LLM Fraud Report Generation for Explainable Content Blocking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ML model-based systems face challenges in generating consistent and user-specific reports explaining the detection of fraudulent content, such as phishing emails, due to the complexity of multiple ML models and diverse features involved.
Innovation Solution
A system utilizing a large language model (LLM) to generate personalized fraud reports by extracting features from content, classifying them into fraud categories, and generating input prompts for the LLM to produce detailed reports without human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple ML models with diverse features are used to detect various categories of fraudulent content, then the detection capability and accuracy are improved, but the complexity of generating consistent and customized reports increases significantly
Solution Approach 1:
The patent merges multiple ML model outputs and diverse features into a unified report generation process. The system consolidates detection results from multiple models and synthesizes them into a single consistent report, eliminating the need for separate interpretation processes for each model.
Solution Approach 2:
The report generation system is designed to be universal, handling multiple fraud categories and diverse ML model outputs through a single multi-functional process. This universal approach allows the system to generate customized reports for different user needs while processing various types of fraudulent content detections.
2Reliability
If multiple ML models with diverse features are used to detect various categories of fraudulent content, then the detection capability is improved, but the ability to generate user-specific customized reports becomes impractical
Solution Approach 1:
The report generation process is made dynamic and adaptive, automatically adjusting the level of detail, format, and content based on user-specific requirements. The system can dynamically customize reports for different users while maintaining consistent detection results from the multiple ML models.
Solution Approach 2:
The system changes report parameters such as detail level, format, and focus areas based on user preferences and requirements. This allows the same detection results to be presented in multiple customized formats, enhancing adaptability without compromising detection reliability.
3Productivity
If traditional ML model-based systems are used for fraud detection, then detection functionality is provided, but user understanding of detection reasons is insufficient
Solution Approach 1:
The system introduces an intermediary report generation process that translates complex ML model detection results into user-friendly explanations. This intermediary layer bridges the gap between technical detection mechanisms and user understanding, providing clear reasons for detections without reducing detection efficiency.
Solution Approach 2:
The system automatically generates comprehensive explanations and reasoning for detection decisions without requiring additional human intervention. The self-service approach provides users with detailed information about detection reasons, maintaining high productivity while reducing information loss.
Data Source
AI summary
A new approach is proposed that supports large language model (LLM)-driven fraud report generation. First, a plurality of features is extracted/derived either directly from an original piece of content/information susceptible of fraud or indirectly from one or more external sources (e.g., statistical data) associated with the piece of content. The plurality of extracted features are then classified into one or more fraud categories using one or more classification models. If a fraud attack is detected, an input prompt is generated based on the plurality of extracted features and the one or more fraud categories related to the specific detection in order to generate a report for a user as to the reason for this detection. Finally, a LLM is utilized to generate a fraud report of the original piece of content for the user based on the input prompt specific to the one or more fraud categories.