Shinsuke Nakajima

Context-Aware SVM for Context-Dependent Information Recommendation

The purpose of this study is to propose Context-Aware Support Vector Machine (C-SVM) for application in a context-dependent recommendation system. It is important to consider users’ contexts in information recommendation as users’ preference change with context. However, currently there are few methods which take into account users’ contexts (e.g. time, place, the situation and so on). Thus, we extend the functionality of a Support Vector Machines (SVM), a popular classifier method used between two classes, by adding axes of context to the feature space in order to consider the users’ context. We then applied the Context-Aware SVM (C-SVM) and the Collaborative Filtering System with Context-Aware SVM (C-SVM-CF) to a recommendation system for restaurants and then examined the effectiveness of each approach.

Reasonable tag-based collaborative filtering for social tagging systems

In this paper, we present a tag-based collaborative filtering recommendation method for use with recently popular online social tagging systems. Combining the information provided by tagging systems with the effective recommendation abilities given by collaborative filtering, we provide a website recommendation system which provides relevant, credible recommendations that match the users changing interests as well as the users bookmarking profile. Based upon user testing, our system provides a higher level of relevant recommendations over other commonly used search and recommendation methods. We describe this system as well as the relevant user testing results and its implication towards use in online social tagging systems.

A ranking method based on users' contexts for information recommendation

We propose a ranking method using a Support Vector Machine for information recommendation. By using the SVM, a recommendation method can determine suitable items for a user from enormous item sets. However, it can decide based on just two classes: whether the user likes a thing or not. When there is a large number of recommended items, it is not easy for the user to find the best item by herself. To resolve this issue, it is desirable to rank the items based on the users preferences. Moreover, the users preferences change depending on the context. Based on the above problem, we propose a context-aware ranking method for information recommendation. Our method considers a users context when ranking items. Our method consists of the following two steps: (1) Predicting important feature parameters for the user. (2) Calculating a ranking score of each item in recommendation candidates. In this paper, we describe our method and show experimental results.

Investigation of the Effectiveness of Tag-Based Contextual Collaborative Filtering in Website Recommendation

As the Internet continues to mature and become more accessible to the commonuser, the amount of available information increases exponentially. Accordingly, find-ing useful and relevant information is becoming progressively difficult. Moreover,a lot of the information available—blogs, various types of reviews, and so forth—is highly subjective and thus, hard to evaluate purely through machine algorithms.Being subjective in nature, one person may absolutely love something while the nextmay loathe the same—no single authority exists. It is in these cases where people—more so than the current ability of machine algorithms—are greatly effective inevaluating and filtering this information.For this reason, the idea of collaborative filtering (CF) was started, extensivelyresearched, and eventually deployed to relatively good amounts of success. Usingthe people and the community, recommendations of subjective information can bemade through the matching of similar users. Sites such as amazon.com [1] or movie-lens [6], etc. utilize such recommendation methods, matching users based upon theirratings and then producing recommendations. Through this, CF provides personal-ized recommendations to the users, while at the same time offering the ability to dealwith subjective material. However, the failing of CF is that it does not consider whya user likes something and what the user is interested in now. In other words, CF canrecommend relevant sites, but does not know why or when it should be appropriate.Similarly, online social tagging systems also employ the masses to evaluate anddescribe information. Instead of relying purely upon machine algorithms, people

Identifying agitators as important blogger based on analyzing blog threads

A blog (weblog) lets people promptly publish content (such as comments) relating to other blogs through hyperlinks. This type of web content can be considered as a conversation rather than a collection of archived documents. To capture ‘hot’ conversation topics from blogs and deliver them to users in a timely manner, we propose a method of discovering bloggers who take important roles in conversations. We characterize bloggers based on their roles in previous blog threads (a set of blog entries comprises a conversation). We provide a definition of agitators as bloggers’ roles who have a great influence on bloggers’ discussion. We consider that these bloggers are likely to be useful in identifying hot conversations. In this paper, we discuss models of blogs and blog thread data, methods of extracting blog threads, discovering important bloggers.

Route recommendation method for car navigation system based on estimation of driver's intent

Nowadays, car navigation systems are widely used in cars to aid drivers by providing directions to a destination. However, these systems do not always recommend a route that perfectly matches the drivers intent. Even when drivers intentionally select a route that is different from the recommended one, the system leads them back to the original route. Such recommendations do not adequately reflect the drivers intent. This study proposes a route recommendation method for a car navigation system that estimates the drivers intent and rerecommends a route that matches this intent when the driver deviates from the originally recommended route. We developed a simulator based on the proposed method and used it to experimentally verify the effectiveness of the proposed method.

Relative Queries and the Relative Cluster-Mapping Method

Most conventional database systems and information retrieval systems force users to specify the qualification conditions in queries in an ”absolute” manner. That is, a user must specify a qualification condition to be satisfied by each retrieval result. To do this properly, users must have sufficient knowledge about the metadata and their structures. In the real world, however, people often specify a qualification in a relative manner such as ”I prefer this one among these.” In this paper, we propose the notion of ”relative queries,” and their query processing method called the ”relative cluster-mapping.” A relative query specifies user’s selection criteria as selecting his/her favorite data among a sample data cluster. This notion is useful when users do not have sufficient knowledge about metadata, or users cannot decide a complete qualification condition. The ”relative cluster-mapping” method maps the relative position of the user-selected data in a sample data cluster to a target data cluster and returns an answer from the target data cluster. The answer‘s relative position in the target data cluster is similar to that of the selected data in a sample data cluster. We believe it is more natural to express the desired data using relative qualifications. We also have developed prototype system, and evaluated its appropriateness.

Amplifying the differences between your positive samples and neighbors in image retrieval

A novel method for retrieving images based on relevance feedback and clustering has been developed. That is, by clustering sets of retrieved data, a user can select some good answers from them by considering the difference between the feature data of the selected images and the feature data of images placed in their neighborhood. This difference information improves previous queries since the user must have found some important difference between their-selected image and similar neighboring images. An image-retrieval system based on a relevance feedback by difference amplification is set up and shown to be more effective than conventional methods.

Cooking Recipe Recommendation Method Focusing on the Relationship Between User Preference and Ingredient Quantity

There are numerous websites on the Internet that recommend cooking recipes. However, these websites order recipes according to date of submission, access frequency, or user ratings for recipes, and therefore, they do not reflect a user’s personal preferences. In this paper, we propose a recipe recommendation method based on the user’s culinary preferences. We employ the user’s recipe browsing and cooking history in order to determine his/her preferences in food. In our previous study on the subject, we considered only the presence of certain ingredients in cooking recipes in order to determine user preferences. However, in order to ascertain them more accurately, we propose a scoring method for cooking recipes based on users’ food preferences and the quantity of the ingredients used.

Evaluating credibility of web information

We describe a new concept and method for evaluating the Web information credibility. The quality control of information (text, image, video etc.) on the Web is generally insufficient due to low publishing barriers. As a result, there is a large amount of mistaken and unreliable information on the Web that can have detrimental effects on users. This calls for technology that facilitates the judging of the credibility (expertise and trustworthiness) of Web content and the accuracy of the information that users encounter on the Web. Such technology should be able to handle a wide range of tasks: extracting several credibility-related features from the target Web content, extracting reputation-related information for the target Web content, such as hyperlinks and social bookmarks and evaluating its distribution, and evaluating features of the target content authors. We propose and describe methodologies of analyzing information credibility of Web information: (1) content analysis, (2) social support analysis and (3) author analysis. We overview our recent research activities on Web information credibility evaluation based on this methodologies.