• ETRI Journal
• /
• v.35 no.5
• /
• pp.838-848
• /
• 2013

This paper presents a novel approach to automatically generate Korean multiword sentiment expressions by using a seed sentiment lexicon and a large-scale domain-specific corpus. A multiword sentiment expression consists of a seed sentiment word and its contextual words occurring adjacent to the seed word. The multiword sentiment expressions that are the focus of our study have a different polarity from that of the seed sentiment word. The automatically extracted multiword sentiment expressions show that 1) the contextual words should be defined as a part of a multiword sentiment expression in addition to their corresponding seed sentiment word, 2) the identified multiword sentiment expressions contain various indicators for polarity shift that have rarely been recognized before, and 3) the newly recognized shifters contribute to assigning a more accurate polarity value. The empirical result shows that the proposed approach achieves improved performance of the sentiment analysis system that uses an automatically generated lexicon.

• The Journal of the Acoustical Society of Korea
• /
• v.20 no.3
• /
• pp.24-33
• /
• 2001

this paper describes a baseline for an implementation of a corpus-based Korean TTS system. The conventional TTS systems using small-sized speech still generate machine-like synthetic speech. To overcome this problem we introduce the corpus-based TTS system which enables to generate natural synthetic speech without prosodic modifications. The corpus should be composed of a natural prosody of source speech and multiple instances of synthesis units. To make a phone level synthesis unit, we train a speech recognizer with the target speech, and then perform an automatic phoneme segmentation. We also detect the fine pitch period using Laryngo graph signals, which is used for prosodic feature extraction. For break strength allocation, 4 levels of break indices are decided as pause length and also attached to phones to reflect prosodic variations in phrase boundaries. To predict the break strength on texts, we utilize the statistical information of POS (Part-of-Speech) sequences. The best triphone sequences are selected by Viterbi search considering the minimization of accumulative Euclidean distance of concatenating distortion. To get high quality synthesis speech applicable to commercial purpose, we introduce a domain specific database. By adding domain specific database to general domain database, we can greatly improve the quality of synthetic speech on specific domain. From the subjective evaluation, the new Korean corpus-based TTS system shows better naturalness than the conventional demisyllable-based one.

• The Journal of the Korea Contents Association
• /
• v.21 no.12
• /
• pp.132-139
• /
• 2021

This corpus-driven research expounded the compilation of the most frequently occurring academic words in the domain of syntax and compared the extracted wordlist with Academic Word List(AWL) of Coxhead(2000) and General Service List(GSL) of West(1953) to examine their distribution and coverage within the syntax corpus. A specialized 546,074 token corpus, composed of widely used must-read syntax textbooks for English education majors, was loaded into and analyzed with AntWordProfiler 1.4.1. Under the parameter of lexical frequency, the analysis identified 288(50.5%) AWL word forms, appeared 16 times or more, as well as 218(38.2%) AWL items, occurred not exceeding 15 times. The analysis also indicated that the coverage of AWL and GSL accounted for 9.19% and 78.92% respectively and the combination of GSL and AWL amounted to 88.11% of all tokens. Given that AWL can be instrumental in serving broad disciplinary needs, this study highlighted the necessity to compile the domain-specific AWL as a lexical repertoire to promote academic literacy and competence.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.6
• /
• pp.828-833
• /
• 2009

WWW which is an applicable massive set of document on the Web is a thesaurus of various information for users. However, Search engines spend a lot of time to retrieve necessary information and to filter out unnecessary information for user. In this paper, we propose the EFAM(the Extended Fuzzy AHP Method) model to manage the Web resource efficiently, and to make a decision in the problem of specific domain definitely. The EFAM model is concerned with the emotion analysis based on the domain corpus information, and it composed with systematic common concept grids by the knowledge of multiple experts. Therefore, The proposed the EFAM model can extract the documents by considering on the emotion criteria in the semantic context that is extracted concept from the corpus of specific domain and confirms that our model provides more efficient decision-making through an experiment than the conventional methods such as AHP and Fuzzy AHP which describe as a hierarchical structure elements about decision-making based on the alternatives, evaluation criteria, subjective attribute weight and fuzzy relation between concept and object.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.10
• /
• pp.1000-1004
• /
• 2010

A large number of current language processing systems use a part-of-speech tagger for preprocessing. Most language processing systems required a tagger with the highest possible accuracy. Specially, the use of domain-specific advantages has become a hot issue in machine translation community to improve the translation quality. This paper addresses a method for customizing an HMM or LHMM based English tagger from general domain to specific domain. The proposed method is to semi-automatically customize the output and transition probabilities of HMM or LHMM using domain-specific raw corpus. Through the experiments customizing to Patent domain, our LHMM tagger adapted by the proposed method shows the word tagging accuracy of 98.87% and the sentence tagging accuracy of 78.5%. Also, compared with the general tagger, our tagger improved the word tagging accuracy of 2.24% (ERR: 66.4%) and the sentence tagging accuracy of 41.0% (ERR: 65.6%).

