Biclustering Algorithms And Its Applications In Collaborative Filtering

With the rapid boosting of recommender system and the further study of gene expression data, how to find the relationship between attributes in cluster is becoming increasingly important. Biclustering algorithm was put forward to deepen probe on this issue. As a new data mining method, it had become a hotspot in the data mining research field at the beginning of the 21st century. Due to the lower computing speed and inaccurate results, this paper presents a fuzzy biclustering algorithm to improve the performance of biclustering algorithms base on mean squared residue clustering. This algorithm aims at solving a single cluster base on fuzzy theory. When the thought of biclustering is introduced to recommender system, this paper proposes three improved collaborative filtering algorithms for the sparsity problem and accuracy of prediction problem. The main work is described as following:1. Biclustering algorithms based on mean squared residue mostly use a greedy strategy, which could not obtain accurate clusters with appropriate size. However, fuzzy theory could improve the performance of coclustering algorithms based on mean squared residue clustering and obtain more accurate clusters with appropriate size. Defined the fuzzy variables in cluster, which is also called the significant indexes to improve the performance of biclustering algorithms based on mean squared residue clustering.. And then provided an algorithm and convergence analysis by setting up a model of fuzzy biclustering.2. Considering the advantages of mean squared residue and the problem of applying the algorithm to sparse data, two new collaborative filtering algorithms in view of simple mean squared residue (SMSR_CF) and co-clustering base on simple mean squared residue (SMSR_CoCF) are presented. The former improves the accuracy of prediction on basis of maintaining the high efficiency of calculation. The latter provides the information of co-clusters in sparse matrix as well as the improvement of dynamic problem in recommender system.3. Defined blocks which are similar to neighborhood and presented two novel improved memory-based collaborative filtering algorithms:collaborative filtering algorithm base on blocks (Block-based CF) and collaborative filtering fuzzy algorithm base on blocks (Block-based Fuzzy CF). The first algorithm not only reduces the running time but increases the accuracy of prediction. The second algorithm introduces the index about accuracy rating of predicting block to heighten the precision of prediction further. 4. Most collaborative filtering algorithms being in existence merely tell us how to gain the accurate predicting ratings without introducing where to obtain them. In order to acquire a subset of users and items which can gain a high accuracy of prediction, this paper presents a collaborative filtering algorithm (CoreCF) base on fuzzy theory.