• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2004.10a
• /
• pp.145-148
• /
• 2004

규칙 기반 추론 시스템에서 규칙의 속성 감축은 다양한 방법으로 제안되어 왔다. 규칙의 속성 감축은 퍼지 추론 시스템을 구현하는데 있어서 처리 시간을 단축시킬 수 있으나 규칙의 종속성 및 상관성을 고려하지 않을 경우 예상하지 못한 추론 결과를 얻을 수 있다. 따라서, 본 논문에서는 복합속성을 가진 규칙의 속성 감축과 상관성을 고려하기 위하여 러프집합의 특성 중 식별가능 행렬과 식별가능 함수를 이용하였다. 그리고 속성 감축에 사용된 규칙은 복합속성(composite attribute)을 가지는 감성 데이터를 이용하였다.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.563-568
• /
• 2020

본 연구에서는 블랙박스로 알려진 딥러닝 모델에 설명 근거를 제공할 수 있는 설명자 모델을 적용해 보았다. 영화평 감성 분석을 위해 MLP, CNN으로 구성된 딥러닝 모델과 결정트리의 앙상블인 Gradient Boosting 모델을 이용하여 감성 분류기를 구축하였다. 설명자 모델로는 기울기(gradient)을 기반으로 하는 IG와 레이어 사이의 가중치(weight)을 기반으로 하는 CAM, 그리고 설명가능한 대리 모델을 이용하는 LIME과 입력 속성에 대한 선형모델을 추정하는 SHAP을 사용하였다. 설명자 모델의 특성을 보기 위하여 히트맵과 관련성 높은 N개의 속성을 추출해 보았다. 설명자가 제공하는 기여도에 따라 입력 속성을 제거해 가며 분류기 성능 변화를 측정하는 정량적 평가도 수행하였다. 또한, 사람의 판단 근거와의 일치도를 살펴볼 수 있는 '설명 근거 정확도'라는 새로운 평가 방법을 제안하여 적용해 보았다.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.586-590
• /
• 2023

인터넷과 소셜미디어 사용량의 급증으로, 제품 리뷰, 온라인 피드백, 소셜 미디어 게시물 등을 통해 고객의 감정을 파악하는 것이 중요해졌다. 인공지능이 활용되어 고객이 제품이나 서비스의 어떤 부분에 만족하거나 불만을 가지는지를 분석하는 연구를 ABSA라고 하며 이미 해외에서는 이런 연구가 활발하게 이루어지는 반면, 국내에서는 상대적으로 부족한 상황이다. 이 연구에서는 ABSA의 두 개의 주요 작업인 ACD와 ASC에 대해 생성 모델 중 하나인 T5 모델을 사용하는 방법론을 제시한다. 이 방법론은 기존 판별 모델을 사용하는 것에 비해 시간과 성능 측면에서 크게 향상되었음을 보여준다.

• Information Systems Review
• /
• v.24 no.1
• /
• pp.1-19
• /
• 2022

Social media-based communication has become crucial part of our personal and official lives. Therefore, it is no surprise that social media sentiment analysis has emerged an important way of detecting potential customers' sentiment trends for all kinds of companies. However, social media sentiment analysis suffers from huge number of sentiment features obtained in the process of conducting the sentiment analysis. In this sense, this study proposes a novel method by using Bayesian Network. In this model MBFS (Markov Blanket-based Feature Selection) is used to reduce the number of sentiment features. To show the validity of our proposed model, we utilized online review data from Yelp, a famous social media about restaurant, bars, beauty salons evaluation and recommendation. We used a number of benchmarking feature selection methods like correlation-based feature selection, information gain, and gain ratio. A number of machine learning classifiers were also used for our validation tasks, like TAN, NBN, Sons & Spouses BN (Bayesian Network), Augmented Markov Blanket. Furthermore, we conducted Bayesian Network-based what-if analysis to see how the knowledge map between target node and related explanatory nodes could yield meaningful glimpse into what is going on in sentiments underlying the target dataset.

