| ATUHURRA JESSE | D, 中間発表 | 自然言語処理学 | 渡辺 太郎, 荒牧 英治, 進藤 裕之 |
|
title: *** RoboNER: a Dataset of Rare Named Entities in Human Robot Interaction via Wikidata Knowledge Graph Triples*** abstract: *** Human robot interaction (HRI) aims to support active dialogue between humans and robots. HRI systems typically include an intent recognition (IR) module to enable the robot to understand what the human is talking about. IR involves detection of terms related to the topic of discussion. Hence, IR is similar to named entity recognition (NER), and NER datasets are candidates to develop HRI/IR systems. However, existing NER datasets are not suited to applications, such as HRI, since HRI requires many NE specific to the domain. Hence, the need to develop a new NER dataset. This paper addresses the problem of rare NE encountered in HRI, and introduces RoboNER, a fully human-annotated NER dataset which comprises nine NE. First, to increase NE coverage, we applied distant supervision. We acquired knowledge related to six NE namely; Drink, Food, Hobby, Job, Pet, and Sport in form of triples from Wikidata knowledge graph. Second, we augmented the six NE with the CoNLL++ dataset by re-annotating dev & test sets resulting in three additional NE: Country, PeopleName, and Organization. Consequently, RoboNER represents nine entity types. RoboNER contains 15K sentences and 257K tokens. The proposed dataset enables us to detect rare NE, and our results on RoboNER indicate how valuable this dataset is. *** language of the presentation: *** English *** | |||
| 岡井 光輝 | M, 2回目発表 | ソフトウェア工学 | 松本 健一, 中島 康彦, 石尾 隆（客員教授）, Raula Gaikovina Kula, 嶋利 一真 |
|
title: An Analysis of Single-Letter Variable Usage Criteria with a Focus on Type and Scope
abstract: Variable names affect the readability and understandability of source code. Therefore, it is important that the name of a variable completely and accurately represent the variable's entity. However, in actual software development, single-letter variable names are often used in limited cases such as local variables. Although the standard naming conventions are useful for consistency in variable naming, they do not specify specific conditions under which single-letter variables are acceptable. In this study, we extracted variable names from eight popular Java GitHub projects and investigated the usage of single-letter variables using scope and type as indices in order to demonstrate the criteria for their use. The results showed that the proportion of single-letter variables in primitive types and exception classes tends to be larger than that in other types, and that the scope of single-letter variables tends to differ depending on the type. By using these trends as indicators for use as single-letter variables, standardized use of single-letter variables becomes possible. language of the presentation: Japanese 発表題目: 型とスコープに着目した一文字変数の利用基準の分析 発表概要: 変数名はソースコードの可読性・理解度に影響を与える．そのため，変数の命名においてその変数の実体を完全かつ正確に表す名前であることが重要とされている．しかし，実際のソフトウェア開発ではローカル変数などの限定的な場合で，一文字の変数名も多く使われている．標準的な命名規則は変数の命名に際して一貫性を持たせることに役立つが，一文字の変数として使うことが容認される具体的な条件については示されていない．そこで本研究では一文字変数の利用基準を示すために，Javaの8つの人気GitHubプロジェクトから変数名を抽出し，スコープと型を指標として一文字変数の利用状況を調査した．その結果，プリミティブ型および例外クラスにおける一文字変数の割合はその他の型に比べて大きい傾向があること，型によって一文字変数のスコープは異なる傾向を持つことが明らかになった．これらの傾向を一文字変数として使う際の指標とすることで，一文字変数の標準的な利用が可能になる． | |||
| ROJPAISARNKIT RUKSIT | M, 2回目発表 | ソフトウェア工学 | 松本 健一, 安本 慶一, 石尾 隆（客員教授）, Raula Gaikovina Kula, 嶋利 一真 |
|
Title: Characterising developer workloads during vulnerability fixing Abstract: With the urgent need to secure supply chains among Open Source libraries, attention has been focused on mitigating vulnerabilities are detected for these libraries. Although such awareness have improved over recent times, most studies still report lags in the mitigation process. This suggests that developer still have to deal with other contributions that occur during the period of fixing vulnerabilities (i.e., coinciding Pull Requests (PRs) and Issues) remains unknown. To characterize these contributions, we conducted a mixed-method empirical study to mine NPM GitHub projects that were affected by 554 different vulnerability advisories. We mined a total of 4,699 coinciding PRs and Issues to answer three research questions. Language of the presentation: English | |||