A novel multiple kernel fuzzy topic modeling technique for biomedical data.

Area of Science:

• Biomedical Informatics
• Computational Linguistics
• Data Science

Background:

• Biomedical text mining is crucial due to vast amounts of unstructured text data.
• Topic modeling uncovers hidden semantic structures but faces challenges like sparsity and redundancy in biomedical data.

Purpose of the Study:

• To propose a novel Multiple Kernel Fuzzy Topic Modeling (MKFTM) technique for enhanced biomedical text mining.
• To address sparsity and redundancy issues in biomedical text data.
• To improve the accuracy of topic discovery, classification, and clustering.

Main Methods:

• Developed a novel MKFTM technique integrating fusion probabilistic inverse document frequency (IPF) and multiple kernel fuzzy c-means clustering.
• Utilized fusion IPF for global term weighting and MKFTM for local and global term frequency generation.
• Applied Principal Component Analysis (PCA) to mitigate higher-order negative effects on term weights.

Main Results:

• MKFTM achieved superior classification accuracy, reaching up to 99.69% on the Muchmore Springer dataset and 94.10% on the Ohsumed dataset.
• The clustering performance, indicated by a higher CH index, surpassed existing state-of-the-art topic models.
• Experiments on six biomedical datasets demonstrated the effectiveness of the proposed method.

Conclusions:

• The MKFTM approach efficiently handles sparsity and redundancy in biomedical text documents.
• MKFTM discovers semantically relevant topics with high accuracy, improving classification and clustering.
• MKFTM offers a flexible new approach to topic modeling adaptable to various clustering methods.