Bridging Social Tags and Semantics: Generating User Interest Ontologies via ID3

TL;DR

The proliferation of Social Networking Services (SNS) has created a goldmine of user interest data, often captured as unstructured "tags." This paper presents a methodology to transform these messy tags into a rigorous User Interest Ontology. By leveraging the ID3 Decision Tree algorithm to define domain boundaries and OWL-DL for semantic representation, the authors achieve over 93% accuracy in classifying domain-specific interests.

Background: The "Domain Interest" Gap

In the academic landscape of user modeling, there is a clear divide:

1. Individual Interests: Highly personalized but lack universality.
2. Public Interests: Universal but lack personality.
3. Domain Interests: The sweet spot—shared interests of users within a specific field (e.g., politics, education).

The problem? Computers don't "understand" tags. A tag like "Obama" might relate to "Politics," "Election," or "The New Yorker." Without a semantic structure, classifying users into specific domains is a manual, inefficient task.

Methodology: From Decision Trees to Logic Rules

The core innovation lies in the automated pipeline that converts statistical importance into logical rules.

1. Attribute Selection via ID3

The authors used data from Delicious (a social bookmarking site). They extracted the top 5 tags for web pages and created an ID3 table. By calculating Information Gain, the system determines which tags (attributes) are the best predictors for a specific domain.

2. Semantic Translation (OWL-DL)

Once the ID3 tree is built, every "pathway" from the root to a leaf represents a classification rule. These pathways are translated into OWL-DL (Web Ontology Language).

Fig 1: A specific pathway in the domain tree used to generate semantic rules.

For example, a pathway consisting of tags like [Obama, Election, 2012] is converted into a formal logic statement (TBox/RBox). This allows the computer to perform Inference—if a new web page has certain tags, the ontology "reasons" that it belongs to the "Obama Webpage" class.

Fig 2: Example of OWL-DL tag rules (TBox) derived from the ID3 algorithm.

Experimental Results: The 2012 Presidential Election

The authors tested their approach on the 2012 US Election domain, focusing on three candidates: Barack Obama, Mitt Romney, and Michele Bachmann.

• Dataset: Nearly 515,000 web pages from Delicious.
• Performance: The system demonstrated high robustness.

The high accuracy proves that the ID3 algorithm effectively captures the "Inductive Bias" needed to group social tags into meaningful semantic categories.

Critical Analysis & Future Work

Why it Works

Traditional Folksonomies suffer from polysemy (one word, many meanings) and synonymy (many words, one meaning). By using OWL-DL, the authors provide a formal structure that resolves these ambiguities. The use of ID3 ensures that the "rules" of the ontology are grounded in actual user behavior (statistical tag frequency) rather than just manual expert design.

Limitations

1. Tag Quality: The system relies on the "Top 5 tags." If users provide low-quality or spam tags, the ID3 table becomes noisy.
2. Temporal Dynamics: Interests in politics change rapidly. The paper does not address how the ontology should "evolve" as new tags emerge (e.g., a new candidate entering the race).

Conclusion

This research provides a bridge between the Social Web (Web 2.0) and the Semantic Web (Web 3.0). For developers building recommendation engines or political sentiment analysis tools, this hybrid approach offers a way to move beyond simple keyword matching toward true semantic understanding.