Semantic Based Text Classification Using Deep Learning

Abstract

In recent years abundant data is being produced over the newlineinternet from multiple sources and thereby emerged the term Big Data. In newlinegeneral these raw data does not produce any information. Hence data is newlineprocessed to derive at meaningful insights. A notable amount of this data is newlinecontributed mainly in the form of text rather than other formats. This newlinescenario has automatically gained importance for Natural language newlineprocessing (NLP). Text classification and Sentiment Analysis are few newlineApplications of NLP. The intent of text classification is to assign newlinepredefined labels (single or multiple) to the sequence of text and it is done newlineby various techniques like knowledge engineering, expert system and newlinemachine learning models. Building a model with traditional method is newlinesimple and easy to train for small datasets. To handle larger datasets, neural newlinenetworks came into existence. Deep neural networks have magnificent newlinelearning capabilities thereby helps to achieve outstanding results in nature newlinelanguage processing. In the traditional text processing methods, the newlinesemantic retrieval and correlation between the words is not handled newlineefficiently. Since the data grows at rapid rate, not all words are processed newlinein an order. Hence it leads to data sparsity issue and weak feature newlineextraction occurs and then manual feature engineering is required. This newlinereduces the accuracy in Text classification for large datasets. To address newlinethe above challenges, the research proposed an approach for text newlineclassification based on semantic retrieval of input terms using Hybrid word newlineembedding and Deep hybrid models. In this work, the text classification is newlineperformed for movie reviews dataset retrieved from twitter database using newlinevi newlineTwitter API. Semantic Graph creation based on disambiguation property newlineand Word2vec vector representation are the major activities applied to the newlinestructured input tweets before passing into neural network model. In the newlineproposed work, the most similar semantic words are retrieved for each of newlinethe input words and represented in vector.