Discovery-oriented Collaborative Filtering for Improving User Satisfaction

Written by: Yoshinori Hijikata, Takuya Shimizu, Shogo Nishida

The authors created a discovery oriented content filtering system so that the result is not information that the user already knows. It was found in the past that people stop using content filtering because it narrows the searches to very similar results and doesn't allow for much discovery. The author's research focused on prediction of unknown items, recommendation of items from the user’s preference and acquaintance, and examination of user satisfaction.

The two main content filtering algorithms are user-based and item-based. The authors used the prediction-combining algorithm and Independently Evaluating Algorithm (IEA) of different types to try and get a better response by joining the common results and filtering out what the user had probably seen before. The authors conducted an experiment using 20000 rating data collected from 100 users. They found that the IEA gave the best results and that combining their predictive algorithms with standard content filtering the quality and uniqueness of results goes up.

This is important because much of what people do on the internet is search for answers, hence google became popular. The problem with this research is that content filtering has never had a wide enough scope for an experienced searcher and this research didn't show very much difference between old CF results and their new one. I would think that you would have an algorithm that lets you vote things down as irrelevant and then sort out what content isnt good after it gets a good idea of what your looking for.