A NOVEL APPROACH FOR TEXT SIMILARITY MEASURE AND CLASSIFICATION

Abstract

In the text processing field finding the similarity between multiple documents is an important operation. In
this paper, we proposed a new similarity measure for document clustering. To figure out the similarity
between multiple documents with respect to a feature, our proposed similarity finding measure takes the
following cases into account:
1) The selected feature may appear in both documents, 2) the selected feature appears in only one document,
and 3) the selected feature appears in none of the documents. In the first case, the documents similarity
actually increases as the difference between the selected involved features values are less. Moreover, the
involvement of the difference is normally scaled by feature values. However in the second case, a constant
value is involved to find the similarity and in the last case, the selected feature are absent between the
documents and thus has no contribution to the document similarity. Our proposed measure is extended to
estimate the appropriate similarity between two document sets to get effective results with better performance.