Software Function Recommender System Based on Collaborative Filtering
Naoki Ohsugi, Akito Monden, Shuuji Morisaki, Kenichi Matsumoto · 2004
†High-Functionality Applications (HFA) includes a large amount of useful yet unbeknown functions. This paper proposes a Software Function Recommender System based on Collaborative Filtering (CF) and evaluates the accuracies of the system’s recommendations. The proposed system recommends useful yet unbeknown functions to each user, as follows: the proposed system automatically collects histories of software function execution from many users, then estimates the usefulness of unused functions using CF algorithms, and recommends users with the functions ranked by their estimated usefulness. We conducted experimental evaluation of six CF algorithms including 2 new algorithms (Rank correlation algorithm and Sequence-based algorithm) by using NDPM (Normalized Distance-based Performance Measure). The result of the experiments showed that the average NDPM of all the algorithms were better than that of randomly produced recommendation. This suggested the proposed system can be used for discovering useful yet unbeknown functions of HFA.