Kenta Oku

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.

Mapping Geotagged Tweets to Tourist Spots for Recommender Systems

We are developing a recommender system for tourist spots. The challenge is mainly to characterize tourist spots whose features change dynamically with trends, events, season, and time of day. Our method uses a one-class support vector machine (OC-SVM) to detect the regions of substantial activity near target spots on the basis of tweets and photographs that have been explicitly geotagged. A tweet is regarded as explicitly geotagged if the text includes the name of a target spot. A photograph is regarded as explicitly geotagged if the title includes the name of a target spot. To characterize the tourist spots, we focus on geotagged tweets, which are rapidly increasing on the Web. The method takes unknown geotagged tweets originating in activity regions and maps these to target spots. In addition, the method extracts features of the tourist spots on the basis of the mapped tweets. Finally, we demonstrate the effectiveness of our method through qualitative analyses using real datasets on the Kyoto area.

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.

Tweet-mapping Method for Tourist Spots Based on Now-Tweets and Spot-photos☆

Abstract Tourism recommender systems suggest suitable tourist spots by matching the characteristics of the tourist spots with those of the user. In this paper, we focus on an essential source of these characteristics—geotagged tweets. To solve the problem of associating geotagged tweets to tourist spots, we propose a mapping method that infers the region of a target spot on the basis of two geotagged items. The first is a geotagged tweet, which demonstrates that the tweeter was indeed at the target spot at the time the tweet was posted. We call this a “now-tweet.” The second item is a geotagged photo of the target spot, which we call a “spot-photo.” We regard these now-tweets and spot-photos as training data, and then determine the region of the tourist spot by inferring the geographical distribution of the training data. Next, we map geotagged tweets from the extracted region to the target spot. To improve the accuracy with which the tourist spot is inferred, we apply a clustering algorithm to the training data. Experimental results indicate that photo-based mapping with sophisticated training data produces the most improved performance over baseline methods. When applied to 4,559,643 geotagged tweets, our method maps them to tourist spots with an average granularity of 144.85 m.

Geographical recommender system based on interaction between map operation and category selection

We propose a geographical information recommender system based on interaction between users map operation and category selection. The system has three interfaces, the layered category interface, the geographical object interface and the digital map interface. Our system interactively updates each interface based on the category interest model and the region interest model. This paper describes each interface and each model, and how to update them by our system.

Searching human actions based on a multi-dimensional time series similarity calculation method

With the rapid performance improvement and popularization of sensor devices, a large amount of human action data can be captured in databases. Classification, recognition, searching, and mining of such human actions are promising applications. Although many of these applications have been developed, searching the large quantity of data, especially given the high dimensionality of the captured temporal data sequence is time-consuming. To reduce this time cost, we use a novel method for approximating a multi-dimensional time-series, named multi-dimensional time-series Approximation with use of Local features at Thinned-out Keypoints (A-LTK). With A-LTK applications for two human motion types, sign language and dancing, we found that the categorization of human action data and the search for the most similar human action became more accurate and reduced the time cost.

Mapping Geotagged Tweets to Tourist Spots Considering Activity Region of Spot

We are developing a recommender system for tourist spots. The challenge is mainly to characterize tourist spots whose features change dynamically with trends, events, season, and time of day. Our method uses a one-class support vector machine (OC-SVM) to detect the regions of substantial activity near target spots on the basis of tweets and photographs that have been explicitly geotagged. A tweet is regarded as explicitly geotagged if the text includes the name of a target spot. A photograph is regarded as explicitly geotagged if the title includes the name of a target spot. To characterize the tourist spots, we focus on geotagged tweets, which are rapidly increasing on the Web. The method takes unknown geotagged tweets originating in activity regions and maps these to target spots. In addition, the method extracts features of the tourist spots on the basis of the mapped tweets. Finally, we demonstrate the effectiveness of our method through qualitative analyses using real datasets on the Kyoto area.

Correlation analysis among the metadata-based similarity, acoustic-based distance, and serendipity of music

With the aim of realizing a serendipity-oriented music recommendation, we analyzed the correlation between music similarity and serendipity. A user may be familiar with the musical piece if its metadata, such as the artists names, and the title, is similar to the music he/she has ever listened to. In addition, a user may prefer the music if it is acoustically similar to the music he/she prefers. Based on these notions, we set up the following hypotheses: Hypothesis I: the user is familiar with the music is if the metadata-based similarity between it and the music he/she prefers is high. Hypothesis II: the music is preferred by the user if the acoustic-based distance between it and the music he/she prefers is low. Hypothesis III: the music is serendipitous (unexpected and useful) if the music has both a low metadata-based similarity and low acoustic-based distance with his/her preferred music. This paper presents our examination of the above hypotheses using data from 1,000 real musical recording.

A Fashion-Brand Recommender System Using Brand Association Rules and Features

Web services selling fashion clothes on Internet are rapidly increasing, so it is becoming difficult for users to find their favorite ones among the enormous number of fashion items available. Although several fashion brand recommender services are available to support the users to search clothes to be bought, the accuracy is so low that they need to check clothes one by one. In this paper, we propose a fashion-brand recommendation method based on both the fashion features and the fashion association rules. The fashion-brand association rules are used to select new brands for a user which are similar to users favorite ones. As the rules represent the frequent occurrences in fashion-brand liking, while the fashion-brand feature can be used to calculate similarities between brands. We also propose a new method which is a combination of these two. We combined these two methods into one in a serial-hybrid way. It is shown that a combined method produces the highest F-measure among other methods including existing services.

A coarse-to-fine method for subsequence matching of human behavior using multi-dimensional time-series approximation

In this paper, a novel method for subsequence matching of human behaviors is proposed. Since the human behavior data taken from many sensors monitoring human motions is a multi-dimensional time series, we extend an existing time-series approximation method, A-LTK (Approximation with use of Local features at Thinned-out Keypoints), to improve its performance as well as its accuracy in subsequence matching. Since A-LTK can change its approximation level using a parameter, the approach introduced in this paper uses two types of A-LTK levels, coarse followed by fine. The A-LTK-based Coarse-to-Fine subsequence matching method, called A-LTK 2.0, is discussed. We also evaluate the method, by comparing it with existing matching methods, DTW, AMSS, and the original A-LTK. The evaluations showed that A-LTK 2.0 is superior to the others in subsequence matching for long human-behavior sequences.