[in Japanese]

View full abstract

[in Japanese]

View full abstract

Latent Dirichlet Allocation (LDA) has been widely used for analyzing latent topics of documents. It assigns a probability distribution over topics to each individual word ofdocuments, and then assigns latent topics as particular words which tend to be appeared in a particular topic. In the method, documents are treated as bag-of-words. It does not deal with the relation between words to precisely express the contents of documents. In this study, to estimate latent topics more precisely, we propose a method to assign a probability distribution over topics to pairs of words. Through document retrieval tasks, we investigate how we should provide constraints on pairs of words which are useful for extracting latent topics, and show that LDA can be improved upon by assigning probability distributions to pairs of words.

View full abstract

Many recommender systems construct user profiles by conducting machine learning on users' preference data for showing users their favorite items. Traditionally, researchers of recommender systems have pursued the accuracy of the recommendation. However, recent research trends are changing to the value improvement on the overall services including recommendation process. We especially focused on users' notices while using recommender systems. If the acquired user profile is shown to the user, he might conceive some notices on his preference because user profiles are constructed implicitly based on the users' usual activities such as browsing and shopping. In this study, we investigate the relationship between the visualization of user profiles and their notices on their preferences.

View full abstract

In recent years, studies have focused on the development of recommender systems that emphasize criteria other than accuracy. One such measure, serendipity, is defined as a measure that indicates how the recommender system can find unexpected and useful items for users. In this study,we propose a fusion-based recommender system as a serendipity-oriented recommender system. Our system possesses mechanisms that can cause extrinsic and intrinsic cues, and it enables users to discover valuable items from such cues through their sagacity.We consider that such mechanisms are required for the development of the serendipity-oriented recommender system. The key idea of this system is the fusion-based approach, through which the system mixes two user-input items to find new items that have the mixed features. The contributions of this paper are as follows: providing a recommender system that adopts a fusion-based approach to improve serendipity; practically evaluating the recommender system through user tests using a real book data set from Rakuten Books; and showing the effectiveness of the system compared to Amazon that is one of the recommender systems on websites from the viewpoint of serendipity.

View full abstract

There is an increasing number of researches of complex networks such as World Wide Web, social networks and biological networks. They have found the property of a scale free, a small world, a large clustering coefficient, and so on. One of the hot topics in this area is community detection. Nodes belonging to a community are likely to have common properties. For instance, in the World Wide Web, a community may be a set of pages which belong to a same topic. Community structure is undoubtedly a key characteristic of complex networks. In this paper, we present a new framework for finding communities in complex networks and evaluate detecting the community. This framework uses the idea of intersection graph and uses semantic information such as text and attributes which appear in networks.

View full abstract

In this paper, we propose a new feature vector for 3D object retrieval, which we call Local Feature Correlation Descriptor (LCoD). Given a 3D object, we first render depth-buffer images from multiple viewpoints. We then extract local features from each depth-buffer image. For every depth-buffer image, we compute the correlation matrix of local features, and define the vector as LCoD, which is obtained by the elements of the correlation matrix. Our experiments on the Princeton Shape Benchmark show that LCoD achieves the First Tier of 0.4708, which exhibits higher search performance than conventional techniques.

View full abstract

News websites are becoming an important information source from which readers can obtain information about daily events as well as experience various impressions such as happiness, sadness, and anger, depending on how the news articles are written. For a contentious news topic, in particular, diverse news websites may have different reporting tendencies that may also vary over time, inducing different emotions in readers. We make use of “emotions” as the evaluation index of the way in which news articles are written and propose an analysis method that can visually compare reporting tendencies in news websites. We first design multidimensional emotions that are suitable for news articles and construct an emotion dictionary. We then calculate emotion values for news articles and news websites and present the differences in websites' reporting tendencies and variations over time. In this study, we propose a method for analyzing news websites' reporting tendencies based on the various emotions they trigger in readers and verify the effectiveness by implementing a prototype for twenty-five news websites (15 in Japan and 10 overseas).

View full abstract

In this paper, we propose the strategy for the distributed database technology to integrate data of diffused databases on the Internet by Peer-to-Peer cloud computing technology, which enables to construct logical data spaces for controlling and managing them by SQL. In recent years, there are several kinds and functions of databases, and they have been constructed original databases including their characteristics and faculties. Utilizing them would be expected to create new wonderful services. The proposal model in this paper has following characteristics: one is realizing controls and managements of data relationships by SQL over the functions of database like RDBMS, KVS and locales of each database. The other is the service scalability by collaborations among distributed nodes. From the result of evaluation, proposal model in this paper clarifies that it combines the advanced data management by SQL and service scalability by collaborations of distributed nodes. As for the next step, we discuss about the improvement of service performance by the replication technology such as cashing manner.

View full abstract

The Great East Japan Earthquake caused devastating damage to infrastructure and loss of life. Many people used social media to communicate and share information through the series of earthquake events and aftershocks. We analyzed over 400 million Tweets and the Twitter network in Japan before and after the earthquake, which revealed the earthquake that occurred in the Twitter milieu itself. We performed analyses of two kinds. As a result, we suggest that the role of Twitter as a means to share information changed during the earthquake period.

View full abstract

Recently, curation of web contents is widely spotlighted. Curation is selection and summarization of contents from one's viewpoint on a theme. It is important for curation to represent the context in which one can interpret curated contents. However, with existing services, such as Naver matome, both representing a variety of context and understanding the context intuitively are difficult. Consequently, one can not curate web contents effectively by trial and error. Therefore, we developed a web content curation system using cover generate engine on two-dimensional layout. Two-dimensional layout enables to layout many contents flexibly and to overlook the whole of the curated page. Cover of a web page, which is composed of images, texts and colors in the web page, enables to understand each content and context intuitively. To evaluate our system, we conducted an comparative experiment with Naver matome. As a result, our system showed more variety of representation of curations, and had more visual elements in the curation pages than Naver matome. Also, users using our system tended to be attentive toappearance and intuition of contents.

View full abstract

YouTube as the most popular video hosting service is growing rapidly, making it difficult for users to ?nd video that fit their particular interest. Implementation of video recommendation method is one way to solve this problem. However, several difficulties remain, one of them is no clear relation between the original and recommended video. Moreover, there are many cases that the recommended videos mostly have similar contents to the original videos. In this paper, we try to solve these problems by concept of words and to explicitly state the reason for recommendation. We build video recommendation application for YouTube using Formal Concept Analysis, and we performed the experiments to evaluate our system. By analyzing the results of experiments, we look at the effect of our system on user's load.

View full abstract

Recently, the number of studies on sentiment analysis has increased. These studies have attempted to determine whether people's impressions of news articles tend to be positive or negative. However, these studies have focused on people's impressions of individual articles and have not taken into account the effects of the reading order of articles. In this study, we conducted questionnaire surveys and examined how people's impressions of an article change in accordance with their impressions of preceding articles. The results showed that in many cases, reading order had no effect on people's impressions of news articles. However, in several cases, people's impressions of an article were enhanced after reading the preceding article.

View full abstract