• Journal of Intelligence and Information Systems
• /
• v.26 no.4
• /
• pp.1-25
• /
• 2020

Sentiment Analysis (SA) is a Natural Language Processing (NLP) task that analyzes the sentiments consumers or the public feel about an arbitrary object from written texts. Furthermore, Aspect-Based Sentiment Analysis (ABSA) is a fine-grained analysis of the sentiments towards each aspect of an object. Since having a more practical value in terms of business, ABSA is drawing attention from both academic and industrial organizations. When there is a review that says "The restaurant is expensive but the food is really fantastic", for example, the general SA evaluates the overall sentiment towards the 'restaurant' as 'positive', while ABSA identifies the restaurant's aspect 'price' as 'negative' and 'food' aspect as 'positive'. Thus, ABSA enables a more specific and effective marketing strategy. In order to perform ABSA, it is necessary to identify what are the aspect terms or aspect categories included in the text, and judge the sentiments towards them. Accordingly, there exist four main areas in ABSA; aspect term extraction, aspect category detection, Aspect Term Sentiment Classification (ATSC), and Aspect Category Sentiment Classification (ACSC). It is usually conducted by extracting aspect terms and then performing ATSC to analyze sentiments for the given aspect terms, or by extracting aspect categories and then performing ACSC to analyze sentiments for the given aspect category. Here, an aspect category is expressed in one or more aspect terms, or indirectly inferred by other words. In the preceding example sentence, 'price' and 'food' are both aspect categories, and the aspect category 'food' is expressed by the aspect term 'food' included in the review. If the review sentence includes 'pasta', 'steak', or 'grilled chicken special', these can all be aspect terms for the aspect category 'food'. As such, an aspect category referred to by one or more specific aspect terms is called an explicit aspect. On the other hand, the aspect category like 'price', which does not have any specific aspect terms but can be indirectly guessed with an emotional word 'expensive,' is called an implicit aspect. So far, the 'aspect category' has been used to avoid confusion about 'aspect term'. From now on, we will consider 'aspect category' and 'aspect' as the same concept and use the word 'aspect' more for convenience. And one thing to note is that ATSC analyzes the sentiment towards given aspect terms, so it deals only with explicit aspects, and ACSC treats not only explicit aspects but also implicit aspects. This study seeks to find answers to the following issues ignored in the previous studies when applying the BERT pre-trained language model to ACSC and derives superior ACSC models. First, is it more effective to reflect the output vector of tokens for aspect categories than to use only the final output vector of [CLS] token as a classification vector? Second, is there any performance difference between QA (Question Answering) and NLI (Natural Language Inference) types in the sentence-pair configuration of input data? Third, is there any performance difference according to the order of sentence including aspect category in the QA or NLI type sentence-pair configuration of input data? To achieve these research objectives, we implemented 12 ACSC models and conducted experiments on 4 English benchmark datasets. As a result, ACSC models that provide performance beyond the existing studies without expanding the training dataset were derived. In addition, it was found that it is more effective to reflect the output vector of the aspect category token than to use only the output vector for the [CLS] token as a classification vector. It was also found that QA type input generally provides better performance than NLI, and the order of the sentence with the aspect category in QA type is irrelevant with performance. There may be some differences depending on the characteristics of the dataset, but when using NLI type sentence-pair input, placing the sentence containing the aspect category second seems to provide better performance. The new methodology for designing the ACSC model used in this study could be similarly applied to other studies such as ATSC.

• Journal of Digital Convergence
• /
• v.17 no.3
• /
• pp.107-117
• /
• 2019

In the era of SNS, many people relies on it to express their emotions about various kinds of products and services. Therefore, for the companies eagerly seeking to investigate how their products and services are perceived in the market, emotion mining tasks using dataset from SNSs become important much more than ever. Basically, emotion mining is a branch of sentiment analysis which is based on BOW (bag-of-words) and TF-IDF. However, there are few studies on the emotion mining which adopt feature selection (FS) methods to look for optimal set of features ensuring better results. In this sense, this study aims to propose FS methods to conduct emotion mining tasks more effectively with better outcomes. This study uses Twitter and SemEval2007 dataset for the sake of emotion mining experiments. We applied three FS methods such as CFS (Correlation based FS), IG (Information Gain), and ReliefF. Emotion mining results were obtained from applying the selected features to nine classifiers. When applying DT (decision tree) to Tweet dataset, accuracy increases with CFS, IG, and ReliefF methods. When applying LR (logistic regression) to SemEval2007 dataset, accuracy increases with ReliefF method.

• The Journal of the Convergence on Culture Technology
• /
• v.6 no.1
• /
• pp.111-120
• /
• 2020

As an application of big data and artificial intelligence techniques, this study proposes an atypical language-based sentimental opinion poll methodology, unlike conventional opinion poll methodology. An alternative method for the sentimental classification model based on existing statistical analysis was to collect real-time Twitter data related to parliamentary elections and perform empirical analyses on the Polarity and Intensity of public opinion using attribute-based sensitivity analysis. In order to classify the polarity of words used on individual SNS, the polarity of the new Twitter data was estimated using the learned Lasso and Ridge regression models while extracting independent variables that greatly affect the polarity variables. A social network analysis of the relationships of people with friends on SNS suggested a way to identify peer group sensitivity. Based on what voters expressed on social media, political opinion sensitivity analysis was used to predict party approval rating and measure the accuracy of the predictive model polarity analysis, confirming the applicability of the sensitivity analysis methodology in the political field.

