Data Mining / Recommendation / Artificial Intelligence / Machine Learning / Optimization

Given vast amount of data, recently, there is a trend to retrieve meaningful information from it. However, no method gives us guarantee to obtain good insight only using such data. Actually, we need additional knowledge of the data at the certain level in advance. For example, a famous method in the domain of recommender system, called Collaborative filtering (CF), uses a prerequisite that users has preferences similar to each others' tend to prefer the same items. This prerequisite enables CF predicts user ratings for items that is still unknown because the user have never reviewed the items, only using the rating history of users in the system. We can categorize those models that uses such this prerequisites to a mathematical model. In this category, My study focuses especially on a probability model with expecting that this model will be able to write most of the known heuristic methods consistently and generally. In this point of view, therefore, I expect that we can establish more suitable model for the real data than another.