• 한국HCI학회:학술대회논문집
• /
• 2008.02b
• /
• pp.475-481
• /
• 2008

본 연구는 어떤 사건에 대한 예측 정확성 여부와 기억을 회상할 때의 정서 상태가 후견지명 편향 (hindsight bias)에 미치는 영향을 알아보고자 하였다. 이에 valence 축에 따라 긍정적 정서와 부정적 정서를 일으키는 두 가지 음악을 제시하고 두 조건에 대하여 기억에 대한 과잉 확신이 얼마나 달라지는가를 분석하였다. 예측 정확성 여부에 대해서는 실험 결과 데이터 중 예측 일치 조건과 불일치 조건으로 나누어 후견지명 편향에 끼치는 영향과 정서와의 상호작용이 있는가를 분석하였다. 사람들은 예측과 반대되는 결과를 접했을 때 결과에 anchoring하여 기억을 회상하려는 편향이 더욱 커졌으며 부정적인 정서보다 긍정적 정서 상태일 때 후견지명 편향이 더욱 커졌음을 밝혔다. 특히 예측과 상이한 결과 피드백을 받고 긍정적 정서 상태일 때 가장 많은 왜곡 현상을 보였으며, 예측 불일치/ 부정적 정서 조건, 예측 일치/ 긍정적 정서 조건, 예측 일치/ 부정적 정서 조건 순으로 후견지명 편향을 보였다. 이 결과는 정서 상태보다 어떤 사건에 대한 예측 정확성 여부가 후견지명 편향에 더 큰 영향을 준다는 것을 시사한다. 본 연구의 실험실 상황을 통하여 자기와 관련이 없는 중립적 과제를 통해서도 후견지명 편향이 나타남을 알 수 있었다. 특히 그 동안 거의 이루어지지 않았던 정서와 후견지명 편향의 관계를 밝히고, 기존의 예측 정확성에 따른 편향을 설명하는 모델간 논쟁이 많았으나 실험 결과가 motivational model을 지지함을 밝혔음에 의의가 있다.

• Survey Research
• /
• v.10 no.1
• /
• pp.169-186
• /
• 2009

The propensity score method is used to minimize the bias level in social survey, which comes from nonresponse. The theoretical concept and the background of the propensity score method is discussed first. The propensity score method was first applied in the epidemiology observational study. I have summarized the process of the three propensity score methods that were used to reduce estimation bias in this study. Matching by propensity score is applied to the relatively large control group. Subclassification has the advantage of using whole control group data and regression adjustment is applied to multiple covariates as well as propensity score of each unit is computable and usable. Lastly, the application procedures of propensity score method to reduce the nonresponse bias is suggested and its applicability to real situation is reviewed with the existing data.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.7
• /
• pp.949-955
• /
• 2022

Deep learning-based text summarization models are not free from datasets. For example, a summarization model trained with a news summarization dataset is not good at summarizing other types of texts such as internet posts and papers. In this study, we define this phenomenon as Data Bias Problem (DBP) and propose two training methods for solving it. The first is the 'proper nouns masking' that masks proper nouns. The second is the 'length variation' that randomly inflates or deflates the length of text. As a result, experiments show that our methods are efficient for solving DBP. In addition, we analyze the results of the experiments and present future development directions. Our contributions are as follows: (1) We discovered DBP and defined it for the first time. (2) We proposed two efficient training methods and conducted actual experiments. (3) Our methods can be applied to all summarization models and are easy to implement, so highly practical.

• Knowledge Management Research
• /
• v.23 no.2
• /
• pp.229-251
• /
• 2022

Online reviews are useful for other consumers to make reasonable purchase decisions by providing previous buyers' experiences. However, when online reviewers are biased, online reviews do not accurately reflect the true quality of the product. Therefore, we investigated the characteristics of reviewers with underreporting bias to cope with the problem of declining reliability of online reviews. In this context, this study attempted to examine the characteristics of reviewers with underreporting bias using 14,165 reviews of Steam, an online game platform. As a result of the analysis, reviewers with underreporting bias mainly write reviews positively, write reviews within a short period from the game release date, but tend to write reviews after playing games for longer time, and write reviews when purchasing high-priced games. Since this study has explored the characteristics of reviewers showing underreporting bias, it will be meaningful as a basic study to cope with the problem caused by underreporting bias.

• The KIPS Transactions:PartD
• /
• v.17D no.4
• /
• pp.245-252
• /
• 2010

In this paper, we propose an effective filtering method for skyline queries in mobile ad hoc networks (MANETs). Most existing researches assume that data is uniformly distributed. Under these assumptions, the previous works focus on optimizing the energy consumption due to the limited battery power. However, in practice, data distribution is skewed in a specific region. In order to reduce the energy consumption, we propose a new filtering method considering the data distribution. We verify the performance of the proposed method through a comparative experiment with an existing method. The results of the experiment confirm that the proposed method reduces the communication overhead and execution time compared to an existing method.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.533-538
• /
• 2023

