A Decision Tree Based Record Linkage for Recommendation Systems

Record linkage merges all the records relating to the same entity from multiple datasets, at the entity level. It is the initial data preparation phase for most of the database projects. Traditionally one to one data linkage is performed among the entities of same type with common unique identifier. The proposed one to many and/or many to many record linkage method is able to link the entities of same or different types with or without availability of common unique identifier. Here a probabilistic record linkage which is based on clustering tree construction that classifies the matching entities by linkage. The tree construction is based on the one of the splitting criterion for the best attribute selection that partitions dataset at each node of the tree. Record Linkage is used in recommender system domain to produce list of recommendations at each leaf of the tree. It is used for matching new user with their product expectations in order to produce list of recommendations. In propose method a decision tree based record linkage is applied to generate book recommendations. This technique is also useful in solving cold start and new user problems. Keywords— Decision tree, Classification, Clustering, Splitting Criterion, Record Linkage, Model Based, Recommendation System.