Insurance industry batch job monitoring system and method based on large language model

Abstract

The invention provides an insurance industry batch job monitoring system and method based on a large language model. The system comprises a data acquisition module, a log analysis module, an intelligent report generation module, an alarm and feedback module and a learning and optimization module. The method comprises the following steps: S1, data acquisition; s2, performing semantic analysis on the collected batch job logs; s3, generating an execution report; s4, alarming and feedback are carried out; s5, performing optimization and learning; according to the insurance industry batch job monitoring system and method based on the large language model, a large amount of data and complex business logic can be rapidly processed, and the batch job monitoring efficiency is greatly improved; compared with a traditional monitoring mode, the system can complete the inspection and analysis of a large amount of batch operation in a short time, and reduces the manpower cost and time cost. The method has high accuracy and reliability, and misjudgment and deviation caused by human factors are avoided.

Description

Technical Field

[0001] The present invention relates to the field of batch operation monitoring in the insurance industry, and in particular to a batch operation monitoring system and method for the insurance industry based on a large language model. Background Art

[0002] In the insurance industry, batch jobs refer to periodic, large-scale operations or tasks, including claims processing, policy verification, report generation, bill settlement, etc. These operations usually require processing a large amount of data and performing a series of automated operations. With the increase in business complexity and data volume, manual operations or rule-based automated systems have been unable to meet the requirements of real-time, accuracy and efficiency of batch job management. At present, batch job monitoring in the insurance industry mainly relies on preset rules, which lack flexibility and adaptability and cannot effectively deal with complex or unknown abnormal situations. Potential problems in batch jobs can only be discovered after the job is completed in the preset rule mode, resulting in delayed problem handling, affecting business continuity and customer experience. With the expansion of business scale, the deficiencies of traditional monitoring methods in real-time, automation and intelligence are becoming more and more obvious, and cannot meet the modern insurance industry's needs for efficient and accurate monitoring. Summary of the invention

[0003] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a batch operation monitoring system and method for the insurance industry based on a large language model.

[0004] To achieve the above object, the present invention adopts the following technical solutions: A batch operation monitoring system and method for the insurance industry based on a large language model, the batch operation monitoring system for the insurance industry based on a large language model comprises a data acquisition module, a log analysis module, an intelligent report generation module, an alarm and feedback module, and a learning and optimization module; The data acquisition module is used to obtain batch operation logs from various systems of the insurance company and transmit them to the central database; The log analysis module performs semantic analysis on the log data based on the large language model; The intelligent report generation module is used to generate a detailed execution report; The alarm and feedback module is used to monitor the various processes of batch job log execution in real time. When an abnormality is detected, the alarm information will be sent to the relevant operation and maintenance personnel through email, SMS or instant messaging tools; The learning and optimization module is used to continuously optimize the model based on the continuously accumulated historical data.

[0005] The insurance industry batch job monitoring method based on a large language model comprises the following steps: S1: Data collection; The following sub-steps are included: S11: Set data source; Set up log files, database tables, file storage, etc. related to batch jobs in various systems of the insurance company as data sources; The systems include core business systems, claims processing systems, actuarial systems, etc.; S12: Determine the batch job log to be collected; Determine the batch job logs to be collected based on actual insurance business needs and pre-set batch job monitoring targets; the batch job logs include basic batch job information, execution status, time-related information, business-related indicators and other data; The basic information of the batch job includes the batch job name, batch job number, the system to which the batch job belongs, etc.; The execution status includes running, failed, successful, and paused; The time-related information includes the start time, end time, estimated execution time, actual execution time, etc.; The business-related indicators include the number of insurance policies processed, the number of claims cases, the amount of premium calculation, etc.; S13: Setting data collection frequency; According to the actual batch job operation cycle and monitoring requirements, set different collection frequencies and corresponding data time ranges for high-frequency batch jobs and low-frequency batch jobs; S14: Collect batch job logs; Pre-set reasonable batch operation data rules and necessary batch operation data elements, including batch operation number, execution status, execution time, number of insurance policies processed, etc.; Collect batch job logs required in batch job tasks from data sources according to the set collection frequency through API interfaces, message queues, log collection tools, etc. The log collection tools include Logstash, Fluentd, etc.; After the collection is completed, the collected batch job logs are checked for integrity and rationality; Specifically, the data verification function is used to check whether each collected batch job log contains the necessary batch job data elements. If the necessary batch job data elements are missing, the integrity check fails, an abnormal reminder is given, and the failure reason is recorded; otherwise, the integrity check succeeds; Compare the collected batch job logs with the set batch job data reasonable rules to check whether the collected batch job logs meet the set batch job data reasonable rules. If not, the rationality check fails, an abnormal reminder is given and the failure reason is recorded; otherwise, the rationality check succeeds; The reasonable rules for batch data set include that the execution time cannot be a negative number, the number of insurance policies processed is an integer, and the timestamp data is within the time range set by the reasonable rules for batch data. Transfer batch job logs to a central data warehouse.

