Comparative Study on Influences of False Negative Ratings in Collaborative Filtering with Fuzzy Co-Clustering

Collaborative filtering-based personalized recommendation can be achieved by two main approaches of memory-based algorithms and model-based algorithms. Many real-world applications are realized with 0-1 type relational data such as purchase history transactions, where rating 1 (one) means a positive preference on an item while rating 0 (zero) can be a negative feeling or unknown including both ‘negative feelings’ and ‘positive preferences but not bought yet’. This paper presents a comparative study on the influences of false negative ratings in the memory-based GroupLens algorithm and the fuzzy co-clustering-induced model-based algorithm. Especially, by investigating the impact of unbalanced false positives among subsets of users and items, this research tries to explore operational techniques to improve performance in real world applications.