Studying the correlation between different word sense disambiguation methods and summarization effectiveness in...

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Information Retrieval

Background:

• Word Sense Disambiguation (WSD) identifies correct word meanings from context, crucial for knowledge-based automatic summarization.
• Previous studies have not investigated the direct correlation between WSD accuracy and summarization performance.

Purpose of the Study:

• To evaluate the impact of WSD techniques on a graph-based automatic summarization system.
• To investigate the correlation between WSD performance and summarization outcomes using the UMLS Metathesaurus.

Main Methods:

• Three knowledge-based WSD approaches and a graph-based summarizer were implemented.
• The Unified Medical Language System (UMLS) Metathesaurus served as the knowledge source for both WSD and summarization.
• WSD methods were evaluated directly against reference datasets (NLM WSD, MSH WSD) and indirectly within the summarization task.

Main Results:

• WSD techniques positively impacted the graph-based summarizer's results.
• A correlation was observed between WSD and summarization task performance on large, homogeneous evaluation collections.
• The top-performing WSD algorithm generally excelled in both WSD and summarization evaluations.

Conclusions:

• While WSD improves summarization, the degree of improvement does not directly correlate with WSD accuracy.
• Disambiguation error importance varies based on concept salience within the document, influencing summarization outcomes.
• The study highlights the nuanced relationship between WSD accuracy and its practical benefit in automatic summarization.