應用層級貝氏理論於跨商品類別之顧客購買期間預測模型

Applying Hierarchical Bayesian Theory to the Multi-Category Inter-Purchase-Time Model

由於消費者需求日漸分歧，大眾媒體式微，利用資料庫進行一對一行銷已成為必然的趨勢。然而，過去以顧客購買期間為主要變數之學術研究鮮少討論到跨商品類別的議題，因此本研究的主要目的在於利用層級貝氏理論提出一跨商品類別之顧客購買期間預測模型。本研究以Generalized Gamma Distribution 來配適顧客購買期間變數，並利用Multiplicative Model 將商品類別變數引入過去之單一商品類別購買期間模型當中。本研究利用國內一型錄購物公司之顧客交易資料進行實證研究，得到以下結論：1. 本研究所提出之跨商品類別購買期間模型之預測命中率優於過去之單一商品類別購買期間模型。2. 跨商品類別顧客購買期間模型可以估計出不同商品類別對購買期間的影響參數，進而瞭解商品類別對購買期間的影響。

Because of the recent diversity of consumers’ demand and the less popularity of mass media, one-to-one database marketing has been utilized by companies to increase their competitive capability. The availability of data warehouses that combine demographic and behavioral information also encourages marketing managers to use database approaches to understand their customers and predict customer purchasing behavior. For virtual shopping industry, such as TV shopping companies, online shopping companies, and catalog shopping companies, the most desirable information to know is when a customer will make their next purchase. This is because those companies’ first priority is to maintain customers’ loyalty, and keep their customers making purchases. Therefore, the information of next-purchasing time can support strategic and marketing decisions; moreover, it can help companies save costs effectively. For deriving the desired information, inter-purchase time takes an important role, a useful variable, to estimate purchase frequency patterns for customers. Over the past few years, a considerable number of studies have investigated this issue using methodologies including decision trees, neural network, genetic algorithms, OLAP and statistical models; however, few of them consider the impact of multi-category of products on inter-purchase time. Under the multi-category condition, each product category has different purchasing period and properties. The price and the relation between different products, say, complementary or substitution, also have influence on customers’ inter-purchase time. Today, many companies sell a variety of product categories. Since the inter-purchase time of products may vary under different products, the single-category product model is not enough to support decision making in some business condition. Because investigation of multi-category inter-purchase time models has been inadequate, the aim of the present study is to analyze the system by building a one-to-one multi-category inter-purchase time model to improve the prediction ability of the single-category model. We also further explore the relationship between product category and inter-purchase time. Our proposed model is based on the Hierarchical Bayesian (HB) model. Bayesian methods are particularly appropriate for decision orientation in marketing problems and the HB model has been widely discussed in the past decade. It can model heterogeneity across customers and estimate a unique parameter value for each customer. Using a HB model, one-to-one marketing can be achieved even if only a small number of purchase records are available for some customers. In this research, the proposed multi-category inter-purchase time HB model is based on Generalized Gamma Distribution and multiplicative model formulations. With the use of Hazard rate function, the model is then used to obtain the purchase probability of each individual customer which can be used to derive the desired information. To validate the effect of the proposed model, field data from a local catalog company in Taiwan are collected. The model’s parameters are estimated through Markov chain Monte Carlo (MCMC) simulation method. Prediction hit rates by different models are compared. Based on the validation results, conclusions are drawn as follows: 1. The multi-category inter-purchase time model has better prediction hit rate than a basic model. 2. By using the multiplicative model, our multi-category model can estimate the influence of product category on customers’ inter-purchase time.