Categorised Neighborhood-based Collaborative Filtering

Bell and Koren produced an efficient and swift Collaborative Filtering algorithm, which improved the original Netflix algorithm by addressing some of the main concerns which were present. We copied the majority of this algorithm and included an extension. The question is whether this extension will provide an improvement to the existing algorithm. The existing model makes use of the so-called jointly derived interpolation algorithm, we left out factorisation which was used in Bell and Koren (2007) and included a categorisation of the ratings. In short this means that user ratings labeled as ’high’, ’middle’ or ’low’, will only be compared to similar ’high’, ’middle’ or ’low’ user ratings. This is in contrast to the original where they are compared to all ratings. The results will be compared to our own reproduction of the existing algorithm by making use of the Root Mean Squared Error term. Lastly, we will discuss the results and shortcomings and make suggestions for future research.