Extracting information and inferences from a large text corpus.

Area of Science:

• Computer Science
• Data Science
• Machine Learning

Background:

• Industry 4.0 has led to massive data accumulation, necessitating efficient machine learning for data interpretability.
• Text collections from scientific, biological, and social media require advanced topic modeling for decision-making.
• Existing topic models like LDA and DTM are effective but may not be optimal for dynamic, large-scale data.

Purpose of the Study:

• To analyze and compare the performance of Latent Dirichlet Allocation (LDA), Dynamic Topic Model (DTM), and Embedded Topic Model (ETM).
• To propose an Incremental Topic Model with Word Embedding (ITMWE) for processing large text data in incremental environments.
• To evaluate ITMWE's efficiency in extracting latent topics and grouping documents.

Main Methods:

• Comparative analysis of LDA, DTM, and ETM.
• Development of an incremental topic model incorporating word embeddings (ITMWE).
• Experimental evaluation in both offline and online settings using large datasets (CORD-19, NIPS papers, Tweet data).

Main Results:

• LDA and DTM effectively discover word-level topics.
• ITMWE demonstrates superior performance in identifying document-level topic groups.
• ITMWE shows greater efficiency in dynamic environments crucial for text mining applications.

Conclusions:

• ITMWE offers an efficient approach for topic discovery in large, dynamic text collections.
• The proposed model enhances data interpretability for better organizational decision-making.
• ITMWE is particularly valuable for real-time text mining applications in scientific and social media domains.