Research Field : Natural Language Processing with Deep Learning

Research Subject: Bilingual Word Embedding with Parallel Corpora

Researcher: SeolHwa Lee

Dense real-valued vector representations of words or word embedding have recently gained increasing popularity in natural language processing, serving as invaluable features in a broad range of NLP tasks.

Research interest has recently extended to bilingual word embeddings. Bilingual word embedding models focus on the induction of a shared bilingual word embedding space where words from both languages are represented in a uniform language-independent manner such that similar words have similar representation.

These research goal make seed lexicon by aligned parallel data and make mapping function between seed lexicon and general-domain corpora such as Wikipedia. Therefore, we expect to applicate these research to other NLP tasks such as machine translation.


Research Subject: Bags of Features: Word Representation for a Limitless Vocabulary

Researcher: ChanHee Lee

‘Bags of Features’ is a simple yet effective way of representing a word as a fixed length vector. This method operates on the character level and therefore does not require a limited or predefined vocabulary. Furthermore, unlike the previous approaches of character level word modeling, this method does not depend on neural network architectures like CNN or RNN. In order to evaluate the effectiveness of our proposed model, we carried out a POS tagging experiment and compared the results with the one-hot encoding baseline as well as other state-of-the-art approaches in the literature. The BOF model outperformed the one-hot encoding baseline, with 49.68\% relative improvement for unseen word tagging accuracy, and 0.96\% improvement for all words. It also showed competitive results with state-of-the-art models that do not exploit extra data.


Research Subject : Korean Dependency Parsing and SyntaxNet

Researcher : Andrew Matteson

BLP is currently developing technology to increase the accuracy of dependency parsing. We are investigating the most effective way to use deep learning in order to process morphologically complex languages like Korean in an efficient and accurate manner. We are also looking into performing POS tagging and dependency parsing simultaneously using a single hybrid model. Dependency parsing paves the way for sentiment analysis and could also improve machine translation accuracy in the future. We are also working on morphological analysis using SyntaxNet, which could aid in processing less formal text such as Twitter. The use of pre-trained word vectors (word2vec) is also being investigated. For more details on the parsing implementation, check our seminar on SyntaxNet available on the Seminars page.


Research Subject : Opinion Mining with a Syntactic Structure Analyzer

Researcher : Yu WonHui

Sentiment analysis or opinion mining refers to the applications of natural language processing, computational linguistics, and text analytics to identify and extract subjective information of a given.Generally speaking, sentiment analysis aims to determine the attitude or intention of a speaker or a writer with respect to the topic or the overall contextual polarity of a document. The attitude may be his or her judgment or evaluation (see appraisal theory), affective state (that is to say, the emotional state of the author when writing), or the intended emotional communication (to say, the emotional effect the author wishes to have on the reader).


Research Subject: A Language Typewriter for Brain-Computer Interface

Researcher: SaeByeok Lee

Brain-computer interface (BCI) is a communication system that translates brain activity into commands for computers or other devices. In other words, BCIs create a new communication channel between the brain and an output device by bypassing the conventional motor output pathways consisting of nerves and muscles. This is particularly useful for facilitating communication for people who suffer from paralysis. Due to the low bit rate, it takes much more time to translate brain activity into commands, with main emphasis on the time it takes to input characters using BCI-based typewriters. Therefore, we propose a new typewriter which is accelerated by a language prediction model.


Research Field : Learning Analytics

Research Subject: Computer Aided Language Learning

Researcher: YeongWook Yang

A computer aided language learning system is a system that automatically controls educational elements. The system consists of a diagnosis, learning and assessment state. The diagnosis state estimates language ability of the present time cognitive abilities. After the learning state, a learning course is created using information gathered from the diagnosis state. Lastly, the assessment state supports the course management and assessment management about learning states. This system can be implemented to effectively learn different languages as well as other academic fields


Research Subject: Learning Conceptual Model Considering Digital Native’s Characteristics

Name: Jaechoon Jo

1. Project Concept

The development of ICT (Information & Communication Technology) due to the Smartphone, Smart pad, smart (intelligent) interface and cloud computing, changed the education environment thus creating a new education environment and learning technique called “Smart Learning”. Therefore, much need research is needed on this newly and ever changing education environment.
This research is to add a new perspective on the education environment by defining smart learning and developing new educational models and teaching-learning methods considering a digital native.

2. Research list

-Case studies on smart education on a national and international level, and the development of the digital native characterization and criteria
-Developing a teaching-learning method and conceptual model of smart learning.
-Designing a smart class environment and publishing smart learning guideline documents.

Support : This work was supported by WBS.

Research Subject : A LDT Based Comparative Analysis of English Vocabulary Learning using Smartphone

Name: Jaechoon Jo

The ability to have a strong English vocabulary, also affects the learning process of learning the English language. There are many ways in which someone can learn English, through schools, private classes, and from online resources. Smart Learning, is the method of learning English using a Smartphone, which in recent years, have been receiving a lot of attention by learners. The purpose of this study is to do a comparative analysis of the learning effects of Smart Learning based on an LDT model. We have done a comparative analysis consisting of two different groups- new methods using a Smartphone and the learning effects of tradition methods using printed text materials. The experiment showed that both of the groups had positive learning effects. In addition, our analysis showed that the learning effects of brain-based learning can further the learning abilities with the use of Smartphone’s rather than traditional methods using printed text materials.


