Cosine Proximity Codebooks for Short-Text Sentiment Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing user text input, such as customer reviews, are inefficient and rudimentary, particularly for freeform text requiring natural language processing, and do not effectively utilize advanced language processing techniques.
Innovation Solution
Utilizing BERT analysis and clustering algorithms to partition text into segments, generate numerical vectors, and classify them based on cosine proximity thresholds, enabling efficient categorization of sentiments in user comments through a graphical user interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If topic modeling is used to analyze user text input, then topics can be found within large subsets of text, but it performs poorly when user text input is short (fifty words or less)
Solution Approach 1:
The patent transforms the text analysis approach by converting short text segments into numerical vectors using BERT embeddings, changing the parameter representation from raw text to dense vector space where semantic similarity can be effectively measured through cosine proximity, enabling accurate analysis even with limited text volume
2Ease of operation
If database text search is used to analyze customer reviews, then reviews containing searched words can be retrieved, but the search is rarely comprehensive due to variations in word forms and spelling errors
Solution Approach 1:
The patent replaces the mechanical keyword-matching search system with a semantic similarity-based system using BERT embeddings and cosine proximity calculation, substituting exact word matching with vector-based semantic comparison that captures meaning regardless of word form variations or spelling errors
3Ease of manufacture
If star-rating system is used to analyze customer reviews, then reviews can be categorized and sorted by star values, but the method is rudimentary and does not provide efficient natural language processing
Solution Approach 1:
The patent introduces BERT embeddings as an intermediary layer between raw text input and analysis output, transforming text into numerical vectors that enable sophisticated semantic comparison and clustering, thereby enhancing the rudimentary star-rating system with advanced natural language processing capabilities
4Measurement precision
If advanced language processing methods are applied to topic modeling, then analysis accuracy may improve, but the complexity of the system increases
Solution Approach 1:
The patent segments the text analysis process into distinct modular steps: text input segmentation, BERT embedding generation, cosine proximity calculation, and clustering algorithm application. This segmentation allows advanced language processing to be applied systematically while managing complexity through clear separation of functions
Data Source
AI summary
Certain aspects of the disclosure provide a system and method for generating codes associated with clusters of sentiments, comprising receiving a text input, partitioning the text input into one or more segments, generating one or more numerical vectors associated with each of the one or more segments, comparing the one or more numerical vectors to generate a plurality of cosine proximity values associated with the one or more numerical vectors, applying a clustering algorithm to the one or more numerical vectors to generate clusters of segments within one or more cosine proximity ranges, generating one or more codes associated with each of the clusters of segments within the one or more cosine proximity ranges, wherein each cluster represents an overall sentiment, and netting the one or more codes into one or more categories by inputting the one or more codes into a large-language model.