Showing papers by "Geun-Sik Jo published in 2015"

Alleviating the cold-start problem by incorporating movies facebook pages

Abstract: Recommender systems are generally known as predictive ecosystem which recommends an appropriate list of items that may imply their similar preference or interest. Nevertheless, most discussed issues in recommendation system research domain are the cold-start problem. In this paper we proposed a novel approach to address this problem by combining similarity values obtain from a movie "Facebook Pages". To achieve this, we first compute users' similarity according to the rating cast on our Movie Rating System. Then, we combined similarity value obtain from user's genre interest in "Like" information extracted from "Facebook Pages". Finally, all the similarity values are combined to produce a new user's similarity value. Our experiment results show that our approach is outperformed in cold-start problem compared to the benchmark algorithms. To evaluate whether our system is strong enough to recommend higher accuracy recommendation to users, we also conducted prediction coverage in this research work.

Personalized advertisement system using social relationship based user modeling

Abstract: The influence of social relationships has received considerable attention in recommendation systems. In this paper, we propose a personalized advertisement recommendation system based on user preference and social network information. The proposed system uses collaborative filtering and frequent pattern network techniques using social network information to recommend personalized advertisements. Frequent pattern network is employed to alleviate cold-start and sparsity problems of collaborative filtering. For the social relationship modeling, direct and indirect relations are considered and relation weight between users is calculated by using six degrees of Kevin Bacon. Weight `1' is given to those who have connections directly, and weight `0' is given to those who are over six steps away and hove no relation to each other. According to a research of Kevin Bacon, everybody can know certain people through six depths of people. In order to improve prediction accuracy, we apply social relationship to user modeling. In our experiments, advertisement information is collected and item rating and user information including social relations are extracted from a social network service. The proposed system applies user modeling between collaborative filtering and frequent pattern network model to recommend advertisements according to user condition. User's types are composed with combinations of both techniques. We compare the performance of the proposed method with that of other methods. From the experimental results, a proposed system applying user modeling using social relationships can achieve better performance and recommendation quality than other recommendation systems.

An Unsupervised Approach for Identifying the Infobox Template of Wikipedia Article

Abstract: Wikipedia infoboxes serve as important structured information source in the web. To author infobox for a particular article, volunteers required a considerable amount of manual effort to identify the respective infobox template. Thus, an automatic process to mark infobox template might be useful and beneficial for the Wikipedia contributors. In this paper, we present a Natural Language Processing (NLP)-based automated approach to identify the infobox template in an unsupervised fashion. The proposed approach has been developed by using semantic relations (hyponym and holonym) and word features of Wikipedia articles. Our approach works in three steps: first it processes the raw text of the article to generate sets of words, next it apply the proposed algorithm to identify the infobox type and finally point out the infobox template from the large pool of template list. The effectiveness of the proposed approach has been proved in terms of autonomous and accuracy, by a data-driven experiment.

Aligned thumbnails-based video browsing system with Chromecast

Abstract: Well-known media applications of Chromecast such as YouTube, Viki usually use a time slider on the second screen for video navigation. However, the time slider does not offer easy way to access particular contents. Here we propose an aligned thumbnails-based video browsing system with Chromecast using content-based video browsing approach by automatically detecting scenes. It uses hierarchical sliding interface to browse the playing video on TV easily. Through the experimental results, our proposed system helps search specific contents quickly.

RANSAC versus CS-RANSAC

Abstract: A homography matrix is used in computer vision field to solve the correspondence problem between a pair of stereo images. RANSAC algorithm is often used to calculate the homography matrix by randomly selecting a set of features iteratively. CS-RANSAC algorithm in this paper converts RANSAC algorithm into two-layers. The first layer is addressing sampling problem which we can describe our knowledge about degenerate features by mean of Constraint Satisfaction Problems (CSP). By dividing the input image into a N×N grid and making feature points into discrete domains, we can model the image into the CSP model to efficiently filter out degenerate features. By expressing the knowledge about degenerate feature samples using CSP in the first layer, so that computer has knowledge about how to skip computing the homography matrix in the model estimation step for the second layer. The experimental results show that the proposed CS-RANSAC algorithm can outperform the most of variants of RANSAC without sacrificing its execution time.

Context-Aware Navigation Element for Collaborative Nonlinear Video Authoring Tool

Abstract: In recent years, online video streaming service has become more popular. High internet bandwidth, powerful mobile devices, and advance of video annotation techniques have raised the popularity of the rapidly growing interactive video genre. This research focuses on enabling collaboration among authors of interactive nonlinear videos that provide alternative story plots for viewers to choose as part of their interactive behaviors. We discuss the “irrelevant navigation elements” problem, which could occur when multiple nonlinear video authors want to reuse a shared interactive video. Then, we propose a system called MAVINS, a managed navigation element for interactive nonlinear videos, to solve the aforementioned problem. The system is implemented as a web-based authoring tool and interactive video player for user-creator and user-viewer, respectively. Experimentation in self-directed learning was conducted to demonstrate the problem that occurs in current approaches as well as to evaluate the effectiveness...

Text-Based Semantic Video Annotation for Interactive Cooking Videos

Abstract: Videos represent one of the most frequently used forms of multimedia applications. In addition to watching videos, people control slider bars of video players to find specific scenes and want detailed information on certain objects in scenes. However, it is difficult to support user interactions in current video formats because of a lack of metadata for facilitating such interactions. This paper proposes a text-based semantic video annotation system for interactive cooking videos to facilitate user interactions. The proposed annotation process includes three parts: the synchronization of recipes and corresponding cooking videos based on a caption-recipe alignment algorithm; the information extraction of food recipes based on lexico-syntactic patterns; and the semantic interconnection between recognized entities and web resources. The experimental results show that the proposed system is superior to existing alignment algorithms and effective in semantic cooking video annotation.

Temporal Interval Reasoning with Korean Historical Event

Abstract: In historical event knowledge base, relationships between time event intervals are complex that is not easy to express its complex relations. We make the time event intervals reasoning in order to express complicated relations among events in Korean historical event based on 13 Allen's temporal interval relations, but it takes too much time to do the reasoning. In pre-computed model, if we have quantitative information in Korean history dataset, we pre-compute time event relations from possible pair of quantitative event intervals to qualitative event relation triples with Allen's operator model. In this paper, we propose effective hybrid algorithm, which is a combination of pre-computed model and backward chaining to get a real time query and perform the reasoning more effectively between time event intervals in Korean history dataset with more than 3 billion triples. As user imposes questions in English, we reformulate it into qualitative structure query in which consists of Allen's operators and then look up for answer in the existing qualitative answers that are already pre-computed. Otherwise, we infer only necessary entries from quantitative temporal information to compute the inferred facts to get the answer during query time based on backward chaining. We implemented this approach with a Spark Scala framework, which is a new parallel system programming that is capable of processing large-scale dataset efficiently and speeding up our reasoning process. With this reasoning process, we get a real time query with response times in a small number of milliseconds.