[0006] S2: Perform semantic analysis on the collected batch job logs; The following sub-steps are included: S21: Processing the batch job log; This includes deduplication, supplementation and modification, removal of irrelevant data, format conversion, etc. of batch job logs; The central database receives the batch job logs and performs deduplication processing on the batch job logs by using a unique constraint in the database; The engineer modifies and supplements the batch job log that fails the integrity check and rationality check according to the failure reason recorded in the batch job log in step S1, combined with the log context content or other related logs. If the modification and supplementation are not possible, the corresponding data of the batch job log is marked as suspicious data; Removing irrelevant data in the batch job log, the irrelevant data including the comment part in the batch job log, automatically generated meaningless character sequences, etc.; Unify the data format in batch job logs; S22: Use a large model to perform semantic analysis on batch job logs; According to the specific requirements and data characteristics of the batch job log, select the corresponding large model for analysis; The large model identifies information in batch job logs based on the training corpus and converts it into feature vectors. It also identifies abnormal situations based on the feature vectors and marks abnormal batch job logs, such as task timeout, failure, or abnormal exit.

[0007] S3: Generate execution report; The alarm and feedback module monitors each process of batch job log execution in real time and generates monitoring data; the monitoring data includes batch job log execution status, configuration information, system resource usage, etc.; The intelligent report generation module collects the monitoring data of batch job logs in real time or periodically, organizes and formats them, converts them into vector forms that can be processed by the report generation model, and inputs the prepared data into the trained report generation model; The report generation model generates the execution report text according to the predefined report template and logical structure, and checks whether the execution report text content is complete and whether it contains all the batch job log monitoring data, and promptly modifies the monitoring data that does not meet the requirements; Furthermore, the report generation model generates corresponding chart descriptions based on the input content, and the chart descriptions include a success rate change trend chart, a resource utilization rate bar chart, etc.; Create corresponding visual charts according to the generated chart description; S4: Providing warning and feedback; The alarm and feedback module issues an alarm based on the data in the visual chart, and the alarm includes three levels: low, medium, and high; Set low-level alarms for minor resource fluctuations and other issues, and set high-level alarms for failures in key batch job logs or abnormal business indicators that seriously affect the business. At the same time, an alarm upgrade mechanism is set. If a low-level alarm is not handled within a certain period of time and the problem persists, it will be upgraded to a medium-level alarm. After the alarm rule is triggered, a report is automatically generated and sent to relevant personnel via email, text message, instant messaging tools, etc.; when a high-level alarm occurs, an email is sent to the batch operation and maintenance team members, and a text message reminder is sent to the on-duty personnel at the same time to increase the timeliness and reliability of the notification.

[0008] S5: Optimize and learn; Continuously use the ever-increasing historical data to comprehensively evaluate the trained model, and adopt corresponding optimization strategies based on the model evaluation results and performance analysis.

[0009] Compared with the prior art, the present invention has the following beneficial effects: The insurance industry batch job monitoring system and method based on the large language model proposed in the present invention can quickly process a large amount of data and complex business logic, greatly improving the efficiency of batch job monitoring; compared with traditional monitoring methods, it can complete the inspection and analysis of a large number of batch jobs in a short time, reducing labor costs and time costs; at the same time, because the analysis results of the large language model are based on data and algorithms, they have higher accuracy and reliability, avoiding misjudgments and deviations caused by human factors.

