Predicting Marathi News Class Using Semantic Entity-Driven Clustering Approach

Abstract

Document management is a need for an era and managing documents in the regional languages is a significant and untouched area. Marathi corpus consisting of news is processed to form Group Entity document matrix Marathi (GEDMM), Vector space model for Marathi (VSMM) and Hysynset Vector space model for Marathi (HSVSMM). GEDMM uses entity group extracted using Condition random field (CRF). The frequent terms are used to construct VSMM using TF-IDF. HSVSMM uses synsets using hypernyms-hyponyms and synonyms. GEDMM and HSVSMM use dimension reduction by selecting significant feature groups. Hierarchical agglomerative clustering (HAC) is used and a dendrogram is produced to visualize the clusters. The performance analysis is carried out using several parameters like entropy, purity, misclassification error and accuracy. The clusters produced using GEDMM shows the minimum entropy and the highest purity. A random forest classifier is applied and the results are evaluated using misclassification error and accuracy.