Enhanced Opinion Mining from Medical Tweets Using an Optimized Penguin Search-Based Feature Selection Algorithm

Opinion mining is the approach of utilizing Natural Language Processing (NLP) concepts to extract the public opinions on specific topics and has gained increasing significance in major text mining applications. Many opinion mining methods have been developed that builds a model to collect and analyse the opinions on topics from the blogs, reviews, comments or tweets. Recently, the application of opinion mining on medical tweets has gained immense research interest due to the challenge of processing unique medical terms in tweets. In this paper, an opinion mining framework has been developed to provide automatic extraction of opinions from medical tweets using improved optimization algorithms. The input tweets undergo pre-processing, and features are extracted by POS tagging and n-grams. Then the feature subset candidates are selected using Penguin Search Optimization algorithm (PeSOA) and Improved PeSOA. In PeSOA, the solution search operation is random and does not utilize exploration concept effectively in order to maintain simplicity. The Improved PeSOA exploits this limitation and introduces a new solution search equation to compliment the traditional search process and an effective feature subset ranking concept. These concepts of Improved PeSOA increase the efficiency of selecting optimal feature subsets. Once the features are selected, the final classification is performed using k-Nearest Neighbor (k-NN), Naïve Bayes (NB) and Support Vector Machine (SVM) classifiers to obtain the opinions. Experiments are conducted on medical datasets containing Cancer and drug tweets. The results prove that the classification accuracy for opinion mining has been increased significantly by the use of Improved PeSOA than the traditional PeSOA.