Research Subject: A study on factor analysis to support knowledge based decisions for a smart class

Name: Jaechoon Jo

A smart class provides an environment that enables collaboration, sharing, and participation between teachers and students. Thanks to the great attention paid to the smart class idea, with a view to providing effective and efficient learning for students, many state-of-the-art tech- nologies have been applied to the field of education. However, simple infrastructure construction and the in- troduction of state-of-the-art technology have many limitations in obtaining the desired effects of a smart class. This study aims to discover the important elements that allow a smart class to achieve positive effects in education and to support the design and application of a smart class based on the derived elements. In the study, an integrated teaching and learning assistance system was applied to a smart class. A smart class environment was constructed that was applied and test operated in an elementary school for 4 weeks. Through the test operation of the smart class, the important elements of an effective smart class were determined to be system playfulness, perceived usefulness, perceived ease of use, and attitude toward class.

Research Subject: A study on Minimal Learning Model for Mind Wandering Judgement System of Online Learner

Name: Jaechoon Jo

In this study, we define minimal learning model, and develop minimal learning judgement system of online learner. this system consists of mind wandering judgement system and automatic words generation system.

Research Field 2

Research Field : Educational Data Mining & Artificial Intelligence in Education

Research Subject: Social Learning Platform COLLA

Researcher: Hyesung Ji

COLLA is a platform to build social networks among students who share the same class, subject or interests. COLLA stands for COLLAboration with people, content, and service. In other words, COLLA is a new social networking site which implement adaptive learning to satistify student’s learning needs. This platform provide students with services, such as: flipped learning, social learning platform, blended learning, contents sharing, learning by experience, and learning by doing.

Demo :

Project Interview : ‘디지털 에듀케이션’으로 만나는 미래 교육 환경


Research Subject: Knowledge-Sharing System Using Smart Devices

Researcher: WonBeom Yoon

This work describes a Knowledge-sharing system using Smart Devices. We provide divide our solutions in three methods; web search, the accumulated knowledge of data retrieval, and the users direct respond. Our system allows users to share their questions and answers through real-time updates and communications. Thus, they can give feedbacks and utilize other’s knowledge. In this sharing process, the qualified information are recommended to the inquiring user. Therefore, users can easily access their desired information without much research on the web.

Support : This work was supported by WBS.

Research Field : Information Retrieval

Research Subject: Intelligent Search Platform for the support of Traditional Culture

Researcher: Hye Sung Ji

To develop a platform that would act as a platform for both kind of users who want to consume traditional culture related contents as well as for users who would produce it. As a platform that would support many forms of traditional culture, it will not only deal its contents as traditional search engines do through keywords, but also generate its own keywords by analyzing the content which would be in the form of either text, image and videos. The technology that is researched is a semantic image detection method and a semantic video detection method as well as an evaluation method to measure its effectiveness.


Research Subject: Fashion Content Based Search & Coordination System

Researcher: Dong Yub Lee

To apply image pattern analysis technology, image big data machine learning (deep learning) technology, and actual image prediction technology done by the analysis of simple sketched images to the fashion (especially clothing) industry which is already in a visualized form of image. And develop a product that supports users shopping online for fashion.


Research Subject: A Personalized Recommendation Model of Content Consumption Routes

Researcher: Yeong Wook Yang

This study proposes a prediction algorithm that tracks users’ preferences and its changes that occur through the course of time and predicts the user’s course of consuming contents and a recommendation algorithm that recommends a series of contents in a predicted order for the user to consume according to their preferences. This study is followed by researches to verify the effectiveness of each algorithm and ultimately aimed to research and develop a personalized content consumption route recommendation model considering the user’s changes in preferences.

Content Route

Research Subject: Extraction of Users Intentions from Web Search Logs

Researcher: Hye Sung Ji

This research proposes a method to extract a user’s intention automatically and implementation of an intention map that supports a user. This method can appropriately search results using the user’s information needs accurately. It selects the user’s intentions based on their web history obtained from previous user’s queries and extracts user’s intentions by using a clustering algorithm and a user intention extraction algorithm. This is then represented in an intention map based on the theory of knowledge representation. We proceed to research useful and convenient information retrieved from general search engines.


User Interface & User Experience (UI/UX)

Research Subject: Fashion Content Based Search & Coordination System

Researcher:You Dong Yun

The purpose of this research and development is to provide seamless cultural enjoyment without any discrimination to the continuously increasing smart senior generation. To do this the development of cognitive response measurement technology that measures and evaluates cognitive response factors in a multidimensional manner and to develop a UI / UX technology tailored to the user’s cognitive response that can provide tailored content based on the analysis of one’s preference and behavior.