TextNetTopics Pro, a topic model-based text classification for short text by integration of semantic and...

Area of Science:

• Computational linguistics
• Bibliometrics
• Machine learning for text analysis

Background:

• Scientific literature is growing exponentially, necessitating automated classification.
• Short texts like article titles pose challenges for traditional text mining due to data sparseness and limited context.
• Previous methods like TextNetTopics showed promise but require enhancement for short-text classification.

Purpose of the Study:

• To evaluate the performance of TextNetTopics on short texts.
• To propose TextNetTopics Pro, a novel framework for short-text classification.
• To improve the classification accuracy of scientific articles using titles.

Main Methods:

• Explored TextNetTopics performance on short texts.
• Developed TextNetTopics Pro, combining lexical features and topic distributions from topic models.
• Evaluated nine state-of-the-art short-text topic models on Biomedical and Computer Science datasets.
• Compared classification performance with and without using article abstracts.
• Assessed robustness on imbalanced data, specifically for Drug-Induced Liver Injury classification.

Main Results:

• TextNetTopics Pro effectively alleviates data sparseness in short-text classification.
• The proposed approach demonstrates robust performance on imbalanced datasets.
• Utilizing semantic information from topic models significantly enhances machine learning classifier performance.
• Comparative evaluation showed the effectiveness of the combined approach over baseline methods.

Conclusions:

• TextNetTopics Pro offers a reliable framework for classifying scientific articles based on titles.
• Leveraging topic models and lexical features is crucial for overcoming challenges in short-text analysis.
• The approach shows strong potential for applications in bibliometrics and scientific information retrieval.