Search

Class imbalance is commonly observed in real-world data, and it is problematic in that it degrades classification performance due to biased supervision. Undersampling is an effective resampling approach to the class imbalance. The conventional undersampling-based approaches involve a single fixed sampling ratio. However, different sampling ratios have different preferences toward classes. In this paper, an undersampling-based ensemble framework, MUEnsemble, is proposed. This framework involves weak classifiers of different sampling ratios, and it allows for a flexible design for weighting weak classifiers in different sampling ratios. To demonstrate the principle of the design, in this paper, a uniform weighting function and a Gaussian weighting function are presented. An extensive experimental evaluation shows that MUEnsemble outperforms undersampling-based and oversampling-based state-of-the-art methods in terms of recall, gmean, F-measure, and ROC-AUC metrics. Also, the evaluation showcases that the Gaussian weighting function is superior to the uniform weighting function. This indicates that the Gaussian weighting function can capture the different preferences of sampling ratios toward classes. An investigation into the effects of the parameters of the Gaussian weighting function shows that the parameters of this function can be chosen in terms of recall, which is preferred in many real-world applications.

Data exploration is a fundamental user task in the information seeking process. In data exploration, users have ambiguous information needs, and they traverse across the data for gathering information. In this paper, a novel data exploration system, called FPX-G, is proposed that uses virtual reality (VR) technology. VR-based data exploration (or immersive analytics) is a recent trend in data analytics, and the existing work approaches involve aggregated information in an interactive and 3D manner. However, exploration for individual pieces of data scarcely has been approached. Traditional data exploration is done on 2D displays, therefore space is limited, and there is no depth. FPX-G fully utilizes 3D space to make individual piece of data visible in the user’s line of sight. In this paper, the data structure in FPX-G is designed as a graph, and the data exploration process is modeled as graph traversal. To utilize the capability of VR, FPX-G provides a first person view-based interface from which users can look at individual pieces of data and can walk through the data (like walking in a library). In addition to the walking mechanism, to deal with limited physical space in a room, FPX-G introduces eye-tracking technology for traversing data through a graph. A simulation-based evaluation reveals that FPX-G provides a significantly efficient interface for exploring data compared with the traditional 2D interface.

We propose a method that assists legislative drafters in locating inappropriate legal terms in Japanese statutory sentences and suggests corrections. We focus on sets of mistakable legal terms whose usages are defined in legislation drafting rules. Our method predicts suitable legal terms using a classifier based on a BERT (Bidirectional Encoder Representations from Transformers) model. We apply three techniques in training the BERT classifier, specifically, preliminary domain adaptation, repetitive soft undersampling, and classifier unification. These techniques cope with two levels of infrequency: legal term-level infrequency that causes class imbalance and legal term set-level infrequency that causes underfitting. Concretely, preliminary domain adaptation improves overall performance by providing prior knowledge of statutory sentences, repetitive soft undersampling improves performance on infrequent legal terms without sacrificing performance on frequent legal terms, and classifier unification improves performance on infrequent legal term sets by sharing common knowledge among legal term sets. Our experiments show that our classifier outperforms conventional classifiers using Random Forest or a language model, and that all three training techniques contribute to performance improvement.

This paper proposes a new search method that supports legislative duties in Japanese local governments where both expert and non-experts draft local ordinances. Since such legislation requires comparison with the ordinances of other local governments, our purpose is to introduce a neural model and show how it effectively performs similarity search tasks for ordinances. Our experimental results show that our method outperforms the previous scheme for similarity search with a F-measure of 0.977. An additional experiment shows that our method is useful for analyzing the legislation of local governments.

The Japanese government has promoted dissemination of Japanese legal information to the world, and it has released English translations of Japanese statutes as well as a Japanese-English bilingual dictionary for statutory terms. However, the number of these translations is insufficient. The bilingual dictionary, the Japanese-English Standard Legal Terms Dictionary, contains only 3782 entries. To expand this key reference work, we focused on the Japanese Official Gazette, English Edition, which was published from 1946 to 1952 under order of the Supreme Commander for the Allied Powers. We conducted an experiment on targeted phrase extraction to acquire legal translations and confirm the usefulness of the Japanese Official Gazette, English Edition.