Sentiment analysis system using conventional neural network in social media

Sentiment analysis systems on social media platforms such as Twitter has become an extremely significant and difficult topic. In the proposed system, the convolution neural network CNN has been employed to classify operations, CNN is one of the fast, accurate, reliable, and efficient networks among other classification networks. This paper used one dataset which are: the Stanford Twitter Sentiment Test “STSTd”. The proposed SACNN model consists of main steps such as: “preprocessing, feature extraction, and classification” steps. There are four steps in preprocessing operation which is: Tokenization, Stopword Removal, Stemming, and Transformation. Following preprocessing, characteristics from a text document are extracted through the use of frequency and “inverse document frequency (TF-IDF)” to feed forward to the proposed SACNN classifier. The proposed SACNN was created to classify the text as “positive” or “negative”, the structure of the proposed CNN consists of 18 layers which are divided into one input layer, eight convolution layers, seven pooling layers, one Flatten layer and finally Dense Layer. CNN layers have been created with multi parameters; this structure made CNN more efficient in the classification process. The proposed CNN classifier has been achieving a higher accuracy metric rate of 96.53 in the classification process by testing STSTd. while the metrics have been scored as precision is 92.88%, recall is 97.45%, and F1-score is 98.88%.