In natural language processing, most prominent branch is sentiment analysis. People’s emotions and attitudes are analyzed using this sentiment analysis towards service, some product, etc. In prediction of the future scope of a product, some benefits are given by sentiment analysis. However, manual analysis of such a huge amount of documents is a highly tedious task, especially with limited time. Hence, for solving this problem, various attempts are made in literature and proposed different sentiment analysis methods. However, in generation of lexicon, popular NLP tools has some drawbacks. The accuracy of lexicons based on humans is less and they are limited too. On the other hand, lexicons based on dictionary are highly general and they are domain specific. So, a technique called Corpus Integrated Autoencoder Convolutional Neural Network based Sentiment Analysis (CI-AECNN) is proposed in this work for solving this issue. The sentiment lexicon generation based on corpus is performed in this work. Candidates sentiment orientation are computed using this and seed lexicon are added with recognized sentiment words and from seed lexicon, words with incorrect sentiment are removed. The long short-term memory (LSTM) is used for performing Word Sense Disambiguation. Conditional random fields are used for extracting features. At last, auto-encoder, convolutional neural network is used for performing classification. In MATLAB simulation environment, conducted this research work’s overall analysis and it indicates that better results are produced by proposed technique when compared with available techniques.