Mined Semantic Analysis - Published Patent Application

In plain English, MSA is able to closely approach human level concept association (scoring over 90% in the Miller Charles noun-pair tests and 70% in more demanding Mechanical Turk word-pair tests). What is MSA? Unlike other corpus-based methods which look for direct associations between concepts and terms MSA discovers latent associations by mining concept-concept association rules and uses these rules to induce implicit associations between terms and concepts. It can identify not only concepts directly related to the given linguistic item but also other latent concepts associated with them (implicitly). It could be applied in semantic search, textual entailment, word sense disambiguation, vocabulary mismatch resolution, concept tracking, technology mappings, and others. The algorithm is efficient because it employs an inverted index to retrieve the semantically related concepts. Additionally, mining for association rules is done offline making it scalable to huge amounts of data.