コロキアムB発表

日時: 9月26日(木)5限(16:50~18:20)


会場: L1

司会: 磯山直也
日高 真人 D, 中間発表 ユビキタスコンピューティングシステム 安本 慶一, 中村 哲, 荒川 豊, 諏訪 博彦, 藤本 まなと
title:On-site Tourism Planning System Based on Dynamic Information of Tourism
Recently, due to the drastic increase of foreign tourists coming to Japan, it is demanded to provide smart tourism services that enable inbound tourists to enjoy comfortable sightseeing. To realize satisfactory tourism for tourists, it is desirable to provide tourist contents in a timely manner by considering the dynamic information which changes depending on time such as current congestion information in destination spots and travel route information, in addition to the static information such as preference and profile of tourists. However, in many existing studies, there are following serious problems such as 1) no support for on-site use, 2) no consideration of dynamic information, and 3) heavy burden on tourists. In previous efforts,we proposed a novel system that can provide tourism plans with high quality to tourists in a timely manner in spots that tourists do sightseeing in order to solve such problems. In addition to the above information, this presentation will take into account the user's personality for recommendations.
language of the presentation: Japanese
 
曽根田 悠介 M, 2回目発表 ユビキタスコンピューティングシステム 安本 慶一, 中村 哲, 荒川 豊, 諏訪 博彦, 水本 旭洋
title: Multi-Modal Sensing and Analysis for Gesture/Behavior on Group Discussion
abstract: In recent years, group communication skills in active-learning are regarded as important. In a group discussion, it is known that non-verbal communication such as nodding has a great impact on the conversation. In this study, we investigate how micro behaviors correlate with the satisfaction of participants, and try to evaluate the quality of communication among multiple people. In order to analyze micro gestures and behaviors during discussions, we created a data set for discussions with four people. This data set consists of three types of data: Sensor data (360-degree camera, inertial measurement device and eye gaze measurement device), Survey data for discussion, and Labeling data for defined actions. In this presentation, we will describe the statistics of the measured data and the direction of future work.
language of the presentation: Japanese
発表題目: 複数センサを使用したグループディスカッション中に発生する行動の分析
発表概要: 近年,アクティブラーニングに代表されるように複数人でのディスカッションといったコミュニケーション能力が重要視されている.このとき,頷きといった非言語でのコミュニケーションは,会話中において大きな影響を与えることが知られている.このような微小な行動が参加者の満足度とどのように相関性があるのか調査し,複数人でのコミュニケーションの質について評価を試みる.本研究では,4人でのディスカッションに関するデータセットの作成を行った.このデータセットは360度カメラ,慣性計測装置,視線計測装置のセンサーデータ,ディスカッションに関するアンケートデータ,定義した行動に関するラベリングデータの3種類のデータから構成されている.本発表では,実際に測定されたデータの統計の確認と考察に加え,今後の解析の方向性について述べる.
 
平野 陽大 M, 2回目発表 ユビキタスコンピューティングシステム 安本 慶一, 中村 哲, 荒川 豊, 諏訪 博彦
title: A Multi-objective Optimization Algorithm for User's Resources and Satisfaction for Tour-Route Decision Support
abstract: The tour route recommendation system is broadly considered as a growing research topic. However, many of the studies do not consider multiple user parameters such as stamina, time, money and satisfaction. Therefore, there is a possibility that a sightseeing route that does not consider cost performance is recommended. In order to solve these problems, there have been proposals for tourism recommendation systems that take into account the trade-off between one user resource and the degree of satisfaction obtained from tourism. In reality, however, users have a wide variety of resources and considering one of them is not enough. Hence, we propose a sightseeing route recommendation system considering multiple resources of users. In our research, we defined and formulate the route recommendation problem as a Multi-Objective optimization problem, and defined money, time, stamina and satisfaction as main elements of tourism. We employed an approach to solving this problem consisting of two phases: a global search and local search. In order to derive semi-optimal solutions in practical time to be used by tourists, we design a new heuristic algorithm to solve this problem and implement it for our route recommendation system. In conclusion, we confirmed that our proposed algorithm calculated 100 diversified solutions for Higashiyama area in Kyoto with 30 tour spots in 114.9 [sec].
language of the presentation: Japanese
 
鶴山 優季子 M, 2回目発表 ユビキタスコンピューティングシステム 安本 慶一, 中村 哲, 荒川 豊, 諏訪 博彦
title: Methods for improving rent estimation accuracy of real estate for restaurants
abstract: Rent estimation for real estate is widely performing with the developing of machine learning. However, there are many complex factors for restaurants, and it is difficult to estimate using only general factors such as location, footprint, year of build. Therefore, previous research has proposed a price estimation model based on three factors, such as static information, dynamic information, and latent information. However, the estimation accuracy was about 0.7, and higher accuracy is required for use in the real world. In this research, we propose a new rent estimation model for improving accuracy. Concretely, we add the new data, add new features, and remove noise. As a result, although our proposal could not increase the accuracy, we got some findings for modifying the model.
language of the presentation: Japanese
発表題目: 飲食店向け不動産物件の賃料推定精度向上に向けた手法の検討
発表概要: 機械学習の発展に伴い,不動産物件の賃料推定は広く行われている.しかし飲食店向け物件に関しては,多くの要因が複雑に絡んでおり,立地や間取りといった一般的な要因のみで価格を推定することは難しい.そこで先行研究では,静的情報,動的情報,潜在的情報といった3つの要因をベースとした価格推定モデルを提案した.しかしその推定精度は0.7程度であり,実世界で使用するには,より高い精度が求められる.そこで本研究では,推定精度向上のための新たな価格推定モデルを提案する.具体的には,データ数の追加,新たな特徴量の追加,ノイズの除去などを行なった.結果として,提案モデルでは推定精度の向上は得られなかったが,モデルを改善するための新たな知見を得ることができた.