Implicit Feedback Embeddings for Recommender Systems on Sparse Data（推薦システムのための疎データの効果的埋め込み表現に関する研究）
Thai Binh Nguyen · Institutional Repositories DataBase (IRDB)
In information recommendation, the preferences of users and the attributes of items to be recommended are represented by feature vectors.The items are recommended based on the similarity of the corresponding feature vectors.Characteristics of users and items are learned from the rating for items of users, etc.In general, the amount of ratings is limited, and extracting effective features from sparse data will be necessary.Also, for users who newly participate in the system, it is difficult to obtain user characteristics because the information is limited.The aim of this thesis is to propose effective models of recommender systems for such sparse data.We first look into the rating prediction problem, one of the essential tasks in recommender systems.Rating is another kind of feedback known as explicit feedback.Different from the implicit feedback, the amount of rating data is limited because it requires users to provide the ratings explicitly.I first proposed a feature extraction method that utilizes the data that is easy to obtain, such as the click history recorded in the log when a user examines an item.In this method, the features of items are extracted based on two sources of feedback: rating data and click data.We show that