Long Short Term Memory (LSTM) based Deep Learning for Sentiment Analysis of English and Spanish Data
Baidya Nath Saha, Apurbalal Senapati
Abstract
In recent years, deep neural networks have acquired a super power in the the field of machine learning, mainly for unstructured data types such as text and image, the demands of which are escalating day by day. This paper presents a Long Term Short Memory (LSTM) based Recurrent Neural Network (RNN), a popular deep learning algorithm for sentiment analysis of English and Spanish data. This domain is a complicated, because people's opinions are very subjective in nature and depends on many unpredictable factors. We tackled two primary bottlenecks of deep learning algorithms: optimal values of the parameters were chosen through cross validation and optimal architecture were selected through dropout based regularization method. Experimental results obtain state-of-the-art performance and it also offers very attractive insights.