TopicWeb: A Novel Approach to Automatic Document Similarity Measurement and Categorization

The project introduces the TopicWeb, a system for improved similarity measurement between two textual documents (news articles, emails, etc). It consists of an interconnected web of documents, where a connection between two documents specifies the strength of the similarity between the topics of those documents. A standard method of calculating similarity between two documents is improved upon by examining not only the two documents themselves, but also their neighbors in the TopicWeb. This method effectively differentiates between documents with topics that may be similar sounding to a computer like ”heart surgery” and ”Valentine’s day heart” but are fundamentally different.