Feature Word Extraction and Topic Grouping from User Reviews
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Enterprises face challenges in analyzing vast amounts of user reviews for products or services, as manually reading each review is impractical and sampling may lead to incorrect conclusions, necessitating a more efficient method for extracting meaningful insights.
Innovation Solution
A system and method for extracting and grouping feature words from a corpus of text bodies, involving the creation of a second corpus with filtered reviews, preliminary topic assignment using Latent Dirichlet Allocation (LDA), and final topic assignment based on topic count values, allowing for the presentation of feature words as groups on a user interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual reading of each review is performed, then analysis accuracy is improved, but time consumption and labor cost increase significantly
Solution Approach 1:
The patent replaces manual mechanical reading and analysis of reviews with an automated computational system that uses text processing algorithms and machine learning models to extract feature words, determine topics, and analyze sentiments, thereby eliminating time-consuming human labor while maintaining or improving analysis accuracy
Solution Approach 2:
The system creates a simplified representation (copy) of the review data by extracting key feature words and topics, allowing analysis to be performed on this condensed version rather than the full text of each review, thus reducing processing time while preserving essential information
2Loss of time
If sampling of reviews is performed, then time consumption is reduced, but analysis accuracy deteriorates due to incomplete representation
Solution Approach 1:
The system extracts the most informative feature words and topics from the review corpus using automated text processing, identifying and isolating key elements that represent the overall sentiment and content without requiring manual selection or sampling of individual reviews
Solution Approach 2:
The patent transforms the review data by converting unstructured text into structured feature word frequencies and topic distributions, changing the parameters from raw text to quantifiable metrics that enable comprehensive analysis of the entire corpus efficiently
3Loss of information
If comprehensive analysis of all reviews is performed, then analysis completeness is improved, but processing complexity increases
Solution Approach 1:
The patent segments the complex task of comprehensive review analysis into distinct modular steps: feature word extraction, topic determination, sentiment analysis, and result aggregation. Each module handles a specific aspect independently, reducing overall processing complexity while maintaining complete analysis coverage
Solution Approach 2:
The system introduces intermediate data structures (feature word lists, topic models, sentiment scores) that serve as mediators between the raw review corpus and the final analysis results, simplifying the processing pipeline by breaking down the complex transformation into manageable intermediate stages
Data Source
AI summary
Various embodiments of systems and methods for extraction and grouping of feature words are described herein. Feature words are obtained from a first corpus of text bodies comprising a plurality of reviews. A second corpus is created using a combination of the obtained feature words, verbs and adjectives from the first corpus. The second corpus comprises filtered reviews and each of the filtered reviews pertains to a review. Topics are preliminarily assigned for words in the filtered reviews of the second corpus. For each of the feature words in the second corpus, a topic count is determined for every preliminarily assigned topic. After determining the topic count, one or more of the topics are finally assigned to the feature words based on a topic count value. At least one topic is presented as a group of the feature words for which the at least one topic is assigned based on the topic count value.