Compromise between quality and effort of recommender systems using the neighbor filtering approach based on reliable neighbors

Collaborative filtering approach recommends new items to users by aggregating opinions of others who present highly similar preferences. Usually similarity computation techniques in collaborative filtering involved with analyzing common aspects between users. Limitation of collaborative filtering techniques is that lacking of reliability can be found in the output when a very small number of common aspects are considered in similarity computation. This conduces to disrepute of opinions and may not only affect to recommendation quality, but the system might also waste time to process poor quality opinions. To alleviate this limitation, we proposed the techniques how to locally estimate reliability of opinions or neighbors based on the number of common aspects presented by users. Next we proposed the Neighbor Filtering Approach to reduce the number of low quality opinions or neighbors who are not likely to be creditable in recommendation process. We also suggested new process structure which incorporates the Neighbor Filtering Approach into the classical collaborative filtering recommender system. Finally we conduct the collaborative filtering system according to the new process structure and evaluate the results by comparing them with the baseline from the classical system. The results show that the effort in recommendation process can be reduced without compromising its recommendation quality.