Automated methodology comprised of supervised techniques to assist product selection

Customer targeted markets are inundated with similar products from multiple vendors and selecting a product of choice is a challenging task. All varieties of a product have pros and cons in plenty, and the task of identifying a suitable product is daunting and cumbersome. To address this, we propose a methodology to identify preferred products in an automated manner. The end result is achieved by analyzing the history of multiple factors involved with the product of study and utilizing a supervised learning algorithm to predict the worthiness of the product with respect to the user. This algorithm is designed by combining and customizing sentiment analysis and automatic ontology construction algorithms. Dependency parsing for ontology construction, HMM/CRF for decision making, and a new personalized algorithm for sentiment analysis were utilized to customize the prediction method. For a product under consideration, the algorithm takes into account all the user specified features and predicts an outcome of it being good (positive) or bad (negative) to the interested user. This outcome is achieved by analyzing the past history of the features specified by the user. Using this algorithm we studied a set of 20 movies released during the period of January - March 2013 and achieved 70% accuracy in predicting their box office outcome. Our results indicate that there is a correlation between the selected features past performance and the overall success of a new product with the same features. Given a wide array of available choices, this algorithm can predict an ideal product for a customer.