• Journal of Digital Convergence
• /
• v.18 no.2
• /
• pp.249-259
• /
• 2020

This study presents a recently obtained social media data set based upon the case study of Electric Vehicles (EV) and looks to implement a sentiment analysis (SA) in order to gain insights. This study uses two methods in order to fully analyze the public's sentiment on EVs. First, we implement a SA tool in which we used to extract the sentiment of comments. Next we labeled the data with these sentiments obtained and classified them. While performing classification we found the problem of dimensionality and also explored the use of feature selection (FS) models in order to reduce the data set's dimensionality. We found that the use of three FS models (Chi Squared, Information Gain and ReliefF) showed the most promising results when used alongside a logistic and support vector machines classification algorithm. the contributions of this paper are in providing an real-world example of social media text analytics which can be adopted in many other areas of research and business. Moving forward researchers can use the methodological approach in this paper to further refine and improve their own case uses in text analytics.

• Science of Emotion and Sensibility
• /
• v.12 no.4
• /
• pp.433-442
• /
• 2009

People are tending to buy products through the Internet rather than purchasing them from the store. Some of the consumers give their feedback on line such as reviews, replies, comments, and blogs after they purchased the products. People are also likely to get some information through the Internet. Therefore, companies and public institutes have been facing this situation where they need to collect and analyze reviews or public opinions for them because many consumers are interested in other's opinions when they are about to make a purchase. However, most of the people's reviews on web site are too numerous, short and redundant. Under these circumstances, the emotion scanning system of text documents on the web is rising to the surface. Extracting writer's opinions or subjective ideas from text exists labeled words like GI(General Inquirer) and LKB(Lexical Knowledge base of near synonym difference) in English, however Korean language is not provided yet. In this paper, we labeled positive, negative, and neutral attribute at 4 POS(part of speech) which are noun, adjective, verb, and adverb in Korean dictionary. We extract construction patterns of emotional words and relationships among words in sentences from a large training set, and learned them. Based on this knowledge, comments and reviews regarding products are classified into two classes polarities with positive and negative using SO-PMI, which found the optimal condition from a combination of 4 POS. Lastly, in the design of the system, a flexible user interface is designed to add or edit the emotional words, the construction patterns related to emotions, and relationships among the words.

• 한국어정보학회:학술대회논문집
• /
• 2016.10a
• /
• pp.189-194
• /
• 2016

본 연구는 다국어 감성사전 및 감성주석 코퍼스 구축 프로젝트인 MUSE 프로젝트의 일환으로 한국어 감성사전을 구축하기 위해 대표적인 영어 감성사전인 SentiWordNet을 이용하여 한국어 감성사전을 구축하는 방법의 의의와 한계점을 검토하는 것을 목적으로 한다. 우선 영어 SentiWordNet의 117,659개의 어휘중에서 긍정/부정 0.5 스코어 이상의 어휘를 추출하여 구글 번역기를 이용해 자동 번역하는 작업을 실시하였다. 그 중에서 번역이 되지 않거나, 중복되는 경우를 제거하고, 언어학 전문가들의 수작업으로 분류해낸 결과 3,665개의 감성어휘를 획득할 수 있었다. 그러나 이마저도 병명이나 순수 감성어휘로 보기 어려운 사례들이 상당수 포함되어 있어 실제 이를 코퍼스에 적용하여 감성어휘를 자동 판별했을 때에 맛집 코퍼스에서의 재현율(recall)이 긍정과 부정에서 각각 47.4%, 37.7%, IT 코퍼스에서 각각 55.2%, 32.4%에 불과하였다. 이와 더불어 F-measure의 경우, 맛집 코퍼스에서는 긍정과 부정의 값이 각각 62.3%, 38.5%였고, IT 코퍼스에서는 각각 65.5%, 44.6%의 낮은 수치를 보여주고 있어, SentiWordNet 기반의 감성사전은 감성사전으로서의 역할을 수행하기에 충분하지 않은 것으로 나타났다. 이를 통해 한국어 감성사전을 구축할 때에는 한국어의 언어적 속성을 고려한 체계적인 접근이 필요함을 역설하고, 현재 한국어 전자사전 DECO에 기반을 두어 보완 확장중인 SELEX 감성사전에 대해 소개한다.