• Journal of Information Processing Systems
• /
• v.17 no.5
• /
• pp.867-878
• /
• 2021

In context awareness and user intention tasks, dataset construction is expensive because specific domain data are required. Although pretraining with a large corpus can effectively resolve the issue of lack of data, it ignores domain knowledge. Herein, we concentrate on data domain knowledge while addressing data scarcity and accordingly propose a multi-channel long short-term memory (LSTM). Because multi-channel LSTM integrates pretrained vectors such as task and general knowledge, it effectively prevents catastrophic forgetting between vectors of task and general knowledge to represent the context as a set of features. To evaluate the proposed model with reference to the baseline model, which is a single-channel LSTM, we performed two tasks: voice phishing with context awareness and movie review sentiment classification. The results verified that multi-channel LSTM outperforms single-channel LSTM in both tasks. We further experimented on different multi-channel LSTMs depending on the domain and data size of general knowledge in the model and confirmed that the effect of multi-channel LSTM integrating the two types of knowledge from downstream task data and raw data to overcome the lack of data.

• Journal of the Korean Society of Clothing and Textiles
• /
• v.46 no.6
• /
• pp.1142-1160
• /
• 2022

Text data plays a significant role in understanding and analyzing trends in consumer, business, and social sectors. For text analysis, there must be a corpus that reflects specific domain knowledge. However, in the field of fashion, the professional corpus is insufficient. This study aims to develop a taxonomy and thesaurus that considers the specialty of fashion products. To this end, about 100,000 fashion vocabulary terms were collected by crawling text data from WSGN, Pantone, and online platforms; text subsequently was extracted through preprocessing with Python. The taxonomy was composed of items, silhouettes, details, styles, colors, textiles, and patterns/prints, which are seven attributes of clothes. The corpus was completed through processing synonyms of terms from fashion books such as dictionaries. Finally, 10,294 vocabulary words, including 1,956 standard Korean words, were classified in the taxonomy. All data was then developed into a web dictionary system. Quantitative and qualitative performance tests of the results were conducted through expert reviews. The performance of the thesaurus also was verified by comparing the results of text mining analysis through the previously developed corpus. This study contributes to achieving a text data standard and enables meaningful results of text mining analysis in the fashion field.

• Journal of the Korea Convergence Society
• /
• v.9 no.10
• /
• pp.45-53
• /
• 2018

Recently, studies about bilingual word embedding have been gaining much attention. However, bilingual word embedding with Korean is not actively pursued due to the difficulty in obtaining a sizable, high quality corpus. Local embeddings that can be applied to specific domains are relatively rare. Additionally, multi-word vocabulary is problematic due to the lack of one-to-one word-level correspondence in translation pairs. In this paper, we crawl 868,163 paragraphs from a Korean-English law corpus and propose three mapping strategies for word embedding. These strategies address the aforementioned issues including multi-word translation and improve translation pair quality on paragraph-aligned data. We demonstrate a twofold increase in translation pair quality compared to the global bilingual word embedding baseline.

• Journal of KIISE:Software and Applications
• /
• v.37 no.4
• /
• pp.297-304
• /
• 2010

Clustering method which based on sentence type or document genre is a technique used to improve translation quality of SMT(statistical machine translation) by domain-specific translation. But there is no previous research using sentence type and document genre information simultaneously. In this paper, we suggest an integrated clustering method that classifying sentence type by syntactic structure similarity and document genre by word similarity information. We interpolated domain-specific models from clusters with general models to improve translation quality of SMT system. Kernel function and cosine measures are applied to calculate structural similarity and word similarity. With these similarities, we used machine learning algorithms similar to K-means to clustering. In Japanese-English patent translation corpus, we got 2.5% point relative improvements of translation quality at optimal case.

• Journal of Intelligence and Information Systems
• /
• v.28 no.4
• /
• pp.251-273
• /
• 2022

Recently, research on applying text analysis to deep learning has steadily continued. In particular, researches have been actively conducted to understand the meaning of words and perform tasks such as summarization and sentiment classification through a pre-trained language model that learns large datasets. However, existing pre-trained language models show limitations in that they do not understand specific domains well. Therefore, in recent years, the flow of research has shifted toward creating a language model specialized for a particular domain. Domain-specific pre-trained language models allow the model to understand the knowledge of a particular domain better and reveal performance improvements on various tasks in the field. However, domain-specific further pre-training is expensive to acquire corpus data of the target domain. Furthermore, many cases have reported that performance improvement after further pre-training is insignificant in some domains. As such, it is difficult to decide to develop a domain-specific pre-trained language model, while it is not clear whether the performance will be improved dramatically. In this paper, we present a way to proactively check the expected performance improvement by further pre-training in a domain before actually performing further pre-training. Specifically, after selecting three domains, we measured the increase in classification accuracy through further pre-training in each domain. We also developed and presented new indicators to estimate the specificity of the domain based on the normalized frequency of the keywords used in each domain. Finally, we conducted classification using a pre-trained language model and a domain-specific pre-trained language model of three domains. As a result, we confirmed that the higher the domain specificity index, the higher the performance improvement through further pre-training.