Topic selection for text classification using ensemble topic modeling with grouping, scoring, and modeling approach.

Area of Science:

• Computational biology
• Bioinformatics
• Natural Language Processing

Background:

• TextNetTopics uses topic modeling for text classification, reducing dimensionality while retaining semantic information.
• Individual topic models can exhibit performance variability in text classification tasks.

Purpose of the Study:

• To introduce Ensemble Topic Model for Topic Selection (ENTM-TS) as an advancement over TextNetTopics.
• To mitigate performance variability by integrating multiple topic models.
• To evaluate ENTM-TS and compare TextNetTopics with various topic modeling algorithms.

Main Methods:

• Developed ENTM-TS, integrating multiple topic models via Grouping, Scoring, and Modeling.
• Conducted a comparative study of eleven state-of-the-art topic modeling algorithms within TextNetTopics.
• Utilized the Drug-Induced Liver Injury and WOS-5736 datasets for comprehensive evaluation.

Main Results:

• Latent Semantic Indexing demonstrated comparable performance with fewer features than other methods.
• ENTM-TS performance matched or surpassed optimal individual topic models on both datasets.
• The study identified Latent Semantic Indexing as a strong performer for feature extraction.

Conclusions:

• ENTM-TS is a robust and effective enhancement for text classification tasks.
• The ensemble approach effectively addresses the performance variability of individual topic models.
• The findings provide valuable insights into selecting optimal topic models for text classification.