Abstract
Background:Sentiment analysis means predicting whether text shows negative or positive opinion in general or over specific entities like person, product or policy which has acquired lots of attention in current year. It plays prominent role in NLP (Natural Language Processing). In past year, people generate manual reviews for everything which consist of large quantity of unstructured data which is tedious to acquire useful data from it. Because of challenges in sequence length, textual order and complex logic, till now it is remained as challenge to detect correct sentiment polarities of users review, mainly for proper classification and prediction of sentiments.
Objective: To perform sentiment analysis with hard dataset which is tedious process to make classification is achieved with help of proposed Lexicon-NCCA-ALSTM, to convert normal word form to numerical form is achieved with support of lexicon and NCCA, to classify sentiments into positive or negative polarity, ALSTM model is used
Methodology: Here in this paper, to enhance polarity classification lexicon is utilized. So in our proposed work lexicon and novel Canonical Correlation Analysis (NCCA) are integrated with Attention-Long Short Term Memory (ALSTM) technique. Here proposed lexicon and NCCA is used to convert normal word text to numerical form which is understandable by machine language. Then proposed ALSTM is designed in a way to categories the sentiments as positive or negative based on the numerical value provided by NCCA.
Result:Experiments was conducted on three kinds of hard dataset to evaluate the performance of proposed work. 1st two dataset are taken from SemEval-2014 Task 4 which consist of data related to laptop and restaurant domain. Whereas 3rd dataset is tweets which are collected from twitter.
Conclusion:Our proposed result illustrates that for laptop dataset it outperforms all other existing techniques for performance metrics accuracy and Macro F1 Score. Whereas for restaurant and twitter dataset it gives significance results compared to other techniques.

Key words: Natural Programming Language, Attention Long Short Term Memory, Lexicon creation, Canonical Correlation Analysis, SemEval, Twitter.