大島 裕明

本研究では，事故の概要を説明したテキストを入力として，当事者が受けた傷病の程度を推定する手法を提案する．入力の対象とするテキストは，数文程度の文書を想定している．機械学習による分類問題を解くことで，その入力に該当する傷病の程度を推定するというのが提案手法の構成となる．本研究で利用するデータは，事故情報データバンクシステムで公開されている事故データである．入力として用いるのは「事故の概要」項目に記載されたテキストである．提案手法では，入力テキストを汎用言語モデルBERTを利用して分散表現として表現する．BERTのモデルとしては，日本語Wikipediaを用いて学習された事前学習モデルを用いる．しかし，傷病の程度を推定するというタスクの正解率を向上させるために，四つの工夫，（1）クラスウェイト，（2）Ordinal Classification，（3）マルチタスクラーニング，（4）トークンラベル推定による追加学習モデル，を導入する．これらの工夫を用いる場合と用いない場合において，傷病の程度の推定の正解率やMacro F1，RMSE，混同行列による評価にどのような影響が出るかを検証した．その結果，（1）クラスウェイト，並びに，（2）Ordinal Classificationを導入した際に，Macro F1の向上とRMSEの改善が得られるという結果となった．また，（3）マルチタスクラーニングを導入した際に正解率の向上が見られた．

本研究では，提示型検索モデル（Ostensive Search Model）に基づくインタフェースによって鑑賞者個人の興味を反映したミュージアム体験を可能にする電子ガイドを提案し，そのログを分析することで実現可能になった事前学習，事後学習支援システムについても提案する．我々は，国立民族学博物館（みんぱく）の展示物のうち3,053点について，展示物の解説やビデオなどの詳細情報を検索し，メモなどのアノテーションを付与できるiPad用アプリケーションである「みんぱくガイド」を作成した．みんぱくガイドは，鑑賞者が新しい展示物に気付いたり，興味を明確化することができるように，一覧性の高い検索結果画面を中心に情報探索を繰り返せるインタフェースを持っている．このような電子ガイドの操作履歴や位置情報などのログを用いることで，個人のミュージアム体験を色濃く反映した事前・事後学習支援を可能にした．事前学習支援システムでは，ミュージアムに行く前にカードを整理しながら鑑賞計画を立てるウェブアプリケーションの利用を通じて，鑑賞者に自分が何を学びに行くかという鑑賞軸を自覚してもらう．また，事後学習を促す仕組みとして，ログから鑑賞者が興味を持った展示物を推定し，後から鑑賞体験を思い出しやすくするためのパーソナライズされたポストカードを自動生成するシステムも作成した． This paper proposes an electronic guide application that enables visitors to get a museum experience that reflects their individual interests through an interface based on an Ostensive Search Model. In addition, we propose a pre-learning and post-learning support system that can be connected to our electronic guide. The system is based on a search result screen with a high level of browsability. We propose a system that allows users to search for detailed information such as explanatory texts and videos of exhibits they are interested in and add annotations such as handwritten notes. We created the “Minpaku Guide,” an iPad application that allows users to search for detailed information such as explanatory text and videos on exhibits of interest and add annotations such as notes. We also developed pre-learning support systems connected to our guide application. As a pre-learning support system, we created a web application that allows users to manually organize cards before visiting the museum to clarify what the visitors will learn beforehand. As a post-learning support system, we implemented a system that summarizes the visitor's operation log of the guide application into a postcard that helps to recall the viewing experience later.

This paper introduces a new test collection for ad-hoc dataset retrieval, which have been developed through a shared task called Data Search in the fifteenth NTCIR. This test collection consists of dataset collections derived from the US and Japanese governments' open data sites (i.e., Data.gov and e-Stat), as well as English and Japanese topics for these collections. Organizing the shared task in NTCIR, we conducted relevance judgments for datasets retrieved by 74 search systems, and included them in the test collection. In addition to the detailed description of the test collection, we conducted in-depth analysis on the test collection, and revealed (1) what techniques were used and effective, (2) what topics were difficult, and (3) large topic variability in the dataset retrieval task.

