O-Joun Lee

Exploiting character networks for movie summarization

Movie summarization focuses on providing as much information as possible for shorter movie clips while still keeping the content of the original movie and presenting a faster way for the audience to understand the movie. In this paper, we propose a novel method to summarize a movie based on character network analysis and the appearance of protagonist and main characters in the movie. Experiments were carried out for 2 movies (Titanic (1997) and Frozen (2013)) to show that our method outperforms conventional approaches in terms of the movie summarization rate.

Sequence Clustering-based Automated Rule Generation for Adaptive Complex Event Processing

In Complex Event Processing (CEP), complex events are detected according to a set of rules that are defined by domain experts. However, it makes the reliability of the system decreased as dynamic changes occur in the domain environment or domain experts make mistakes. To address such problem, this study proposes a Sequence Clustering-based Automated Rule Generation (SCARG) that can automatically generate rules by mining decision-making history of domain experts based on sequence clustering and probabilistic graphical modeling. Furthermore, based on a two-way learning approach, the proposed method is able to support automated regular or occasional rule updates. It makes self-adaptive CEP system possible by combining the rule generation method and the existing dynamic CEP systems. This technique is verified by establishing an automated stock trading system, and the performance of the system is measured in terms of the rate of return. The study solves the aforementioned problems and shows excellent results with an increase of 19.32% in performance when compared to the existing dynamic CEP technique. The paper presents a novel framework for complex event processing.The proposed method has been designed by temporal probabilistic model.It has been applied to stock trading system.

A computational model of transmedia ecosystem for story-based contents

Story-based contents (e.g., novel, movies, and computer games) have been dynamically transformed into various media. In this environment, the contents are not complete in themselves, but closely connected with each other. Also, they are not simply transformed form a medium to other media, but expanding their stories. It is called as a transmedia storytelling, and a group of contents following it is called as a transmedia ecosystem. Since the contents are highly connected in terms of the story in the transmedia ecosystem, the existing content analysis methods are hard to extract relationships between the contents. Therefore, a proper content analysis method is needed with considering expansions of the story. The aim of this work is to understand how (and why) such contents are transformed by i) defining the main features of the transmedia storytelling and ii) building the taxonomy among the transmedia patterns. More importantly, computational transmedia ecosystem is designed to process a large number of the contents, and to support high understandability of the complex transmedia patterns.

Event-Driven Trust Refreshment on Ambient Services

Since trust among entities can change according to various conditions, it is necessary for ambient services to determine when and how the trust has to be updated. Therefore, our contribution in this paper is to present: 1) a new definition of trust that can be extended to various domains; 2) a novel method based on social events and patterns to trigger trust refreshment in ambient services; and 3) a web application framework (called SocioScope) for collecting and analyzing data from multiple data sources. Finally, the case study suggests that this proposal could be applied to trust-aware ambient and recommendation systems.

Predictive Clustering for performance stability in collaborative filtering techniques

Model-based collaborative filtering improves the fundamental limitations of the collaborative filtering facing the issues of data sparsity and scalability while presenting other constraints of high costs of model building and the tradeoff between performance and scalability. Such tradeoff results in reduced coverage, which is one sort of the sparsity issue. Furthermore, high model building costs lead to unstable performance driven by cumulative changes in the domain environment. To solve these problems, we propose Predictive Clustering-based CF (PCCF) that incorporates the Markov model and fuzzy clustering with Clustering based CF (CBCF). The method improves performance instability by tracking the changes in user preferences and bridging the gap between the static model and dynamic users. Furthermore, the issue of reduced coverage is also improved by expanding the coverage based on transition probabilities. The proposed method has been validated by testing the robustness of performance instability and scalability-performance tradeoff. In comparison with the existing techniques, the suggested method shows slight performance improvement. Notwithstanding, it is more advanced than the existing techniques in terms of the range that indicates the level of performance fluctuation. This signifies that the proposed method, despite the slight performance improvement, clearly offers better performance stability compared to the existing techniques.

