Sentiment analysis is a process of identification of opinion and thought related to any product, people or an organization. The sentiment analysis is mainly done to understand the others opinion related to some entity. This concept is mainly used in the large organizations and E-commerce to track the user’s activity and their response related to the product. The reviews of the product help the other users to know about the product more. If the reviews of other users are positive its sale is enhanced and if reviews are negative, then it affects the product sale. Sentiment analysis is done on the basis of text and images posted by the users on the social media website. In this analysis, sentiments are classified into positive, negative and neutral. sentiment analysis can be characterized as a procedure that helps in mining of feelings, emotions, views, and opinions from content, tweets, database, and speech in an automatic way by mean of NLP i.e. Natural Language Processing. SA examination includes the classification of opinions in content into classifications like "positive" or "neutral" or "negative". It's likewise indicated as opinion-based mining, subjectivity examination, and the extraction based on judgement. This polarity is assigned according to the meaning of words and after these score of all words is combining to understand the total score and then decides the comment is positive or negative. Sentiment analysis is a challenging task because it is not easy to analyses the exact views, opinions, and feeling from the text. The way of writing the feelings are different for every people in different context and topics. This issue solved by combining the text and prior knowledge. This research work proposes the deep convolutional neural network that uses character- to sentence-level information to perform sentiment analysis of tweets. This model presented a new approach for the initialization of the weights of convolutional neural network which helps to train the network effectively and helps to add new features. The model train by using unsupervised neural language and further tuned by deep learning model on a distant supervised corpus.