Dongjin Choi

Least Slack Time Rate First: New Scheduling Algorithm for Multi-Processor Environment

Real-time systems have to complete the execution of a task within the predetermined time while ensuring that the execution results are logically correct. Such systems require scheduling methods that can adequately distribute the given tasks to a processor. Scheduling methods that all tasks can be executed within a predetermined deadline are called an optimal scheduling. In this paper, we propose a new and simple scheduling algorithm (LSTR: least slack time rate first) as a dynamic-priority algorithm for a multi-processor environment and demonstrate its optimal possibility through various tests.

Text analysis for detecting terrorism-related articles on the web

Classifying web documents is considered as one of the most important tasks to reveal the terrorism-related documents. Internet provides a lot of valuable information to the users and the amount of web contents is progressively increasing. This makes it very difficult to identify potentially dangerous documents. Simply extracting keywords from documents is not enough to classify the contents. To build automated document classification systems, many techniques have been studied so far, but they are mostly statistical and knowledge-based approaches. These methods, however, do not yield satisfactory results because of complexity of natural languages. To overcome this deficiency, we propose a method to use word similarity based on WordNet hierarchy and n-gram data frequency. This method was tested with the sampled New York Times articles by querying four distinct words from four different areas. Experimental results show our proposed method effectively extracts context words from the text and identifies terrorism-related documents.

Information Retrieval Techniques to Grasp User Intention in Pervasive Computing Environment

This paper contains a method to construct context data which can help an application grasp user intention in pervasive computing environment (PCE). There are various devices in which a user is interested for the user intention (intended behavior such as entering, reading and sleeping). And the core of PCE is to provide appropriate services adapted for grasped user intention through processing context information received from device sensors. Therefore, this paper suggests an approach based on co-occurrence and statistical method, kinds of information retrieval technique, to grasp user intention based on diverse device sensors (context information), including both physical and logical objects.

Automatic Image Annotation Using Semantic Text Analysis

This paper proposed a method to find annotations corresponding to given CNN news documents for detecting terrorism image or context information. Assigning keywords or annotation to image is one of the important tasks to let machine understand web data written by human. Many techniques have been suggested for automatic image annotation in the last few years. Many researches focused on the method to extract possible annotation using low-level image features. This was the basic and traditional approach but it has a limitation that it costs lots of time. To overcome this problem, we analyze images and theirs co-occurring text data to generate possible annotations. The text data in the news documents describe the core point of news stories according to the given images and titles. Because of this fact, this paper applied text data as a resource to assign image annotations using TF (Term Frequency) value and WUP values of WordNet. The proposed method shows that text analysis is another possible technique to annotate image automatically for detecting unintended web documents.

Extracting User Interests on Facebook

Recently, through the rapid development of smart devices, Facebook has been publicly recognized as a representative social network service. Facebook has profile information that is the basis to form network relationship between people as well as closed information sharing between users. Also, social plug-in “Like” is a feature that Facebook includes. Current researches based on Facebook have a problem of not considering this “Like” feature. This paper has proposed the method of extracting user’s interest by using Term Frequency of nouns and “Likes.” “Posts” and “Likes” were collected through Facebook Open API. Collected “Posts” were preprocessed and Term Frequency of nouns was calculated. After calculating weights of user interests by using Term Frequency of nouns and “Likes,” user interests were extracted by higher ranked user interests weighting.

Sentiment analysis for tracking breaking events: a case study on twitter

Social media such as Twitter and Facebook can be considered as a new media different from the typical media group. The information on social media spread much faster than any other traditional news media due to the fact that people can upload information with no constrain to time or location. People also express their emotional status to let others know what they feel about information. For this reason many studies have been testing social media data to uncover hidden information under textual sentences. Analyzing social media is not simple due to the huge volume and variety of data. Many researches dealt with limited domain area to overcome the size issue. This study focuses on how the flow of sentiments and frequency of tweets are changed from November to December in 2009. We analyzed 110 million tweets collected by Stanford University and LIWC (Linguistic Inquiry Word Count) for sentiment analysis. We did find that people were not happy in afternoon but they were happy in night time as many psychologists suggested before. After analyzing large volume of tweets, we were also able to find the precise day when breaking events occurred. This study offer diverse evidence to prove that Twitter has valuable information for tracking breaking news over the world.

Solving English Questions through Applying Collective Intelligence

Many researchers have been using n-gram statistics which is providing statistical information about cohesion among words to extract semantic information in web documents. Also, the n-gram has been applied in spell checking system, prediction of user interest and so on. This paper is a fundamental research to estimate lexical cohesion in documents using trigram, 4gram and 5gram offered by Google. The main purpose of this paper is estimating possibilities of Google n-gram using TOEIC question data sets.

Text Editor based on Google Trigram and its Usability

This paper contains contents about a text editor based on Google trigram and its performance. Google has constructed n-gram data from uni- to 5-gram through analyzing huge document set. The n-gram data is a representative type of collective intelligence. Especially, big grams such as 4 and 5-gram contain more detail semantic context (semantics). However, 4 and 5-grams are hard to deal in computer due to enormous size. Therefore, in this paper, we apply only trigram to the text editor and evaluate its usability. In the experiment, we evaluate how much this text editor efficient according to the typing speed of people. Usability of the Google trigram was estimated by time cost which is required to type into the text editor. Keywords-component; N-Gram; Text Editor; Word Recommendation; Performance Evaluation

LDA-based face recognition using multiple distance training face images with low user cooperation

The LDA-based face recognition using face images by actual distance as training images showed good performance. However, it causes user inconvenience as it requires the user to move multiple distance in person to acquire face images for initial user registration. In this paper, LDA-based face recognition applicable to robotic environments is proposed. The proposed method can get face images by distance by zooming in on face of users without their cooperation and use those images for training. The experimental results show that the face recognition performance of the proposed method was as good as that of the method using face images by distance, which are taken after users are located at fixed positions within a fixed distance, for training.

Automatic Evaluation of Document Classification Using N-Gram Statistics

Due to the development of World Wide Web technologies, people are living in the place flooding trillions of web pages in every moment. The amount of web size has been increasing dramatically. For this reason, it is getting more difficult to find relevant web documents corresponding to what users want to read. Classifying documents into predefined categories is one of the most important tasks in Natural Language Processing field. Over the years, many statistical and linguistical approaches have been applied to overcome traditional classification machine. However, it still remains in unsolved problem. There is a no perfect solution to machine understand human language yet. We have to consider every possibility for making machine think like human does. In this paper, we propose a method for classifying textural document using n-gram co-occurrence statistics which have a great possibility to find similarities between given documents. We also compare our proposed method with traditional method suggested by Keselj. This paper only covers simple approaches and still needs more sophisticated experiments. However, the performance using this method is better than the Keselj approach.