Cross-Domain Recommendation Model based on Hybrid Approach

Abstract

The demand of recommendation has aroused severely since there are huge number of choices available and the end user desires to extract information in least time and with high accuracy. The traditional recommendation systems generate recommendations in the same domain but now cross domain recommendations are gaining importance. The cross domain recommendations address well the limitations of single domain analysis such as data sparsity and cold start problem. Under this research work cross domain recommendation model is designed based on the study of various supervised classification algorithms. 3 domains are under consideration music, movie and book. Model is capable of generating one to many cross domain recommendations exploiting movie domain knowledge to generate recommendations for books and music. Data is collected through survey and data pre-processing has been performed. Study is carried out over K-Nearest Neighbor, Decision Tree, Gaussian Naïve Bayes and Support Vector Machine classifiers and also over majority voting Ensembling, cross validation and data sampling by applying these classifiers to choose the best classifier to form the base of content-based recommendation. Recommendation model uses a hybrid approach of combination of content-based recommendation, user to user collaborative filtering and personalized recommendation techniques. The model perform Twitter sentiment analysis over the recommended entities generated by the model to help the user in decision making by knowing the positive, negative and neutral polarity percentage based on tweets done by people. The designed model achieved good accuracy on testing.