Dynamic Traffic Light Control System Based on Process Synchronization Among Connected Vehicles

Vehicular traffic is tremendously increasing around the world, especially in large urban areas. The resulting congestion has become a key issue and emerging research topic to transportation specialist and decision makers. In this study, inspired by recent advanced vehicle technologies, we take into account in improving traffic flow in real-time problem. In order to solve the problem, we propose a new approach to manage traffic flow at the intersection in real-time via controlling by traffic light scheduling. In particular, the proposed method is based on process synchronization theory and connected vehicle technology where each vehicle is able to communicate with others. The traffic deadlock is also taken into consideration in case of high traffic volume. The simulation shows the potential results comparing with the existing traffic management system.

Spatio-temporal query contextualization for microtext retrieval in social media: Spatio-temporal query contextualization for microtext retrieval in social media

Generally, texts on the social media (eg, Twitter and Facebook) are too short (microtexts) to understand the meaning and to search for relevant texts. It is difficult for the conventional information retrieval systems to conduct the searching tasks. Thereby, in this paper, we propose a novel approach on query contextualization by integrating all possible microtexts by considering spatio‐temporal contexts. The proposed approach consists of two steps, which are (i) to understand and process microtexts in social media and (ii) to reformulate the queries for searching for relevant microtexts in these social media. To evaluate the performance of the query contextualization approach, microtexts from Twitter have been collected during 4 months.

Adaptive Complex Event Processing Based on Collaborative Rule Mining Engine

Complex Event Processing (CEP) detects complex events or patterns of event sequences based on a set of rules defined by a domain expert. However, it lowers the reliability of a system as the set of rules defined by an expert changes along with dynamic changes in the domain environment. A human error made by an expert is another factor that may undermine the reliability of the system. In an effort to address such problems, this study introduces Collaborative Rule Mining Engine (CRME) designed to automatically mine rules based on the history of decisions made by a domain expert by adopting a collaborative filtering approach, which is effective in mimicking and predicting human decision-making in an environment where there are sufficient data or information to do so. Furthermore, this study suggests an adaptive CEP technique, which does not hamper the reliability since it prevents potential errors caused by mistakes of domain experts and adapts to changes in the domain environment on its own as it is linked to the system proposed by Bharagavi [10]. In a bid to verify this technique, an automated stocks trading system will be established and its performance will be measured using the rate of return.

Owner-Borrower Model for Recommenders in O2O Services

With remarkable successes of sharing economy services (e.g., UBER (https://www.uber.com), Airbnb (https://www.airbnb.com), and so on), the amount of items which are distributed through these services is rapidly increasing. Therefore recommender systems for the sharing economy services are required. However, the existing recommenders are hard to support the sharing economy services, since they have focused on a ‘Item-User’ model that the recommenders provide satisfiable items to consumers (users) in accordance with only the consumers’ preferences. In this regard, we suggest a novel recommendation model, ‘Owner-Borrower’ model which considers the preferences of both sides: owners and borrowers of properties (items). Also, we propose a recommendation method based on the proposed model by applying a tensor factorization method and the Gale-Shapley algorithm. The tensor factorization is used for estimating preferences of the owners and the borrowers. With the estimated preferences, the Gale-Shapley algorithm makes optimal matches between the borrowers and the owners’ properties.

Modeling affective character network for story analytics

Abstract Consideration of the stories included in the narrative works is important for analyzing and providing narrative works (e.g., movies, novels, and comics) to users. In this study, we analyzed the stories in a narrative work with three goals: (i) eliciting, (ii) modeling, and (iii) utilizing the stories. Based upon our previous studies regarding ‘character networks’ (i.e., social networks among characters in the stories), we elicited the stories with three methods: (i) composing affective character networks with affective relationships among the characters, (ii) measuring temporal changes in tension according to the flows of the stories, and (iii) detecting affective events which refer to dramatic changes in the tension. The affective relationships contain emotional changes of the characters on each segment of the stories. By aggregating the characters’ emotional changes, we measured the tension of each segment. We called it ‘Affective Fluctuation’ and represented it as a discrete function (Affective Fluctuation Function, AFF). The AFFs enable us to detect affective events by using gradients of them and measure similarities among the stories by comparing their shapes. Also, we proposed a computational model of the stories by annotating the affective events and characters involved in those events. Finally, we demonstrated a practical application with a recommendation method which exploited the similarities between stories. Additionally, we verified the reliabilities and efficiencies of the proposed method for narrative works in the real world.