Semantic content based news recommendation system for cross-lingual context

Abstract

With the advancement of information technology, a huge amount of heterogeneous information of different languages are available to us. But handling these multilingual information effi¬ciently is still a great challenge in current web technology. In overcoming these challenges, semantic recommendation system can play a vital role. In this thesis, we propose an approach for an automated Bengali-English semantic recommendation system based on ontology by ana¬lyzing news domain. News ontology is designed automatically by using information extraction techniques. Both the news title and news body are considered separately in the ontology cre¬ation process. First, important information from news is extracted and ontology is created from the source language document. Then, ontology is created from target language document fol¬lowing similar technique. Next, ontology matching is performed between the translated source ontology and target English ontology. Matching can also be done with synonymous documents. A matching factor is calculated which can be taken as the semantic similarity measure between the cross-lingual documents. Recommendation of news items is done based on this matching factor. Our proposed method can recommend similar news items written in different languages. Validity of our claims have been substantiated through different experiments on news items collected from different news portals.