A new sentiment classification method based on hybrid classification in Twitter

Social media including Twitter create a space for expression and dissemination of thoughts and opinions on various topics and various events and they have created opportunity to apply theories and technology leading to search and explore the trends. Mining stream data needs to be balanced in three different branches: accuracy, time and memory. The optimal accuracy rate has obtained in the stream data using stochastic gradient descent algorithm. In this paper, we show that replacing the stochastic gradient descent algorithm in tree leaves causes improvement and reliability of the forecast. The proposed algorithm, Hoeffding stochastic gradient descent, has not changed the time while increasing the accuracy. Studies on the tweets that have been randomly selected shows that the proposed algorithm outperforms when assessing the sentiment stream data.