A vector reconstruction based clustering algorithm particularly for large-scale text collection.

Area of Science:

• Computer Science
• Data Mining
• Machine Learning

Background:

• Internet technology generates vast amounts of textual data daily.
• Organizing text into categories is crucial for information retrieval.
• Traditional clustering algorithms struggle with large-scale, high-dimensional text data due to semantic similarities.

Purpose of the Study:

• To develop an effective and efficient clustering algorithm for large-scale text collections.
• To address the limitations of traditional clustering methods in high-dimensional and semantically similar text data.
• To improve the quality and performance of text categorization.

Main Methods:

• Proposes a vector reconstruction based clustering algorithm.
• Utilizes a partial tuning sub-process with feature weight fine-tuning (similar to Self-Organizing-Mapping).
• Incorporates an intersection-based similarity measurement and neuron adjustment function for accelerated clustering.
• Employs an overall tuning sub-process to reallocate features and remove non-representative ones.

Main Results:

• The algorithm preserves only cluster-representative features in representative vectors.
• Experimental results show high-quality performance on both small-scale and large-scale text collections.
• Demonstrates effectiveness in handling high-dimensional vector spaces and semantic similarity.

Conclusions:

• The proposed vector reconstruction clustering algorithm is effective for categorizing large-scale text collections.
• The algorithm overcomes limitations of traditional methods in handling complex text data.
• Achieves high-quality clustering performance, enhancing information organization and retrieval.