| HERMAN BERNARDIM ANDRADE GABRIEL | D, 中間発表 | ソーシャル・コンピューティング | 荒牧 英治, 渡辺 太郎, 若宮 翔子, 矢田 竣太郎 |
|
Title: Improving Named Entity annotation efficiency by adopting fuzzy entity boundaries
Abstract: When applying NLP models in real-world applications, a usually overlooked aspect is the impact caused on a model's performance by differences in characteristics between the training and the target datasets. Consequently, some way to adjust the model to the new domain is important to achieve good results. Although model fine-tuning with target domain labeled data can be effective, it relies on an annotation process, which is typically a manual and time-consuming task, especially when annotating Named Entities. "Boundary words," such as articles or modifiers, can induce ambiguity and inconsistency in a process that is already slow and error-prone. We assess the text annotation process to identify methods to enhance its efficiency. More specifically, we evaluate the impact caused by the entity boundary ambiguity in the annotation process and attempt to mitigate them by proposing annotation guidelines that relax the need for precise annotation by adopting fuzzy entity boundaries. Language of the presentation: English | |||
| ZHANG ZHOUQING | M, 2回目発表 | ソーシャル・コンピューティング | 荒牧 英治, 中村 哲, 若宮 翔子, 矢田 竣太郎 |
|
title: Text analyses of social media posts about breast cancer: Patient identification and sentiment analyses
abstract: Breast cancer is a major disease that threatens human life worldwide. Each year new diagnoses and deaths continue to increase, which triggers a great number of issues, including recognition to breast cancer, emotional expression, mental health, etc. In this study, we aim at studying how were the texts’ contents about breast cancer on social platforms (Chinese Weibo). Our goal is to examine whether emotional expression presents a difference among different breast cancer identity patients. We applied the language model to do the text classification and then we used the LIWC to conduct the sentiment analysis. Our results show that emotional expressions present obvious differences among breast cancer identity patients. language of the presentation: English | |||
| WU DONGMING | M, 2回目発表 | 知能コミュニケーション | 中村 哲, 荒牧 英治, 鳥澤 健太郎, 飯田 龍 |
|
title: Pretraining Language Model for Multi-hop Question Answering
abstract: Multi-hop question answering (QA) is a task to find answers to a given question using information from more than one document. In this research, we focus on improving language model (LM) pretraining to achieve better performance on downstream multi-hop QA. We proposed two pretraining methods to let LMs learn multiple types of link information and relation semantics between documents. The experiments showed that LMs that learned multiple types of link information achieve better downstream multi-hop QA performance. The experiments to show the effectiveness of learning relation semantics are still in progress. language of the presentation: English | |||
| WATTANAKRIENGKRAI SUPATSARA | D, 中間発表 | ソフトウェア工学 | 松本 健一, 飯田 元, 石尾 隆, Raula Gaikovina Kula |
|
title: Understanding The Role of Dependency Updates in Sustaining Software Ecosystem
abstract: Popular adoption of third-party libraries for contemporary software development has led to the creation of large inter-dependency networks, where sustainability issues of a single library can have widespread network effects. These dependencies need to constantly update , in response to breaking changes or apply critical security patches from elsewhere in the ecosystem. To understand the role of dependency updates in sustaining software ecosystem, I conduct two studies with a specific focus on two critical aspects (1) sustained contributions and (2) security. In this colloquium, I present these two studies, their progresses, and my future plan for doctoral course. language of the presentation: English | |||
| 栄木 誠 | D, 中間発表 | ディペンダブルシステム学 | 井上 美智子, 中島 康彦, 新谷 道広 |
|
title: *** Improving Efficiency and Robustness of Gaussian Process Based Outlier Detection via Ensemble Learning ***
abstract: *** Although automotive semiconductors must comply with the standard dynamic part average testing (DPAT) defined by the Automotive Electronics Council, it remains challenging to detect outliers that deviate from the spatial trend within a wafer. Outlier detection using Gaussian process (GP) regression has recently been proposed and outperformed DPAT. However, the detection performance degrades when faulty large-scale integrations are densely included in the regression. Furthermore, the applicable test items are limited because of the long computation time for regression. We propose an outlier detection method by applying ensemble learning to GP regression for simultaneously improving the detection performance and shortening the learning time. Experimental results on industrial production test data demonstrate that the proposed method improves the robustness against latent faulty chip detection by 15.6% while reducing the computation time by 98.6% compared with the conventional GPbased method. *** language of the presentation: *** Japanese *** | |||
| MUHAMMAD RADIFAN FITRACH | D, 中間発表 | 大規模システム管理 | 笠原 正治, 安本 慶一, 笹部 昌弘 |
|
title: A Trust Modeling Approach in Analyzing and Evaluating Social Networking Service Users in Exchanging Information.
