Microblog topic identification using Linked Open Data.

Area of Science:

• Social Media Analysis
• Natural Language Processing
• Knowledge Representation

Background:

• Social media posts contain valuable information but are challenging to utilize due to volume and distributed nature.
• Existing topic detection methods often require separate interpretation steps, limiting machine interpretability.

Purpose of the Study:

• To propose a novel approach for identifying machine-interpretable topics from social media microposts.
• To represent topics using an ontology and Linked Open Data (LOD) for enhanced processing and insights.

Main Methods:

• Defined topics as sets of related elements posted in similar contexts.
• Introduced a W3C-standard ontology for topic representation.
• Utilized entity linking to Linked Open Data (LOD) resources to identify topic elements.
• Generated topics from over one million tweets.

Main Results:

• Demonstrated the feasibility of the approach with a large Twitter dataset.
• Showcased the utility of the generated topics in various topic-related tasks.
• Achieved 81.0% precision and 93.3% F1 score in manual evaluations.

Conclusions:

• The proposed ontology-based approach effectively identifies machine-interpretable topics from social media.
• This method provides richer insights compared to traditional word-list-based topic representations.
• The approach enhances automated analysis and understanding of collective interests expressed online.