Tae-Bok Yoon

A music recommendation system based on personal preference analysis

In this paper, we propose a music recommendation system based on user preference analysis. The system builds music models using hidden Markov models with mel frequency cepstral coefficients, which are features of sound wave. Each song is modeled with an HMM and the similarity measure between songs are defined based on the models. With the similarity measure, the songs the user listened to in the past are grouped and analyzed. The system recommends pieces of music to the user based on the result of the analysis. We evaluate our system with virtual users who have various preferences, and observe which recommendation lists the system generates. In most cases, the system recommends the pieces of music which are close to userpsilas preference.

A personalized music recommendation system with a time-weighted clustering

We propose a music recommendation system which provides personalized services. The system keeps a userpsilas listening list and analyzes it to select pieces of music similar to the userpsilas preference. For analysis, the system extracts properties from the sound wave of music and the time when the user listens to music. Based on the properties, a pieces of music is mapped into a point in the property space and the time is converted into the weight of the point. The more recently the user listens to the music, the more the weight increases. We apply the K-means clustering algorithm to the weighted points. The K-means algorithm is modified so that the number of clusters are dynamically changed. By using our K-means clustering algorithm, we can recommend pieces of music which are close to userpsilas preference even though he likes several genres. We also consider the time when music is released. When recommending, the system selects pieces of music which is close to and released contemporarily with the userpsilas preference. We perform experiments with one hundred pieces of music. In this paper we present and evaluate algorithms to recommend system.

Cooperative learning by replay files in real-time strategy game

In real-time strategy game, the game artificial intelligence is not smart enough. That makes people feel boring. In this paper, we suggest a novel method about a cooperative learning of build-order improving the artificial intelligence in real-time strategy game in order to make games funny. We use the huge game replay file for it.

The method for extraction of meaningful places based on behavior information of user

Recently, the advance of mobile devices has made various services possible beyond simple communication. One of services is the predicting the future path of users and providing the most suitable location based service based on the prediction results. Almost of these prediction methods are based on previous path data. Thus, calculating similarities between current location information and the previous trajectories for path prediction is an important operation. The collected trajectory data have a huge amount of location information generally. These information needs the high computational cost for calculating similarities. For reducing computational cost, the meaningful location based trajectory model approaches are proposed. However, most of the previous researches are considering only the physical information such as stay time and the distance for extracting the meaningful locations. Thus, they will probably ignore the characteristics of users for meaningful location extraction. In this paper, we suggest a meaningful location extracting and trajectory simplification approach considering the stay time, distance, and additionally interaction information of user. The method collects the location information using GPS device and interaction information between the user and the others. Using these data, the proposed method defines the proximity of the people who are related with the user. The system extracts the meaningful locations based on the calculated proximities, stay time and distance. Using the selected meaningful locations the trajectories are simplified. For verifying the usability of the proposed method, we collect the behavioral data of smart phone users. Using these data, we measure the suitability of meaningful location extraction method, and the accuracy of prediction approach based on simplified trajectories. Following these result, we confirmed the usability of proposed method.

Behavior Pattern Modeling based Game Bot detection

Korean Game industry, especially MMORPG(Massively Multiplayer Online Game) has been rapidly expanding in these days. But As game industry is growing, lots of online game security incidents have also been increasing and getting prevailing. One of the most critical security incidents is `Game Bots`, which are programs to play MMORPG instead of human players. If player let the game bots play for them, they can get a lot of benefic game elements (experience points, items, etc.) without any effort, and it is considered unfair to other players. Plenty of game companies try to prevent bots, but it does not work well. In this paper, we propose a behavior pattern model for detecting bots. We analyzed behaviors of human players as well as bots and identified six game features to build the model to differentiate game bots from human players. Based on these features, we made a Naive Bayesian classifier to reasoning the game bot or not. To evaluated our method, we used 10 game bot data and 6 human Player data. As a result, we classify Game bot and human player with 88% accuracy.

A Personalized Music Recommendation System with a Time-weighted Clustering

We propose a music recommendation system which provides personalized services. The system keeps a userpsilas listening list and analyzes it to select pieces of music similar to the userpsilas preference. For analysis, the system extracts properties from the sound wave of music and the time when the user listens to music. Based on the properties, a pieces of music is mapped into a point in the property space and the time is converted into the weight of the point. The more recently the user listens to the music, the more the weight increases. We apply the K-means clustering algorithm to the weighted points. The K-means algorithm is modified so that the number of clusters are dynamically changed. By using our K-means clustering algorithm, we can recommend pieces of music which are close to userpsilas preference even though he likes several genres. We also consider the time when music is released. When recommending, the system selects pieces of music which is close to and released contemporarily with the userpsilas preference. We perform experiments with one hundred pieces of music. In this paper we present and evaluate algorithms to recommend system.

Connected Component-Based and Size-Independent Caption Extraction with Neural Networks

Captions which appear in images include information that relates to the images. In order to obtain the information carried by captions, the methods for text extraction from images have been developed. However, most existing methods can be applied to captions with fixed height of stroke`s width. We propose a method which can be applied to various caption size. Our method is based on connected components. And then the edge pixels are detected and grouped into connected components. We analyze the properties of connected components and build a neural network which discriminates connected components which include captions from ones which do not. Experimental data is collected from broadcast programs such as news, documentaries, and show programs which include various height caption. Experimental result is evaluated by two criteria : recall and precision. Recall is the ratio of the identified captions in all the captions in images and the precision is the ratio of the captions in the objects identified as captions. The experiment shows that the proposed method can efficiently extract captions various in size.

A method for learning users` preference on fuzzy values using neural networks and k-means clustering

Fuzzy sets are good for abstracting and unifying information using natural language like terms. However, fuzzy sets embody vagueness and users may have different attitude to the vagueness, each user may choose difference one as the best among several fuzzy values. In this paper, we develop a method teaming a user`s, preference on fuzzy values and select one which fits to his preference. Users` preferences are modeled with artificial neural networks. We gather learning data from users by asking to choose the best from two fuzzy values in several representative cases of comparing two fuzzy sets. In order to establish tile representative comparing cases, we enumerate more than 600 cases and cluster them into several groups. Neural networks ate trained with the users` answer and the given two fuzzy values in each case. Experiments show that the proposed method produces outputs closet to users` preference than other methods.