A Hybrid Machine Learning Approach for Analysis and Classification of Social Network Sentiments

With the evolutionary improvement and establishment of social networking sites huge amount of data associated with user's perceptions, emotions, posts, comments, reactions etc. Meanwhile, extraction of valuable information like user's opinion, sentiment from those massive and higher dimensional dataset have become one of the most crucial, complex and convoluted tasks nowadays. Moreover, immediate realization of user's perspective regarding any topic or product can ensure feasible as well as user friendly environment throughout the social networking sites. In this respect, from current point of view available natural language processing approaches are less applicable because of their limitations. Therefore, effective machine learning approaches long with the concept of NLP need to be invented to negotiate with this sort of situation. Realizing worth of this situation and for negotiating with the concept of big data, we have proposed a machine learning approach by combining unsupervised machine learning approach (Self-Organizing Map) for clustering as well as dimensionality reduction and classification approach (Adam Deep Learning) together. Moreover, for better outcome we have adopted MaxAbsScaler initially. For further clarification we have also considered principal component analysis based deep learning, Logistic Regression, Normalization less SOM based CNN mechanism for comparative study. To ensure effective performance and better feasibility in case of big data we have also employed our approach on different sizes of social network datasets. Overall analysis and demonstration interprets superiority of our proposed approach with accuracy of 88.83%.