Integration of Semantic Features for Short and Long Text Classification using Attention Based DL

The categorization of text is an essential and pivotal step in the process of natural language understanding and processing. There are a lot of different approaches to solving the problem of classifying text, but very few of them leverage the semantic integration of several views to boost the classification performance. A dual-DL attention network model, referred to as DDL, is proposed in the study. In order to address the knowledge gap, this paradigm employs semantic complementarity. For more precise extraction of textual logical semantics, DDLfirst employs transductive learning and graph structure. The next step is to use logical semantics in the attention fusion layer (Channel) to train other semantics jointly, which will gradually fix the predictions produced using unlabeled test data. Important for upcoming text mining jobs, experiments show that DDL can achieve better classification across a variety of text classification datasets.