[0010] The present invention can also conduct in-depth mining and analysis of insurance batch operation data; it can not only discover surface problems and anomalies, but also reveal the underlying laws and trends behind the data; this helps insurance companies better understand business operations, discover bottlenecks and optimization points in business processes, and provide strong support for business decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 The present invention is a flowchart of the steps of a batch operation monitoring system and method for the insurance industry based on a large language model. DETAILED DESCRIPTION

[0012] In order to provide a further understanding of the purpose, structure, features, and functions of the present invention, the following detailed description is given in conjunction with the embodiments.

[0013] like Figure 1 As shown, a batch operation monitoring system and method for the insurance industry based on a large language model, the batch operation monitoring system for the insurance industry based on a large language model includes a data acquisition module, a log analysis module, an intelligent report generation module, an alarm and feedback module, and a learning and optimization module; The data acquisition module is used to obtain batch operation logs from various systems of the insurance company and transmit them to the central database; The log analysis module performs semantic analysis on the log data based on the large language model; The intelligent report generation module is used to generate a detailed execution report; The alarm and feedback module is used to monitor the various processes of batch job log execution in real time. When an abnormality is detected, the alarm information will be sent to the relevant operation and maintenance personnel through email, SMS or instant messaging tools; The learning and optimization module is used to continuously optimize the model based on the continuously accumulated historical data.

[0014] The insurance industry batch job monitoring method based on a large language model comprises the following steps: S1: Data collection; The following sub-steps are included: S11: Set data source; Set up log files, database tables, file storage, etc. related to batch jobs in various systems of the insurance company as data sources; The systems include core business systems, claims processing systems, actuarial systems, etc.; The log file related to the batch job includes the name, number, start time, end time, number of insurance policies processed, execution status, etc. of the batch job; The database table contains various insurance business data, such as policy information, customer information, policy change records, compensation amount calculation, etc.; The file storage includes scanned copies of insurance policies, claim material documents, customer emails, contract agreements, etc.; S12: Determine the batch job log to be collected; Determine the batch job logs to be collected based on actual insurance business needs and pre-set batch job monitoring targets; the batch job logs include basic batch job information, execution status, time-related information, business-related indicators and other data; The basic information of the batch job includes the batch job name, batch job number, the system to which the batch job belongs, etc.; The execution status includes running, failed, successful, and paused; The time-related information includes the start time, end time, estimated execution time, actual execution time, etc.; The business-related indicators include the number of insurance policies processed, the number of claims cases, the amount of premium calculation, etc.; Furthermore, for batch operations related to risk monitoring, special indicators such as risk assessment values ​​and the number of abnormal transactions should be collected; S13: Setting data collection frequency; According to the actual batch job operation cycle and monitoring requirements, set different collection frequencies and corresponding data time ranges for high-frequency batch jobs and low-frequency batch jobs; The high-frequency batch jobs include premium calculation batch jobs that run multiple times a day, and the collection frequency is set to every hour or less; The low-frequency batch jobs include monthly actuarial model updates, etc., and the frequency is set to daily, etc.; At the same time, determine the time range for collection, whether to collect data for the past week or month or all historical data since the start of the batch job. This depends on storage resources and analysis requirements, and ensure that the collection process does not affect the operation of the batch job itself.