In this study, we tackle the problem of retrieving questions from a corpus archived in a Community Question Answering service that a consultant having distress can feel empathy with them. We hypothesize that the consultant feels empathy with the questions having a similar situation with that of the consultant’s distress, and propose a method of retrieving similar sentences focusing on the situation of the distress. Specifically, we propose two approaches to fine-tuning the pre-trained BERT model so that the learned model better captures the similarity of the situation between distress. One tries to extract only the words representing the situation of the distress, the other tries to predict whether the two sentences show the same situation. The data for training the models are gathered by the crowdsourcing task where the workers are asked to gather the sentences whose situation is similar to the given sentence and to annotate the words in the sentences that represent the situation. The data is then used to fine-tune the BERT model. The effectiveness of the proposed methods is evaluated with the baselines such as TF-IDF, Okapi BM25, and the pre-trained BERT. The results of the experiment with 20 queries showed that one of our methods achieved the highest nDCG@5 while we could not observe any significant differences among the methods.

In this study, we propose a method to predict whether a web searcher will purchase a camera in a near future based on his/her web search log. With the increasing popularity of online shopping at EC sites, more and more users are searching for products through web searches and actually purchasing them at EC sites. This indicates that, by analyzing the query log of a searcher, it is possible to predict whether the searcher will purchase the product in the near future. Therefore, we construct a classifier by collecting past web search query logs of searchers who have purchased cameras and those who have not purchased them. In the experiment, we used a web search query log of Yahoo! JAPAN and the product purchase histories of Yahoo! JAPAN Shopping to verify the results. We collected thousands of users who purchased cameras in a certain period and other users in the same number who didn’t purchase but issued queries related to cameras. By analyzing the classifier trained with the prepared dataset, we verify the accuracy of the prediction, the period of time required for the prediction, and whether there are any characteristic words that suggest the purchase.

In this study, we developed an Internet of Things (IoT) monitoring device to monitor over the people inside a room.We collected sen-sor data at a specific location using the device. Based on the data, we tried to predict the behavior of the person at that location. Mon-itoring and predicting human daily behavior is trivial task. Most of the research on monitoring and predicting daily life behavior are based on the data available from smart home [7] [16] [17]. But smart home is expensive compare to normal home, as different kind of sensor are attached in the room and have more facilities. So, we developed a low cost IoT monitoring device and predict the daily life behavior of human from the sensor data taken from the device.We can extract information from the daily life behavior and share it with the family living in distant places.

In this study, we propose a method to classify genre-based films using film screenplays. The proposed method vectorizes films into two aspects, i.e., film content (i.e., what films tell viewers) and screenplay structure (i.e., how the films narrate stories), and classi-fies film genres using the support vector machine method. We ap-plied the Doc2Vec algorithm to screenplay structure and to handle film content. In film production, for vectorizing films, we used the statistics of the four screenplay elements: Scene, action, dialogue, and transition. Compared with baseline methods, the evaluation showed that the proposed method is better for classifying films of specific genres.

In this paper, given the subject-verb pairs, we propose a compu-tational model to express the difference of the meaning of a verb when the subject has changed. We propose a method to gener- A te metaphorical expressions consist of subject-verb pairs from the model. "Airship swims" is one of the example. It is the expression about the event that an airship flies in the sky gracefully. There are a few reasons why the expression is accepted for people. "airship flie" and the motion of a sea creature, for example "whale swims," represent both "the normal move in a space" and there is a simi-larity. Given the input ("airship," "fly"), we propose a method to detect a verb "swims" to generate metaphorical expressions consid-ering these similarity. At first, we test which vectorization method is the best as the vectorization of a subject-verb pair. We calcu-late a transformation matrix to conserve between the meaning of (non-human subject, verb) pairs and the meaning of ("man", verb) pairs. We calcurate the transformation matrix between them us-ing the stable meaning verbs as the anchors. In this paper, we test an hypothesis that we can use these transformation matrices to find an appropriate verb considering the difference of the mean-ing occured from the subjects.We gather 67 cases of target figura-tive expressions fromWeb.We evaluated the proposed method by defining the information retrieval problem of verbs.

We proposed a search method with a query by example in a known domain for information in an unknown domain. It seems natural the relation that 'Pikachu in 2000s' which is similar to 'Dumbo in 1980s.' Pikachu and Dumbo are the popular animation characters in these decades. For example, given three entities, such as Pikachu, 2000s and 1980s, we search Dumbo by the relation 'Pikachu in 2000s as Dumbo in 1980s.' The proposed method is based on transformation matrices to bridge word vector spaces. Especially, we focus on transitive relations such as the relation that 'Dumbo in 1980s' is similar to 'Garfield in 1990s' and to 'Pikachu in 2000s' to solve the problem. Experimental results show that the proposed method has some advantage when there are a little common words between source and target terms.