Long Text Clustering via External Label Information
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The 'collapse' phenomenon in long text clustering, where sentence vectors are encoded into a smaller spatial region, leading to high similarity scores for semantically unrelated words, resulting in poor performance due to lack of 'difference' marking training in semantic vector representation methods.
Innovation Solution
Introducing external label information using open-source text labeling models to perform clustering at a dual level, combining tags and original semantics, and employing a graph structure-based method for optimized clustering.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If sentence vectors are encoded into a smaller spatial region, then the clustering process can be performed more efficiently, but the similarity scores become inflated for semantically unrelated words, leading to poor clustering performance
Solution Approach 1:
The patent segments the clustering process into two distinct levels: a coarse-grained level using external label information (tags) and a fine-grained level using semantic vector similarity. This segmentation allows the system to first filter and organize data using high-level labels, then perform precise similarity matching within relevant subsets, thereby maintaining both efficiency and accuracy.
Solution Approach 2:
The patent introduces external label information (tags) as an intermediary between the raw text data and the final clustering results. These tags act as a mediator that provides semantic structure and organization, enabling the system to achieve accurate similarity measurement without being overwhelmed by the computational complexity of processing all text vectors directly.
2Device complexity
If semantic vector representation methods are used without external label information, then the clustering process is simpler, but the 'collapse' phenomenon occurs where semantically unrelated words are incorrectly grouped together
Solution Approach 1:
The patent merges two previously separate approaches: semantic vector representation and external label-based classification. By combining these methods, the system leverages the computational efficiency of semantic vectors while incorporating the discriminative power of external labels, thereby preventing the collapse phenomenon and improving clustering reliability.
Solution Approach 2:
The patent creates a composite clustering approach that integrates multiple data representations: semantic vector embeddings and external label tags. This composite structure allows the system to benefit from both the semantic richness of vectors and the categorical organization of labels, preventing the harmful effects of using either method alone.
3Measurement precision
If external label information is introduced for clustering, then the clustering accuracy improves, but the computational resources and processing time increase
Solution Approach 1:
The patent applies preliminary action by first extracting and organizing external label information before performing the main clustering computation. By pre-processing and structuring the label data, the system creates an organized framework that guides the subsequent clustering process, reducing the computational burden and improving efficiency.
Solution Approach 2:
The patent implements a dynamic two-level clustering architecture where the system can adaptively adjust the depth and scope of analysis. The coarse-grained label-based clustering provides a flexible first pass that can be refined with fine-grained semantic vector matching only when necessary, allowing the system to optimize computational resource consumption based on data characteristics and requirements.
Data Source
AI summary
In an approach for using an open source of existing text labeling models to label sentences that need to be clustered with multiple external tags and then to use the tags as auxiliary information to perform the clustering at a dual level, a processor receives a set of text, wherein the set of text contains one or more sentences. A processor tags each sentence of the set of text with one or more tags using a plurality of open-source text classification models. A processor performs a preliminary clustering of one or more nodes under strict conditions using a canopy clustering algorithm.