Clustering Tree based Implementation of Record Linkage on Many-to-Many Relation

Record linkage or entity resolution are emerging strategy to avoid duplication and other purposes. Recommender domain uses the linkage method to provide efficient results in terms of accuracy. This paper introduces a new Many-to-Many Record Linkage (MMRL) algorithm which links records from one table with a set of records from another table. MMRL algorithm is based on clustering tree which forms the group on each table separately that to be linked. Hierarchical structure such as tree is suitable to understand and execute the linkage process. Intermediate nodes are having less similarity value than end nodes. Each node of the clustering tree contains a cluster instead of a single classification. Prediction accuracy depends on the end node. Jaccard similarity and metaphone similarity are used as distance measures. Prediction result shows whether the records are matched or not. This result proves the efficiency of MMRL algorithm. A data set from movie recommender domain was evaluated for this paper. This MMRL algorithm gives better performance and results.