Category-based similarity algorithm for semantic similarity in multi-agent information sharing systems

Similarity measures are mechanisms that assign a numeric score indicating how closely two documents, or a document and a query match. Most similarity measures such as cosine measure, which treat a document as a vector of weighted keywords, consider exact matching of keywords when determining the similarity among documents and they do not consider the semantic similarity among the keywords of the documents. This paper presents a category-based similarity algorithm (CSA) to determine the semantic similarity between any two pieces of information. CSA is implemented inside the ACORN (agent-based community oriented routing network) system, which is a multi-agent system for information retrieval and provision in a community of users. CSA can also be used in any information sharing system in which the information content is represented as vectors of weighted keywords.