Computing Semantic Relatedness Based on Search Result Analysis

Automatically computing the semantic relatedness of two words is an essential step for many tasks in natural language processing, including information retrieval. Previous approaches to computing semantic relatedness used statistical techniques or lexical resources. We propose Searcher Result Analysis (SRA), a novel method that captures related text from search engine by issuing proper queries. Inferring the relatedness is then based on word occurrences in certain number of pages. Compared with the previous state of the art, using SRA to computing semantic relatedness based on Wikipedia can achieve competitive results with no need to maintain a local copy of remote resources. It is also shown that the correctness can be further improved by selecting proper knowledge resources or corpora for SRA.