This item is available under a Creative Commons License for non-commercial use only
The diversity; versus accuracy trade off, has become an important area of research within recommender systems as online retailers attempt to better serve their customers and gain a competitive advantage through an improved customer experience. This dissertation attempted to evaluate the use of diversity measures in predictive models as a means of improving predicted ratings. Research literature outlines a number of influencing factors such as personality, taste, mood and social networks in addition to approaches to the diversity challenge post recommendation. A number of models were applied included DecisionStump, Linear Regression, J48 Decision Tree and Naive Bayes. Various evaluation metrics such as precision, recall, ROC area, mean squared error and correlation coefficient were used to evaluate the model types. The results were below a benchmark selected during the literature review. The experiment did not demonstrate that diversity measures as inputs improve the accuracy of predicted ratings. However, the evaluation results for the model without diversity measures were low also and comparable to those with diversity indicating that further research in this area may be worthwhile. While the experiment conducted did not clearly demonstrate that the inclusion of diversity measures as inputs improve the accuracy of predicted ratings, approaches to data extraction, pre-processing, and model selection could inform further research. Areas of further research identified within this paper may also add value for those interested in this topic.
Browne, G. An evaluation of the use of diversity to improve the accuracy of predicted ratings in recommender systems Dissertation submitted in partial fulfilment of the requirements of Dublin Institute of Technology for the degree of M.Sc. in Computing (Data Analytics). 2015.