• KIPS Transactions on Software and Data Engineering
• /
• v.5 no.11
• /
• pp.563-568
• /
• 2016

The word sense disambiguation problem is to find the correct sense of an ambiguous word having multiple senses in a dictionary in a sentence. We regard this problem as a multi-class classification problem and classify the ambiguous word by using Support Vector Machines. Context words of the ambiguous word, which are extracted from Sejong sense tagged corpus, are represented to two kinds of vector space. One vector space is composed of context words vectors having binary weights. The other vector space has vectors where the context words are mapped by word embedding model. After experiments, we acquired accuracy of 87.0% with context word vectors and 86.0% with word embedding model.

• Journal of Korea Multimedia Society
• /
• v.25 no.3
• /
• pp.493-501
• /
• 2022

This paper is a study on context-sensitive spelling error correction and uses the Korean WordNet (KorLex)[1] that defines the relationship between words as a graph to improve the performance of the correction[2] based on the vector information of the word embedded in the correction technique. The Korean WordNet replaced WordNet[3] developed at Princeton University in the United States and was additionally constructed for Korean. In order to learn a semantic network in graph form or to use it for learned vector information, it is necessary to transform it into a vector form by embedding learning. For transformation, we list the nodes (limited number) in a line format like a sentence in a graph in the form of a network before the training input. One of the learning techniques that use this strategy is Deepwalk[4]. DeepWalk is used to learn graphs between words in the Korean WordNet. The graph embedding information is used in concatenation with the word vector information of the learned language model for correction, and the final correction word is determined by the cosine distance value between the vectors. In this paper, In order to test whether the information of graph embedding affects the improvement of the performance of context- sensitive spelling error correction, a confused word pair was constructed and tested from the perspective of Word Sense Disambiguation(WSD). In the experimental results, the average correction performance of all confused word pairs was improved by 2.24% compared to the baseline correction performance.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.10
• /
• pp.1300-1306
• /
• 2018

The purpose of this study is to see possibility of Char2Vec as alternative of Word2Vec that most famous word embedding model in Sentence Similarity Measure Problem by Deep-Learning. In experiment, we used the Siamese Ma-LSTM recurrent neural network architecture for measure similarity two random sentences. Siamese Ma-LSTM model was implemented with tensorflow. We train each model with 200 epoch on gpu environment and it took about 20 hours. Then we compared Word2Vec based model training result with Char2Vec based model training result. as a result, model of based with Char2Vec that initialized random weight record 75.1% validation dataset accuracy and model of based with Word2Vec that pretrained with 3 million words and phrase record 71.6% validation dataset accuracy. so Char2Vec is suitable alternate of Word2Vec to optimize high system memory requirements problem.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.1
• /
• pp.24-29
• /
• 2022

Semantic role labeling task used in various natural language processing fields, such as information extraction and question answering systems, is the task of identifying the arugments for a given sentence and predicate. Predicate used as semantic role labeling input are extracted using lexical analysis results such as POS-tagging, but the problem is that predicate can't extract all linguistic patterns because predicate in korean language has various patterns, depending on the meaning of sentence. In this paper, we propose a korean predicate recognition method using neural network model with pre-trained embedding models and lexical features. The experiments compare the performance on the hyper parameters of models and with or without the use of embedding models and lexical features. As a result, we confirm that the performance of the proposed neural network model was 92.63%.

• Journal of KIISE
• /
• v.42 no.12
• /
• pp.1568-1574
• /
• 2015

Relation extraction plays a role in for the process of transforming a sentence into a form of knowledge base. In this paper, we focus on predicates in a sentence and aim to identify the relevant knowledge base properties required to elucidate the relationship between entities, which enables a computer to understand the meaning of a sentence more clearly. Distant Supervision is a well-known approach for relation extraction, and it performs lexicalization tasks for knowledge base properties by generating a large amount of labeled data automatically. In other words, the predicate in a sentence will be linked or mapped to the possible properties which are defined by some ontologies in the knowledge base. This lexical and ontological linking of information provides us with a way of generating structured information and a basis for enrichment of the knowledge base.

