One-to-one Marketing


One-to-one Marketing (Recommender Systems)

VisProfile: Visualizing User Profiles for Eliciting Users’ Discoveries
IEEE/WIC/ACM International Conference on Web Intelligence (IEEE/WIC/ACM WI’15)
Researchers: Kazunori Okubo

VisProfileS When users understand their preferences and interests, they may find it easier to make a decision to buy an item new to them. We developed a recommender system with a visualization function of the user profile to elicit the user’s discoveries for his/her preferences or interests. Because user profiles are usually implicitly created by machine learning techniques based on the users’ usual activities such as browsing and shopping, they might include user preferences or interests of which the users are not explicitly aware. We conduct a user experiment to know whether users can aware new knowledge about their interests or preferences when they use our system. [PDF]

Individualizing User Profiles from TV Set Viewing Logs
ACM International Conference on Ubiquitous Information Management and Communication (ACM IMCOM’15)
Researchers: Koichi Iguchi

iguchi_res1sA recommendation function is not popular in a TV set (recommending TV programs) whereas it is widely used in the Web-based services. The most crucial reason is that it is difficult to identify an individual sitting in front of the TV, which are usually shared by several members of the family. In this study, we propose a technique to individualize a user profile from the viewing logs mixed with several users. This technique identifies time intervals when the TV is often turned on. We assume that there exists a dominant user to a specific time. We create a user profile in each identified active time interval. [PDF]

Study of the Relationship among User Intervention and User Satisfaction for Recommender Systems
ACM Symposium on Applied Computing: TRECK track (ACM SAC’12)
Researchers: Yuki Kai

kai_res2sUser satisfaction in recommender systems is considered to be influenced by many factors. Among these factors, we focus on user intervention. User intervention is a user control over a recommender system. We provide three hypotheses: i) user intervention in the recommendation process itself improves the user satisfaction, ii) user intervention improves the user satisfaction when the intervention is reflected in the recommendation results, and iii) the degree of improvement in user satisfaction differs among the types of user interventions applied. In this study, we conducted a user experiment using several kinds of interventions, and clarify the relationship between user intervention and user satisfaction. [PDF]

Discovery-oriented Collaborative Filtering: Matching the trade off between Novelty and Accuracy
ACM International Conference on Intelligent User Interfaces (ACM IUI’09)
Researchers: Takuya Shimizu

shimizu_res2sThe main goal of traditional collaborative filtering techniques is improvement of the accuracy of recommendation. Nevertheless, such techniques present the problem that they include many items that the user already knows. In our work, we use not only ratings of preference but those of acquaintance. Ratings of acquaintance indicates what item a user knows. We infer items that a user does not know by calculating the similarity of users or items based on ratings of acquaintance. We also infers items that a user likes in the same way. Finally, we combine those relevance score by considering the trade off between accuracy and novelty. [PDF]

C-baseMR: Content-based Music Filtering System with Editable User Profile
ACM Symposium on Applied Computing: IAR Track (ACM SAC’06)
Researchers: Kazuhiro Iwahama

kai_res1sC-baseMR is a music recommendation system. This is implemented using content-based filtering technique. We extracted musical performance features like tempo, move between notes, major chord, and minor chord from MIDI data. Decision tree is used as machine learning algorithm. This outputs the user profile representing the user’s preference of music graphically. Users can edit their user profile when they find some learning errors in the profile. The correction increases the recommendation accuracy for some users. [PDF]