Enabling Semantic Topic Modeling on Twitter Using MetaMap.

Area of Science:

• Computational Linguistics
• Social Media Analysis
• Bioinformatics

Background:

• Topic modeling struggles with short, informal text common on social media platforms like X.
• Existing methods often fail to capture nuanced meanings in noisy datasets.

Purpose of the Study:

• To evaluate MetaMap's effectiveness in enhancing semantic topic modeling for social media data.
• To compare topic modeling performance using original tweets versus tweets with standardized UMLS Concept Unique Identifiers (CUIs).

Main Methods:

• Utilized 56,017 tweets mentioning "hydroxychloroquine" (03/01/2020-12/31/2021) as a case study.
• Applied MetaMap to extract and encode concepts as UMLS CUIs.
• Employed Latent Dirichlet Allocation (LDA) for topic modeling on original and CUI-encoded datasets.

Main Results:

• MetaMap-assisted LDA models demonstrated superior coherence and representativeness compared to baseline models.
• The semantic approach successfully identified topics relevant to concurrent social and political discourse.
• Standardization via UMLS CUIs significantly improved topic model performance.

Conclusions:

• Integrating MetaMap with topic modeling enhances semantic understanding of social media text.
• UMLS concept standardization is a viable strategy to mitigate noise and improve topic discovery in platforms like X.
• This approach offers improved performance for analyzing complex, real-world textual data.