문서에서 저자의 의도와 주제, 그 안에 포함된 감성을 분석하는 것은 자연어 연구의 핵심적인 주제이다. 이와 유사하게 특정 글에 포함된 정치적 문화적 편향을 분석하는 것 역시 매우 의미 있는 연구주제이다. 우리는 최근 발생한 한 사건에 대하여 여러 신문사와 해당 신문사에서 생산한 기사를 중심으로 해당 글의 정치적 편향을 정량화 하는 방법을 제시한다. 그 방법은 선택된 주제어들의 문장 공간에서의 거리를 중심으로 그래프를 생성하고, 생성된 그래프의 기계학습을 통하여 편향과 특징을 분석하였다. 그리고 그 그래프들의 시간적 변화를 추적하여 특정 신문사에서 특정 사건에 대한 입장이 시간적으로 어떻게 변화하였는지를 동적으로 보여주는 그래프 애니메이션 시스템을 개발하였다. 실험을 위하여 최근 이슈에 대하여 12개의 신문사에서 약 2000여 개의 기사를 수집하였다. 그 결과, 약 82%의 정확도로 일반적으로 알려진 정치적 편향을 예측할 수 있었다. 또한, 학습 데이터에 쓰이지 않은 신문기사를 활용하여도 같은 정도의 정확도를 보임을 알 수 있었다. 우리는 이를 통하여 신문기사에서의 정치적 편향은 작성자나 신문사의 특성이 아니라 주제어들의 문장 공간에서의 거리 관계로 특성화할 수 있음을 보였다. 할 수 있다.

• Communications for Statistical Applications and Methods
• /
• v.17 no.4
• /
• pp.515-525
• /
• 2010

We investigated design-based properties of the ordinary least square estimator(OLSE) and the weighted least square estimator(WLSE) in a panel regression model. Given a complex data we derive the magnitude of the design-based bias of two estimators and show that the bias of WLSE is smaller than that of OLSE. We also conducted a simulation study using Korean welfare panel data in order to compare design-based properties of two estimators numerically. In the study we found the followings. First, the relative bias of OLSE is nearly two times larger than that of WLSE and the bias ratio of OLSE is greater than that of WLSE. Also the relative bias of OLSE remains steady but that of WLSE becomes smaller as the sample size increases. Next, both the variance and mean square error(MSE) of two estimators decrease when the sample size increases. Also there is a tendency that the proportion of squared bias in MSE of OLSE increases as the sample size increase, but that of WLSE decreases. Finally, the variance of OLSE is smaller than that of WLSE in almost all cases and the MSE of OLSE is smaller in many cases. However, the number of cases of larger MSE of OLSE increases when the sample size increases.

• Journal of KIISE:Databases
• /
• v.33 no.1
• /
• pp.69-85
• /
• 2006

Broadcast has been often used to disseminate the frequently requested data efficiently to a large volume of mobile clients over a single or multiple channels. Conventional concurrency control protocols for mobile transactions are not suitable for the wireless broadcast environments due to the limited bandwidth of the up-link communication channel. In wireless broadcast environments, the server often broadcast different data items with different frequency to incorporate the data access patterns of mobile transactions. The previously proposed concurrency control protocols for mobile transactions in wireless broadcast environments are focused on the mobile transactions with uniform data access patterns. However, these protocols perform poorly when the data access pattern of update mobile transaction are not uniform but skewed. The update mobile transactions with skewed data access patterns will be frequently aborted and restarted due 4o the update conflict of the same data items with a high access frequency. In this paper, we propose an energy-efficient concurrence control protocol for mobile transactions with skewed data access as well as uniform data access patterns. Our protocol use a random back-off technique to avoid the frequent abort and restart of update mobile transactions. We present in-depth experimental analysis of our method by comparing it with existing concurrency control protocols. Our performance analysis show that it significantly decrease the average response time, the amount of upstream and downstream bandwidth usage over existing protocols.

• Journal of the Korea Society for Simulation
• /
• v.33 no.2
• /
• pp.27-35
• /
• 2024

We aimed to shed light on the issue of perpetuating societal disparities by analyzing the impact of inherent biases present in datasets used for training artificial intelligence models on the predictions generated by Artificial Intelligence(AI). Therefore, to examine the influence of data bias on AI models, we constructed an original dataset containing biases related to gender wage gaps and subsequently created a de-identified dataset. Additionally, by utilizing the decision tree algorithm, we compared the outputs of AI models trained on both the original and de-identified datasets, aiming to analyze how data de-identification affects the biases in the results produced by artificial intelligence models. Through this, our goal was to highlight the significant role of data de-identification not only in safeguarding individual privacy but also in addressing biases within the data.

• Journal of the Korean Geotechnical Society
• /
• v.30 no.8
• /
• pp.25-37
• /
• 2014

In load and resistance factor design (LRFD) method, resistance factors are typically calibrated using resistance bias factors obtained from either only the data within ${\pm}2{\sigma}$ or the data except the tail values of an assumed probability distribution to increase the reliability of the database. However, the data selection approach has a shortcoming that any low-quality data inadvertently included in the database may not be removed. In this study, a data quality evaluation method, developed based on the quality of static load test results, the engineering characteristics of in-situ soil, and the dimension of aggregate piers, is proposed for use in constructing database. For the evaluation of the method, a total 65 static load test results collected from various literatures, including static load test reports, were analyzed. Depending on the quality of the database, the comparison between bias factors, coefficients of variation, and resistance factors showed that uncertainty in estimating bias factors can be reduced by using the proposed data quality evaluation method when constructing database.