A Hybrid Method for Hindi News Recommendation

Abstract

Hindi language has always been a center of attraction for research industries. There are more than 20 Hindi news portals and papers, whereas there are more than 20 crore Hindi readers. The news has become radical these days and hence one area can be important for one reader, whereas it can be completely useless for another. This paper developed a software aided recommendation system (SARS) by applying a three block architecture for Hindi news. A training mechanism of dual channel Long Short Term Memory (LSTM) is outcaste over 30,000 news headlines. The classifications channel is developed by applying pre-processing and weight vector generation followed by simulation against the trained vector. The evaluation of the proposed work is done on the base of precision, recall and f-measure and is compared with one of recent article. The average improvement in the precision is noted to be 31.9% whereas for recall and F-measure, it is 34 and 35% respectively.