Creative Commons

This dissertation is the final part required for the completion of the MSc in Data Science at the International Hellenic University (IHU). The main goal of this study is to develop a book recommendation system based on collaborative filtering. Furthermore, a comparison is held between the performance of the system when the recommendations are based on simple ratings and when they are based on the result of sentiment analysis over customer comments. The algorithms examined are based on the KNN methodology and Singular Value Decomposition (SVD). The technique used to conclude in the algorithm with the best performance and generalization was cross-validation and the metrics that were examined were precision and recall. Our findings show that the utilization of user comments as the input in our system instead of ratings, results in an increase in precision and a decrease in recall. For a book recommendation system, precision can be a more important indicator as it refers to the number of relevant recommendations which, eventually is the purpose of such a system. In particular the best performance was achieved with a 10-fold cross validation for the top 10 suggested items and resulted in an increase of 10% in the prediction score. As mentioned already the was a simultaneous decrease in the recall score which also influenced the F-score negatively and in particular it led to a decrease of 0.06%.