Linked Data Triples Enhance Document Relevance Classification

Standardized approaches to relevance classification in information retrieval use generative statistical models identify the presence or absence of certain topics that might make a document relevant searcher. These have been used better predict on basis what is “about”, rather than simple-minded analysis bag words contained within document. In more recent times, this idea has extended by using pre-trained deep learning and text representations, such as GloVe BERT. an external corpus knowledge-base conditions model help about. This paper adopts hybrid approach leverages structure knowledge embedded corpus. particular, reports experiments where linked data triples (subject-predicate-object), constructed from natural language elements are derived learning. evaluated additional latent semantic features for classifier customized news-feed website. The research synthesis current thinking NLP predicate web research. Our indicate increased F-score baseline representations 6% show significant improvement over state-of-the art models, like findings tested empirically validated experimental dataset two standardized pre-classified news sources, namely Reuters 20 News groups datasets.