Yamashita Haruka

Collaborative Metric Learning (CML) is a recommendation model based on implicit data, i.e. behavior history such as clicks and browsings. CML learns an metric space to embed not only the relationship between users and items, but also the similarity between items and that between users. Moreover, CML recommends the items which are close to each user in the trained embedding space. However, CML tends to learn by focusing on items that are popular among many users, and the accuracy of embedded representations of other minor items is often neglected. On the other hand, it is necessary to learn embedded representations of many minor items that match the user's preferences with high accuracy in order to provide unexpected recommendations that users may not have recognized. In this study, we propose a method to learn the embedded representations that capture user's preferences by weighting according to the number of observations of implicit data, and to make unexpected recommendations that include minor items. Finally, we apply the proposed method to actual movie evaluation data set, and show the usefulness of the proposed method in making unexpected recommendations based on the users' preferences.

In recent years, many e-commerce cites have been accumulating data on users' purchase histories and comment of products and stores. By utilizing such data and appropriately analyzing the user's behavioral history, effective marketing can be conducted, such as understanding the market and introducing a recommendation system customized for each user. In general, a network among users is constructed on the Internet called social network, and it is thought that there is a tendency in preferences depending on the structure of the network. Therefore, when analyzing user behavior, considering not only the behavior of each user, but also the relationships among users should be desirable. In this study, we integrate these approaches and propose a behavior analysis model that considers relationships among users. Specifically, by using Graph Attention Network, we construct a graph that considers the influence of surrounding users who are connected to each other. By extracting the characteristics of the subgraphs in the proposed analytical model, we can represent the behavior of user exactly. Furthermore, by analyzing the actual data, we show that the user's preferences and the relationships among users are properly represented.

Due to the accumulation of browsing history data on EC sites, Web marketing techniques are of growing significance. Most previous studies analyzed differences in overall browsing pages between purchasing and non-purchasing sessions by constructing a discriminative model and proposed measures for all users. However, it is difficult to utilize this model when considering personalized measures for each user. In this situation, a generative model, which infers browsing-behavior conditioned by whether a user purchases or not, is effective. Conditional VAE infers the data from the label and features of input data. In this paper, we apply Conditional VAE to browsing history data and identify important pages by generating a pseudo session assuming that a user in a non-purchasing session purchases. We propose a method to analyze important browsing pages that contribute to each user's purchase. We clarify the effectiveness of our proposed method by using real browsing history data.

Visualizing social relationships by a network is useful for understanding the behavior of groups and individuals. The target of this study is a network between employees in the workplace. The construction of this network enables us to understand human relationships and managing a team. To build this network, the questionnaire and e-mail data were conventionally used. However, in this work, we use conversation history data on a chat application(Slack, etc.). We propose a method of quantifying the relationship between employees from conversation data on a chat application and visualizing it as a network between employees. Specifically, we assume that strongly related employees will make remarks at adjacent times on the chat, quantify the relationship by multivariate hawkes process and build a network model. To verify the effectiveness of the proposed model, we used Slack conversation data of a real company and extracted knowledge about team management from the network.

In recent years, the importance of recommendation system has been increasing from the development of information technology. One of the important technologies for the recommendation is collaborative filtering. In this study, we focus on EM-NMF which is an effective model for the collaborative filtering. The approach is based on matrix decomposition. Generally, evaluation values by users are biased to some number of items. therefore, EM-NMF tends to learn emphatically to items with many evaluations. The prediction accuracy of evaluation for items with a small number of evaluation data tends to be undesirable. In this study, we propose a method to assemble two matrices; (i)predicted evaluation matrix based on the approach of items with many evaluation oriented and (ii)the matrix based on the approach of items with small number of evaluation obtained. This approach is expected to improve the prediction accuracy for the evaluation.

With the development of information technology, a huge amount of users' action history data has been accumulated on web sites.On such background, recommender system making use of these rich data has become important tool for searching contents or products. Diversifying the recommendation lists in recommender systems could potentially satisfy users' needs. In a previous research, the diversity is raised by the topic diversification method using Latent Dirichlet Allocation, but since the items belonging to the same topic are not diversified, there is a high possibility that they are similar. Therefore, this reserach proposes a recommendation method considering item diversification. Experimental results on MovieLens datasets demonstrate that our approach keeps accuracy produces more diversified results.

近年，データベースやインターネットテクノロジーの進化により顧客がいつ，どこで，何を買ったのかについての詳細なデータを蓄積することが可能となった．このようなデータを活用した解析の中には，優良顧客をどのように獲得するのか，どのような施策を講じれば，離反顧客になることを抑止できるのか，といった様々な観点からの解析が存在する．本研究では，某小売店の購買履歴データを分析対象とし，顧客ごとに存在する会員ステージに着目する．非優良顧客と優良顧客にはどのような購買傾向の違いがあり，顧客を成長させるために重要度が高い商品は何か，をクラスタ分析及び重要度分析から明らかにすることで，顧客の成長への示唆が与えられる．

近年，購買履歴データから商品毎の売上パターンを分析し，マーケティング活動に結びつけようとする取り組みが多々なされている．その中で，食品などを扱う小売店においては，気象条件が売上に強い影響を与える商品も多く，その関係性を考慮した分析が望まれる．本研究では，購買時刻とアイテム情報のデータを用いて，曜日効果及び季節性を取り除いた日付ごと，商品カテゴリごとの売上個数を要素とする行列を生成し，この行列表現されたデータの中に潜在するパターンの抽出を非負値行列因子分解(NMF)の適用により，気象要素による購買パターンの抽出を行い，ある小売店のデータを対象とした実験により，その有効性を示す．

In statistics,there are many studies on principal points.The concept of principal points which is proposed by Flury allows us to carry out such an analysis in a variety of applications and also properties of principal points have been studied. Although principal points of a multivariate distribution have widely studied,there is no discussion of principal points for a multivariate binary distribution.<BR> Yamashita and Suzuki have define the principal points for a multivariate binary distribution. Since principal points for a multivariate binary distribution are selected from multivariate binary region,there is a problem of the amount of calculation,since this problem is an NP-hard problem. Yamashita and Suzuki have shown the submodularity of principal points for a multivariate binary distribution and proposed an approximation method based on the greedy algorithm.Using the property of submodularity of principal points for a multivariate binary distribution,the accuracy of approximations is at least(1-1/e)times the optimal solution proved by Nemhauser et al.Finally, we show the result of an application of the methods to questionnaire survey data.

Analysis with Principal Points is a useful statistical tool for summarizing large data. Principal Points is defined as several representative binary patterns of given multivariate binary data in the binary region. The problem of finding Principal Points is NP-hard in general. In this study, we formulate Principal Points for multivariate binary data as the p-median problem and propose a subgradient-based method. This enables us to find a globally optimal set of Principal Points or to evaluate upper and lower bounds of a solution hi the middle of the calculation. We investigate the applicability of the proposed framework with real-world data.