Tsubasa Takahashi

A Ranking Method for Web Search Using Social Bookmarks

Recently, Social Bookmark, which allows us to register and share our own bookmarks on the web, is attracting attention. Social Bookmark makes it possible to retrieve structured data such as (URL, Username, Timestamp, Set of tags). More importantly, the retrieved data represents user interests. There are two aspects of bookmark usage: data for reuse and data for hot issues. This paper, focusing on timestamps of bookmarks, proposes a way to measure the freshness of a web page. It further proposes a page evaluation method that improves S-BITS, our previously proposed method to evaluate informativeness of web pages using social bookmarks, using the freshness evaluation. Finally, it demonstrates the effectiveness of the proposed method through experiments.

Top-down itemset recoding for releasing private complex data

Complex data, which has single-valued attributes and set-Valued attributes, enables us to associate these attribute values and analyze these relationships. Before releasing such complex data, ensuring anonymity for these data owners should be required. However, existing data anonymization methods are not work well because they only assume that quasi-identifiers are either multidimensional single-valued attributes or one set-value attribute. This paper proposes an anonymization method which integrates recodings for single-valued attributes and set-valued attributes into a whole top-down anonymization. Especially, in order to integrate recoding for set-valued attributes, this paper also proposes top-down itemset recoding which follows top-down manner and does not obfuscate items to ensure k-anonymity. In the experiment part, using real dataset, we clarify characteristics and effectiveness of proposed method. This method does not require the generalization hierarchy for set-valued attributes, thus the anonymized itemsets are not obfuscated and can be analyzed by standard data mining tools.

CMOA: continuous moving object anonymization

This paper proposes a continuous anonymization method for a trajectory stream. In todays mobile environment, positions of moving objects are frequently sensed and collected. For real-time movement pattern analyses of people and automobiles, trajectory streams have attracted a lot of attention. Trajectory streams lead to sensitive locations, such as homes and personal hospitals. Additionally, a set of spatio-temporal data might identify a user from a trajectory stream. Therefore, publishing original trajectory streams may cause critical breaches of privacy. To protect privacy of users, we need a mechanism which makes it difficult to identify users from crowds of trajectory streams. Several techniques for anonymizing trajectories have been proposed. Anonymized trajectories can be published without concerning about privacy issues. However, for the continuous publishing of trajectory streams, existing trajectory anonymization methods are not suitable because they anonymize the overall trajectories at a time. If the existing methods are applied in the continuous publishing, the resolution of anonymized trajectory is hugely degraded or trace-ability is lost. In this paper, we propose an anonymization technique for a trajectory stream. The method continuously anonymizes trajectory streams one by one, and dynamically reforms anonymized trajectory streams to improve the resolution. The experiments showed that our method could keep the resolution at a constant level.

AutoCyclone: Automatic Mining of Cyclic Online Activities with Robust Tensor Factorization

Given a collection of seasonal time-series, how can we find regular (cyclic) patterns and outliers (i.e. rare events)? These two types of patterns are hidden and mixed in the time-varying activities. How can we robustly separate regular patterns and outliers, without requiring any prior information? We present CycloneM, a unifying model to capture both cyclic patterns and outliers, and CycloneFact, a novel algorithm which solves the above problem. We also present an automatic mining framework AutoCyclone, based on CycloneM and CycloneFact. Our method has the following properties; (a) effective: it captures important cyclic features such as trend and seasonality, and distinguishes regular patterns and rare events clearly; (b) robust and accurate: it detects the above features and patterns accurately against outliers; (c) fast: CycloneFact takes linear time in the data size and typically converges in a few iterations; (d) parameter free: our modeling framework frees the user from having to provide parameter values. Extensive experiments on 4 real datasets demonstrate the benefits of the proposed model and algorithm, in that the model can capture latent cyclic patterns, trends and rare events, and the algorithm outperforms the existing state-of-the-art approaches. CycloneFact was up to 5 times more accurate and 20 times faster than top competitors.

Improvement of Pk-Anonymization

Pk-anonymization is a data anonymization method that employs randomization. Pk-anonymization guarantees Pk-anonymity, which is an extension of probabilistic k-anonymity. To implement this method, we assign random noise to records to reduce the probability of identifying record owners to less than 1/k. Existing methods assign noise using a Laplace distribution, and determine the variance of the Laplace distribution at a desired value of k to satisfy Pk-anonymity. In this paper, we propose an algorithm that improves the implementation of Pk-anonymization with smaller variance. We demonstrate the advantage of the proposed method by comparing it with an existing method.

A Video Manager for Relational Stream Processing Systems

Video streams are increasing by the development of network environments and price reduction of camera devices. Therefore needs of applications for querying video streams such as surveillance systems are on the increase. Meanwhile, relational stream processing systems have been researched for querying tuple streams. It is hard to deal with video streams by relational stream processing systems because video streams have time dependency which tuple streams do not have. In this paper, we propose a video manager for relational stream processing systems to integrate video streams and tuple streams. It can be used to query video streams using relational operators by relational stream processing systems and we can obtain resulting video data using RTSP. A video manager manages video data as MPEG-4 files so that they can be played easily by common media players. Finally we implemented a prototype system and evaluated its performance.