Exploring Term Weighting Methods for Kannada Document Classification

The documents classification is one of the vital natural language understanding tasks. The vector space model is used to represent the documents in terms space as vectors. The term weighting schemes are part of vector space model to represent the documents as vectors. In this paper, extensive empirical evaluations are conducted by applying various term weighting methods in combination of symbolic representation and selection method for two standard Kannada document datasets. Further, their results are compared with each other. In all experiments six different state-of-the-art classifiers are used for Kannada document classification. On the other hand, since documents analysis suffers from dimensionality problems the Regularized Locality Preserving Indexing (RLPI) technique is applied, and further experimental results are conducted with various term weighting methods. The symbolic classifier for Symbolic Representation and Selection Positional encoded-Term Frequency-Inverse Document Frequency (SRS_PE-TF-IDF) representation yielded highest accuracy of 93.57% for large dataset with only 40 dimensions.