[0015] S14: Collect batch job logs; Pre-set reasonable batch operation data rules and necessary batch operation data elements, including batch operation number, execution status, execution time, number of insurance policies processed, etc.; Collect batch job logs required in batch job tasks from data sources according to the set collection frequency through API interfaces, message queues, log collection tools, etc. The log collection tools include Logstash, Fluentd, etc.; Preferably, it includes two forms: real-time collection and batch collection. Real-time collection is used to collect real-time batch operation data, and batch collection is used to collect historical batch operation data. The corresponding collection form is selected according to actual needs. Through these two collection methods, not only can the execution status, progress and results of each job be obtained in real time, but also the historical batch job data can be retrospectively queried and stored. When a batch job task is started, data collection monitors the execution status of the task in real time and promptly feeds back the execution progress and processing results of the task. After the collection is completed, the collected batch job logs are checked for integrity and rationality; Specifically, the data verification function is used to check whether each collected batch job log contains the necessary batch job data elements. If the necessary batch job data elements are missing, the integrity check fails, an abnormal reminder is given, and the failure reason is recorded; otherwise, the integrity check succeeds; Compare the collected batch job logs with the set batch job data reasonable rules to check whether the collected batch job logs meet the set batch job data reasonable rules. If not, the rationality check fails, an abnormal reminder is given and the failure reason is recorded; otherwise, the rationality check succeeds; The reasonable rules for batch data set include that the execution time cannot be a negative number, the number of insurance policies processed is an integer, and the timestamp data is within the time range set by the reasonable rules for batch data. Transfer batch job logs to a central data warehouse; Furthermore, the storage format, protocol and structure of batch data are different in each system where the data source is located. After the batch data is collected, format conversion and standardization are performed to make the batch data format unified. For example, converting data in JSON format into CSV or Parquet format, or converting structured database table data into a unified data model; data transmission between different systems may use different protocols, the API interface may use the RESTful protocol, and the message queue may use protocols such as AMQP, which require protocol adaptation; data fields in different systems may have inconsistent naming and structures, such as field names, data types, etc., which require field mapping and data structure conversion to ensure that all data conforms to a unified standard format.

[0016] Preferably, sensitive information such as identity information, policy information, and claim records in the batch job data is encrypted and transmitted through an encryption protocol; Encrypted transmission ensures that data is not stolen or tampered with during transmission, thus ensuring data security and privacy.

[0017] S2: Perform semantic analysis on the collected batch job logs; The following sub-steps are included: S21: Processing the batch job log; This includes deduplication, supplementation and modification, removal of irrelevant data, format conversion, etc. of batch job logs; The central database receives the batch job logs and performs deduplication processing on the batch job logs by using a unique constraint in the database; The engineer modifies and supplements the batch job log that fails the integrity check and rationality check according to the failure reason recorded in the batch job log in step S1, combined with the log context content or other related logs. If the modification and supplementation are not possible, the corresponding data of the batch job log is marked as suspicious data; Removing irrelevant data in the batch job log, the irrelevant data including the comment part in the batch job log, automatically generated meaningless character sequences, etc.; Unify the data format in the batch job log; including unifying the timestamp data into the ISO8601 date and time format, formatting all IP addresses into a unified form, and encoding string type messages into the UTF-8 format; S22: Use a large model to perform semantic analysis on batch job logs; According to the specific requirements and data characteristics of the batch job log, select the corresponding large model for analysis; For example, if the batch job log content is in natural language, select the large language model of the Transformer architecture; The large model identifies information in batch job logs based on the training corpus and converts it into feature vectors. It also identifies abnormal situations based on the feature vectors and marks abnormal batch job logs, such as task timeout, failure, or abnormal exit.

[0018] Batch job logs can also be classified to mark different types of batch job events, thereby providing support for subsequent anomaly detection.

[0019] S3: Generate execution report; The alarm and feedback module monitors each process of batch job log execution in real time and generates monitoring data; the monitoring data includes batch job log execution status, configuration information, system resource usage, etc.; The intelligent report generation module collects the monitoring data of batch job logs in real time or periodically, organizes and formats them, converts them into vector forms that can be processed by the report generation model, and inputs the prepared data into the trained report generation model; The report generation model generates the execution report text according to the predefined report template and logical structure, and checks whether the execution report text content is complete and whether it contains all the batch job log monitoring data, and promptly modifies the monitoring data that does not meet the requirements; Furthermore, the report generation model generates corresponding chart descriptions based on the input content, and the chart descriptions include a success rate change trend chart, a resource utilization rate bar chart, etc.; Create corresponding visual charts according to the generated chart description; Use Matplotlib data visualization tools to present data in an intuitive graphical form. The bar chart shows the success rate comparison of different types of batch jobs, the line chart shows the trend of success rate changes over time, and the pie chart shows the distribution of batch jobs in various business modules. Embed these visualization charts in the corresponding positions of the report to enhance the intuitiveness and persuasiveness of the report.

