Detection of Compatibility, Proximity and Expectancy of Bengali Sentences using Long Short Term Memory

Text classification is known to be a supervised machine learning technique used in one or more predefined categories to classify sentences or text archives. To be a perfect phrase to convey one's feelings or to be significant, a Bengali sentence must have three properties i.e. Compatibility, Proximity and Expectancy. In this paper, we have proposed a method that is able to detect whether a Bengali sentence has compatibility, proximity and expectancy using Long Short Term Memory network. Our model is trained with word embedding layer and LSTM layer for the detection of Compatibility of a sentence but POS tagging is included for assuring the syntactic structure in case of Proximity and Expectancy detection. The model is tested on around 75000 Bengali simple sentences. The proposed framework achieves an accuracy of 97.5 percent, 85.5 percent and 97 percent for Compatibility, Proximity and Expectancy respectively. The result analysis proves that our model gives better performance.