Uncovering Flat and Hierarchical Topics by Community Discovery on Word Co-occurrence Network

Area of Science:

• Computational linguistics
• Data mining
• Social sciences

Background:

• Topic modeling is crucial for uncovering latent themes in text collections.
• Applications span sociology, opinion analysis, and media studies.
• Interpretability, diversity, and coherence are key requirements for topics.

Purpose of the Study:

• To introduce Community Topic, a novel algorithm for efficient topic modeling.
• To enable the identification of both flat and hierarchical topics.
• To facilitate on-demand exploration of topic hierarchies (sub- and super-topics).

Main Methods:

• Exploits word co-occurrence networks to mine communities and produce topics.
• Evaluated using multiple metrics and compared against standard baselines.
• Demonstrates effectiveness on multilingual datasets.

Main Results:

• Community Topic successfully identifies flat topics and topic hierarchy.
• The algorithm confirms good performance in evaluations.
• Facilitates efficient exploration of topic structures.

Conclusions:

• Community Topic offers an effective approach to topic modeling.
• The method supports the discovery of interpretable, diverse, and coherent topics.
• Its ability to handle topic hierarchies enhances its utility in various disciplines.