To correctly evaluate the effects of treatments, conducting randomized controlled trials (RCTs) is a reasonable approach. However, because it is generally difficult to implement RCTs for all treatments, methods to estimate the treatment effects using observational data have been actively studied and used in various decision-making processes. Observational data accumulated in business activities and elsewhere contains the results of various previously implemented treatments, and correctly estimating the effects of any given treatment without separating the impacts of other treatments is challenging. Against this background, this paper proposes a method to estimate the effects of multiple treatments of various types while considering various causal relationships. Specifically, the proposal is a variational inference method that estimates the effect of multiple treatments using four latent factors estimated from observations, making assumptions that are independent of the type and number of treatments. The proposed method makes it possible to appropriately estimate the effects of measures even in situations with complex causal relationships. In addition, in situations where measures with continuous parameters are being implemented, it is possible to estimate the effects of measures that have not been implemented in the past by treating the content of the measures as a continuous variable.

統計的因果推論に基づくビジネス施策の効果検証では，条件付き平均処置効果（以下，CATE）と呼ばれる，処置の有無やアウトカムに関連のある変数が同じ顧客群における処置効果の推定が有用であることが多い．このCATEを推定するための基本的な手法であるCausal Treeは解釈性が高く，観察研究においても適応可能であるため非常に有用である．ただし，Causal Treeを観察研究で用いる際には顧客を人為的に選択することによる系統的な誤差（以下，選択バイアス）を除去する必要があり，傾向スコアを用いた改良手法であるCausal Tree-Transformed Outcome（以下，CT-TO）が提案されている．しかしCT-TOは，傾向スコアを推定するモデルが正しく指定されていない場合には，最終的に誤ったCATEの推定に結び付きやすいことなどの改善点がある．そこで本研究では，Causal Treeをベースとし，高い頑健性を持って選択バイアスが存在する状況に対応したCATE推定手法を提案する．具体的には，Doubly Robust Estimatorという，ある集団全体の処置効果を推定可能な手法を活用したCATEの推定に基づいて，Causal Treeの目的関数の導出を行う．これにより，CT-TOが有する高い解釈性を維持しつつ，選択バイアスが存在する状況において，より精度の高いCATEの推定が可能になる．最後に，人工データと実データに対して提案手法を適用し，その有効性を検証する．
In the effectiveness testing of business measures based on statistical causal inference, it is often useful to estimate the conditional average treatment effect (CATE). Causal Tree, which is the basic method for estimating CATE, is highly interpretable and can be adapted to observational studies, making it very useful. However, when Causal Tree is used in observational studies, selection bias needs to be removed, and an improved method using propensity scores, Causal Tree-Transformed Outcome (CT-TO), has been proposed. However, CT-TO has some points to be improved, such as the fact that it tends to lead to incorrect CATE estimation if a model for estimating propensity scores is not correctly specified. Therefore, this study proposes a Causal Tree-based CATE estimation method for situations where selection bias exists with a high degree of robustness. Specifically, the objective function of Causal Tree is derived based on the estimation of CATE using doubly robust estimator, a method that can estimate the treatment effect for an entire population. This enables more accurate CATE estimation in situations where selection bias exists, while maintaining the high interpretability of CT-TO. Finally, the proposed method is applied to artificial and real data to verify its effectiveness.

With the spread of EC sites, it has become an important work for companies to analyze user preferences contained in accumulated purchase history data and utilize them in marketing measures. A topic model is well known as a method for analyzing user preferences from purchase history data, and a model assuming hierarchy of topics has been proposed as an extension method. The previously proposed PAM (Pachinko Allocation Model) is a highly expressive model in which all upper and lower topics are connected by a network and the relationships between multiple topics can be analyzed. However, PAM is easily affected by the initial values of learning parameters, and it is difficult to obtain stable topics, so the interpretation of the estimated topics becomes unstable. It is dangerous to make business decisions based on the interpretation of such unstable results. Therefore, in this research, instead of using the hierarchy of topics estimated based on the user’s purchasing behavior, we use information with a hierarchical structure of “product categories” given by the EC site for managing items. Therefore, we propose a method that is useful for studying measures and that enables hierarchical topic analysis. Finally, the proposed method is applied to the evaluation history data of the actual EC site to analyze the user’s preference and show its usefulness.