• KIISE Transactions on Computing Practices
• /
• v.23 no.12
• /
• pp.698-702
• /
• 2017

Since the introduction of Deep Neural Networks to the Natural Language Processing field, two major approaches have been considered for modeling text. One method involved learning embeddings, i.e. the distributed representations containing abstract semantics of words or sentences, with the textual context. The other strategy consisted of composing the embeddings trained by the above to get embeddings of longer texts. However, most studies of the composition methods just adopt word embeddings without consideration of the optimal embedding unit and the optimal method of composition. In this paper, we conducted experiments to analyze the optimal embedding unit and the optimal composition method for modeling longer texts, such as documents. In addition, we suggest a new discourse-based composition to overcome the limitation of the sequential composition method on composing sentence embeddings.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.361-363
• /
• 2018

FAQ 시스템은 주어진 질문과 가장 유사한 질의를 찾아 이에 대한 답을 제공하는 시스템이다. 질의 간의 유사도를 측정하기 위해 문장을 벡터로 표현하며 일반적으로 TFIDF, Okapi BM25와 같은 방법으로 계산한 단어 가중치 벡터를 이용하여 문장을 표현한다. 하지만 단어 가중치 벡터는 어휘적 정보를 표현하는데 유용한 반면 단어의 의미적인(semantic) 정보는 표현하기 어렵다. 본 논문에서는 이를 보완하고자 딥러닝을 이용한 문장 임베딩을 구축하고 단어 가중치 벡터와 문장 임베딩을 조합한 문장 유사도 계산 모델을 제안한다. 또한 문장 임베딩 구현 시 self-attention 기법을 적용하여 문장 내 중요한 부분에 가중치를 주었다. 실험 결과 제안하는 유사도 계산 모델은 비교 모델에 비해 모두 높은 성능을 보였고 self-attention을 적용한 실험에서는 추가적인 성능 향상이 있었다.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.652-657
• /
• 2018

With the rapid development of e-commerce, many customers can now express their opinion on various kinds of product at discussion groups, merchant sites, social networks, etc. Discerning a consensus opinion about a product sold online is difficult due to more and more reviews become available on the internet. Opinion Mining, also known as Sentiment analysis, is the task of automatically detecting and understanding the sentimental expressions about a product from customer textual reviews. Recently, researchers have proposed various approaches for evaluation in sentiment mining by applying several techniques for document, sentence and aspect level. Aspect-based sentiment analysis is getting widely interesting of researchers; however, more complex algorithms are needed to address this issue precisely with larger corpora. This paper introduces an approach of knowledge representation for the task of analyzing product aspect rating. We focus on how to form the nature of sentiment representation from textual opinion by utilizing the representation learning methods which include word embedding and compositional vector models. Our experiment is performed on a dataset of reviews from electronic domain and the obtained result show that the proposed system achieved outstanding methods in previous studies.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.161-166
• /
• 2021

본 연구는 한국어 문장 임베딩(embedding)에 담겨진 언어적 속성을 평가하기 위한 프로빙 태스크(Probing Task)를 소개한다. 프로빙 태스크는 임베딩으로부터 문장의 표층적, 통사적, 의미적 속성을 구분하는 문제로 영어, 폴란드어, 러시아어 문장에 적용된 프로빙 테스크를 소개하고, 이를 기반으로하여 한국어 문장의 속성을 잘 보여주는 한국어 문장 임베딩 프로빙 태스크를 설계하였다. 언어 공통적으로 적용 가능한 6개의 프로빙 태스크와 한국어 문장의 주요 특징인 주어 생략(SubjOmission), 부정법(Negation), 경어법(Honorifics)을 추가로 고안하여 총 9개의 프로빙 태스크를 구성하였다. 각 태스크를 위한 데이터셋은 '세종 구문분석 말뭉치'를 의존구문문법(Universal Dependency Grammar) 구조로 변환한 후 자동으로 구축하였다. HuggingFace에 공개된 4개의 다국어(multilingual) 문장 인코더와 4개의 한국어 문장 인코더로부터 획득한 임베딩의 언어적 속성을 프로빙 태스크를 통해 비교 분석한 결과, 다국어 문장 인코더인 mBART가 9개의 프로빙 태스크에서 전반적으로 높은 성능을 보였다. 또한 한국어 문장 임베딩에는 표층적, 통사적 속성보다는 심층적인 의미적 속성을 더욱 잘 담고 있음을 확인할 수 있었다.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.1
• /
• pp.1-8
• /
• 2021

In the political field of news articles, there are polarized and biased characteristics such as conservative and liberal, which is called political bias. We constructed keyword-based dataset to classify bias of news articles. Most embedding researches represent a sentence with sequence of morphemes. In our work, we expect that the number of unknown tokens will be reduced if the sentences are constituted by subwords that are segmented by the language model. We propose a document embedding model with subword tokenization and apply this model to SVM and feedforward neural network structure to classify the political bias. As a result of comparing the performance of the document embedding model with morphological analysis, the document embedding model with subwords showed the highest accuracy at 78.22%. It was confirmed that the number of unknown tokens was reduced by subword tokenization. Using the best performance embedding model in our bias classification task, we extract the keywords based on politicians. The bias of keywords was verified by the average similarity with the vector of politicians from each political tendency.