Prompt Compression Using Weighted Topic Terms for Document Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current topic modeling techniques face limitations in efficiently extracting meaningful insights from large volumes of unstructured data, particularly in generating accurate topic labels and descriptions.
Innovation Solution
A system and method involving an unsupervised machine learning model to generate topics from a set of documents, selecting relevant topic terms based on weight values, computing inverse document frequency weights, and inputting these terms into a language model to produce compressed representations for generating topic labels and descriptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If all topic terms are included in the prompt for language models, then the completeness of topic information is improved, but the prompt length and processing complexity increase
Solution Approach 1:
The patent extracts only the most relevant topic terms from the complete topic model output by applying weight thresholds and selection criteria. This extraction process removes unnecessary information while retaining the essential topic characteristics, thereby reducing prompt length and processing complexity without significant loss of topic information.
Solution Approach 2:
The patent applies different selection criteria and weight thresholds to different topics based on their specific characteristics. Each topic is processed with locally optimized parameters rather than a uniform approach, allowing the system to maintain high information quality for each topic while adapting to varying prompt requirements and reducing overall complexity.
2Measurement precision
If more topic terms are selected for each topic, then the accuracy of topic labels and descriptions is improved, but the computational resources and time required increase
Solution Approach 1:
The patent selects a partial subset of topic terms that exceeds the minimum required for accurate topic representation but does not include all available terms. This partial action approach achieves sufficient topic accuracy while maintaining reasonable processing efficiency by stopping selection at an optimal threshold rather than processing all terms.
Solution Approach 2:
The patent dynamically adjusts the weight threshold parameter and term selection criteria based on topic characteristics and performance requirements. By changing these parameters, the system can optimize the balance between topic label accuracy and processing efficiency for different scenarios without requiring complete term sets.
3Loss of information
If comprehensive topic analysis is performed on all documents, then the depth of insight is improved, but the processing time and computational load increase
Solution Approach 1:
The patent performs preliminary filtering and weighting of topic terms before the main analysis phase. By pre-computing term weights and identifying high-value terms in advance, the system prepares the data structure to enable deep insight generation from a reduced set of critical terms, thereby reducing the computational burden during subsequent processing while maintaining insight depth.
Data Source
AI summary
A data processing system and method include receiving a set of documents having unstructured data, executing the unsupervised machine learning model for outputting topics, selecting a first subset of topic terms, computing an inverse document frequency weight value for each topic term in the first subset of topic terms, computing a second weight value for each topic term in the first subset of topic terms, selecting a second subset of topic terms from the first subset of topic terms, generating a compressed representation of the set of documents from the second subset of topic terms to include in a prompt, inputting the prompt into a language model, and executing the language model based on the prompt to generate the topic label and the topic description.