[0020] S4: Providing warning and feedback; The alarm and feedback module issues an alarm based on the data in the visual chart, and the alarm includes three levels: low, medium, and high; Set low-level alarms for minor resource fluctuations and other issues, and set high-level alarms for failures in key batch job logs or abnormal business indicators that seriously affect the business. At the same time, an alarm escalation mechanism is set up. If a low-level alarm is not handled within a certain period of time (1 hour) and the problem persists, it will be upgraded to a medium-level alarm, and so on, to ensure that important issues can receive timely attention; After the alarm rule is triggered, a report is automatically generated and sent to relevant personnel via email, text message, instant messaging tools, etc.; when a high-level alarm occurs, an email is sent to the batch operation and maintenance team members, and a text message reminder is sent to the on-duty personnel at the same time to increase the timeliness and reliability of the notification.

[0021] The title of the alarm information clearly states the alarm level and batch job name, and the body of the alarm information describes the abnormal situation and related data in detail. At the same time, the alarm message is highlighted in red in the instant messaging tool, and a simple chart is added to the email to assist in understanding the alarm content and provide preliminary processing guidance.

[0022] S5: Optimize and learn; We continuously use the ever-increasing historical data to comprehensively evaluate the trained model, calculate the accuracy, recall, F1 value and other evaluation indicators, and calculate the confusion matrix to analyze the model's predictions in different categories in detail; we regularly conduct in-depth analysis of the model's performance to identify possible performance bottlenecks. This includes analyzing the performance differences of the model when processing different types of batch job data, such as logs generated by different business systems and batch jobs of different complexity; by analyzing the model's computational complexity, memory usage, and data processing speed, we identify factors that affect the model's performance, such as overly complex model structure, insufficient data preprocessing, and training data quality issues.

[0023] According to the model evaluation results and performance analysis, take corresponding optimization strategies. If the model has low accuracy in some categories, you can increase the training data related to these categories, or adjust the structure of the model to enhance the processing ability of these categories of data. For performance bottleneck problems, such as slow model calculation speed, you can use model compression technologies such as pruning and quantization to reduce the model's calculation amount and storage requirements, while keeping the model's performance within an acceptable range.

[0024] The present invention has been described by the above-mentioned relevant embodiments, however, the above-mentioned embodiments are only examples for implementing the present invention. It must be pointed out that the disclosed embodiments do not limit the scope of the present invention. On the contrary, changes and modifications made without departing from the spirit and scope of the present invention are all within the scope of patent protection of the present invention.

Claims

1. A batch job monitoring method for the insurance industry based on a large language model, characterized by: The following steps are involved: S1: Data collection; The following sub-steps are included: S11: Set data source; Set up log files, database tables, and file storage related to batch jobs in various systems of the insurance company as data sources; S12: Determine the batch job log to be collected; Determine the batch job logs to be collected based on actual insurance business needs and pre-set batch job monitoring targets; the batch job logs include basic information of the batch job, execution status, time-related information, and data of business-related indicators; S13: Setting data collection frequency; According to the actual batch job operation cycle and monitoring requirements, set different collection frequencies and corresponding data time ranges for high-frequency batch jobs and low-frequency batch jobs; S14: Collect batch job logs; Collect the batch job logs required in the batch job tasks from the data source according to the set collection frequency through API interface, message queue, and log collection tool; transmit the batch job logs to the central data warehouse; S2: Perform semantic analysis on the collected batch job logs; The following sub-steps are included: S21: Processing the batch job log; S22: Use a large model to perform semantic analysis on batch job logs; S3: Generate execution report; The alarm and feedback module monitors the various processes of batch job log execution in real time and generates monitoring data, which is input into the trained report generation model; The report generation model generates the execution report text, generates the corresponding chart description based on the input content, and creates the corresponding visual chart; S4: Providing warning and feedback; The alarm and feedback module issues an alarm based on the data in the visual chart, and the alarm includes three levels: low, medium, and high; Set low-level alarms for minor resource fluctuations and other issues, and set high-level alarms for failures in key batch job logs or abnormal business indicators that seriously affect the business. At the same time, an alarm upgrade mechanism is set. If a low-level alarm is not handled within a preset time and the problem persists, it will be upgraded to a medium-level alarm. After the alarm rule is triggered, a report is generated and sent to relevant personnel; when a high-level alarm occurs, an email is sent to the batch operation and maintenance team members, and a text message reminder is sent to the on-duty personnel at the same time; S5: Optimize and learn; Continuously use the ever-increasing historical data to comprehensively evaluate the trained model, and adopt corresponding optimization strategies based on the model evaluation results and performance analysis.

