A Semantic Network Approach to Measuring Relatedness

Humans are very good at judging the strength of relationships between two terms, a task which, if it can be automated, would be useful in a range of applications. Systems attempting to solve this problem automatically have traditionally either used relative positioning in lexical resources such as WordNet, or distributional relationships in large corpora. This paper proposes a new approach, whereby relationships are derived from natural language text by using existing nlp tools, then integrated into a large scale semantic network. Spreading activation is then used on this network in order to judge the strengths of all relationships connecting the terms. In comparisons with human measurements, this approach was able to obtain results on par with the best purpose built systems, using only a relatively small corpus extracted from the web. This is particularly impressive, as the network creation system is a general tool for information collection and integration, and is not specifically d...

• Type: Journal
• Year: 2010
• Where: COLING
• Authors: Brian Harrington
• Keywords: COLING 2010, Computational Linguistics, Natural Language Text, Purpose Built Systems, Scale Semantic Network