abstract: In recent decades, social networking services (SNS) have become a powerful information exchange platform worldwide. Along with its convenience and flexibility, exchanging information via SNS has a lot of issues, including news credibility and reliability. However, deciding which users can be trusted and which are not is getting tough. We propose a trust model based on the big five personality traits to understand how news is disseminated among the SNS and evaluate SNS users' trustworthiness. We form this into an agent-based simulator, assessing how the trust model with the big-five personality traits capture the users' trust and their behavior in the SNS. language of the presentation: English | |||
| 笹田 大翔 | D, 中間発表 | サイバーレジリエンス構成学 | 門林 雄基, 安本 慶一, 笠原 正治, 妙中 雄三 |
|
title: Practical Pivacy-Enhancing Technology to Support Spatio-Temporal Data Collection by Adapting to Spatio-Temporal Characteristics and Protection Preference Heterogeneity
abstract: Spatio-temporal data are utilized in various applications such as epidemiology, natural disaster management, and urban planning. However, there is a risk of sensitive information, such as personal residences or workplaces, leaking from the collected data stored on servers. Local Differential Privacy (LDP) based data collection is a promising technique for protecting sensitive information. By modifying each data point in a way that server cannot distinguish it from others, privacy can be preserved. However, when applying LDP to spatio-temporal data, excessive or insufficient noise addition can compromise either data utilization or privacy protection. LDP determines the privacy protection strength (privacy strength) based on the statistical characteristics of the entire dataset under consideration and adds an appropriate amount of noise to individual data to make them statistically indistinguishable. On the other hand, spatio-temporal data exhibit different statistical characteristics depending on location and time, and the data characteristics change as time progresses. Therefore, simply applying LDP may result in excessive or insufficient noise addition depending on the location and time, making it difficult to achieve an optimal amount. In other words, excessive noise degrades the quality of the data, making data utilization challenging, while insufficient noise may leave residual privacy information, jeopardizing privacy protection. To address this issue, this study designs and implements a method to control the privacy protection strength based on temporal and spatial trends. By adjusting the privacy strength according to the constantly changing nature of spatio-temporal data, a consistent level of privacy protection is ensured. language of the presentation: Japanese | |||
| MOHAMMAD HAFIZ HERSYAH | D, 中間発表 | サイバーレジリエンス構成学 | 門林 雄基, 笠原 正治, 林 優一, 妙中 雄三 |
|
title: Addressing Multi Human-System Perspectives of Cloud Computing Research
abstract: A business resiliencies in cloud computing focus on Confidentiality, Availability, and Integrity toward diverse aspects should be rigorously and holistically addressed. Our approach toward fulfilling the purpose comprises several focuses: first, conducting a risk assessment focusing on assets-oriented aspects. Earlier publications only focus on implementing single guidelines, and subsequently, the attack scenario is based on simulation. We must comprehensively assess our assets, threats, and vulnerabilities to address gaps effectively. We utilized methodologies from ISO 27K family, E-Bios Risk Manager, MITRE ATT&CK, and NIST to ensure the utmost protection against potential risks and threats. As for the attack scenario, we referenced a threat group called teamTNT to precisely describe what a basic risk assessment should be done and anticipate best practices for a risk assessment activity. Secondly, the awareness of malicious insider threats due to the recent shift in the business process strategy to mitigate the cloud offers considerable flexibility that sometimes may lead to hostile altercations, such as abuse of authority that can activate damage in the form of intellectual property theft, sabotage, or fraudulent activities. Several existing kinds of research are limited in empirical testing/validation and utilize inaccurate psychometrics traits that fail to consider individual differences in personal-related behavior. We attempt to propose a multi-model insider threat by using pathological personality traits such as Dark Traits Personality to correlate with insider threat events. Thirdly, through Cloud Continuity and Disaster Recovery, the formulation of crucial action strategies towards uncertain situations/incidents in the cloud to protect assets and to continually provide the established service. It is essential to conduct a comprehensive risk assessment that analyzes threats and vulnerabilities to ensure the availability of existing technology through proper initial mitigation. Previous publications mostly separate the cloud continuity plan dan disaster recovery activity, which would trigger limited scope to identify disruptive, even-driven events. Lastly, the Digital Cloud Capability and Maturity Dimension as the upstream points from the previous three activities. There are no sufficient scientific publications that endeavor the mapping between cloud capabilities against the maturity model. Moreover, the current era of Industry 5.0 requires sustainabilities, resiliency, and human-centered design, which will be the end point of our research. language of the presentation: English | |||