Building a case - based recommender system

Abstract

The amazing successful of recommender system in retrieving information that is relevant to user’s needs and preferences is proven through the successful of E-commerce setting, which supplying a well-defined knowledge domain of products and helping save time for online-shopping as well as making decision easily and conveniently. The revenue getting from this utility is a huge number and enriching for big companies such as Amazon, Netflix. Case-based recommendation is a form of recommender systems and gives a well-defined set of features (e.g. food, color, price, etc.). These thorough presentations contribute for motivation the development quality of recommender systems as well as the assessment of product similarities. In this thesis, building a simple case-based recommender system is based on restaurant recommender system applications. Same experiments are carried out in training set and testing sets. Training set includes two files, which are the features of restaurants file and the features definition file. The testing sets are collected from session files consist of all user’s sessions which are portioned into quarters of year and stretching from 1996 to 1999. In training file, the similarity of restaurants is represented by Cosine Similarity. However, finding the K groups of restaurants is first and foremost job. Association Rule mining algorithm takes over as the role of finding the rules and then clustering restaurants. APRIORI algorithm is used for calculating frequent items or frequent features having support that is greater or equal the minsupthreshold. At that time, K-NN got the group for specified feature then calculating viii the distance and given the predicted similar next restaurants’ s lists in a session. Session files are used to evaluate the prediction list of restaurants found in session. The experimental results show that the proposed method outperforms applying K-NN solely.