformation path and the total number of edges; Obtain the preset conversion rate calculation formula; The number of positive conversion edges and the total number of edges are calculated based on the conversion rate calculation formula to obtain the corresponding calculation results; The calculation result is used as the path conversion rate of the conversion path.

5. The data processing method based on artificial intelligence according to claim 1, characterized in that, The step of generating corresponding breakpoint detection data based on the temporal anomaly detection results and the path transformation analysis results specifically includes: Based on a preset report generation strategy, the time-series anomaly detection results and the path transformation analysis results are processed to generate a report, resulting in a corresponding breakpoint report. Obtain the breakpoint location information from the breakpoint report; Based on the breakpoint location information, the corresponding breakpoint context data is extracted from the behavior data; Construct a corresponding risk heatmap based on the breakpoint location information; The breakpoint report, the breakpoint context data, and the risk heatmap are integrated and processed to obtain the corresponding breakpoint detection data.

6. The data processing method based on artificial intelligence according to claim 5, characterized in that, The step of generating suggestions based on the breakpoint detection data using a preset target intelligent agent to obtain corresponding optimization suggestions specifically includes: Based on the target intelligent agent, a preset knowledge base, a large language model, and a policy attribution network are invoked; Based on the breakpoint report, the knowledge base is retrieved and processed to obtain the corresponding similar historical cases; Based on the large language model, the breakpoint context data and the similar historical cases are used for reasoning to obtain the corresponding first optimization suggestion; Based on the policy attribution network, causal inference processing is performed on the breakpoint detection data to obtain the corresponding causal inference results; A second optimization suggestion is generated based on the causal inference results; The first optimization suggestion and the second optimization suggestion are integrated to obtain the corresponding optimization suggestion.

7. The data processing method based on artificial intelligence according to claim 6, characterized in that, The step of integrating the first optimization suggestion and the second optimization suggestion to obtain the corresponding optimization suggestion specifically includes: The first optimization suggestion and the second optimization suggestion are integrated to obtain the integrated target optimization suggestion. Invoke the pre-defined knowledge graph; Based on the knowledge graph, the target optimization suggestion is verified for compliance to determine whether the target optimization suggestion meets the corresponding constraints. If the target optimization suggestion satisfies the constraints, then the target optimization suggestion shall be adopted as the optimization suggestion.

8. A data processing device based on artificial intelligence, characterized in that, include: The acquisition module is used to acquire behavioral data of target users from preset channels; The processing module is used to perform data cleaning and sorting on the behavioral data to obtain the corresponding user behavior sequence; The detection module is used to perform time-series anomaly detection on the user behavior sequence based on a preset time-series anomaly detection model, and obtain the corresponding time-series anomaly detection results; The analysis module is used to perform path transformation analysis on the user behavior sequence and obtain the corresponding path transformation analysis results; The first generation module is used to generate corresponding breakpoint detection data based on the timing anomaly detection results and the path transformation analysis results. The second generation module is used to perform suggestion generation processing on the breakpoint detection data based on a preset target intelligent agent to obtain corresponding optimization suggestions. The output module is used to process the breakpoint detection data and the optimization suggestions.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the data processing method based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the data processing method based on artificial intelligence as described in any one of claims 1 to 7.

Images