2. The insurance industry batch job monitoring method based on a large language model as claimed in claim 1, characterized in that: In step S14, reasonable rules for batch operation data and necessary batch operation data elements are set in advance, and after the collection is completed, integrity check and rationality check are performed on the collected batch operation logs; Specifically, the data verification function is used to check whether each collected batch job log contains the necessary batch job data elements. If the necessary batch job data elements are missing, the integrity check fails, an abnormal reminder is given, and the failure reason is recorded; otherwise, the integrity check succeeds; Compare the collected batch job logs with the set batch job data reasonable rules to check whether the collected batch job logs comply with the set batch job data reasonable rules. If not, the rationality check fails, an abnormal reminder is given and the reason for the failure is recorded; otherwise, the rationality check succeeds.

3. The insurance industry batch job monitoring method based on a large language model as claimed in claim 1, characterized in that: The specific contents of step S2 are as follows: S21: Processing the batch job log; The central database receives the batch job logs and performs deduplication processing on the batch job logs by using a unique constraint in the database; The engineer modifies and supplements the batch job log that fails the integrity check and rationality check according to the failure reason recorded in the batch job log in step S1, combined with the log context content or other related logs. If the modification and supplementation are not possible, the corresponding data of the batch job log is marked as suspicious data; Remove irrelevant data from batch job logs and unify the data in batch job logs into a unified format; S22: Use a large model to perform semantic analysis on batch job logs; According to the specific requirements and data characteristics of the batch job log, select the corresponding large model for analysis; The large model identifies information in batch job logs based on the training corpus and converts it into feature vectors; it also identifies anomalies based on the feature vectors and marks abnormal batch job logs.

4. The insurance industry batch job monitoring method based on a large language model as claimed in claim 1, characterized in that: The specific contents of step S3 are as follows: the alarm and feedback module monitors each process of batch job log execution in real time and generates monitoring data; The intelligent report generation module collects the monitoring data of batch job logs in real time or periodically, organizes and formats them, converts them into vector forms that can be processed by the report generation model, and inputs the prepared data into the trained report generation model; The report generation model generates the execution report text according to the predefined report template and logical structure, and checks whether the execution report text content is complete and whether it contains all the batch job log monitoring data, and promptly modifies the monitoring data that does not meet the requirements; Furthermore, the report generation model generates a corresponding chart description based on the input content, and creates a corresponding visual chart according to the generated chart description.

5. The insurance industry batch job monitoring method based on a large language model as claimed in claim 1, characterized in that: It includes two forms: real-time collection and batch collection. Real-time collection is used to collect real-time batch job data, and batch collection is used to collect historical batch job data. Select the corresponding collection form according to actual needs.

6. A batch operation monitoring system for the insurance industry based on a large language model, used to implement any of the batch operation monitoring methods for the insurance industry based on a large language model described in claims 1-5, characterized in that: Including data collection module, log analysis module, intelligent report generation module, alarm and feedback module, learning and optimization module; The data acquisition module is used to obtain batch operation logs from various systems of the insurance company and transmit them to the central database; The log analysis module performs semantic analysis on the log data based on the large language model; The intelligent report generation module is used to generate a detailed execution report; The alarm and feedback module is used to monitor the various processes of batch job log execution in real time. When an abnormality is detected, the alarm information will be sent to the relevant operation and maintenance personnel through email, SMS or instant messaging tools; The learning and optimization module is used to continuously optimize the model based on the continuously accumulated historical data.

Images