• Technical Name
  • 探勘金融消費資料於客戶消費行為預測與個人化電子廣告標題生成
  • Operator
  • National Yang Ming Chiao Tung University, Dept. of CS, ADSLab
  • Booth
  • AIoT&智慧應用 AIoT & Smart Applications
  • Contact
  • 陳宜勤
  • Email
  • ellie1007@nctu.edu.tw
Technical Description Our technologies are able to analyze customer behaviorsprovide personalized automatic services. We take two directions: (1) Establish a customer behavior predictionrecommendation system by analyzing consumption records, exploring behavior features,strengthening the link between marketing strategiesbehavior analysis (2) Collaborative EDM subjects generation by analyzing the relationship between customers’ click recordsconsumption for understanding the relationship between customer intentionsfinancial products,for generating personalized marketing strategies.
Scientific Breakthrough One of our research results has been published on the ACM WSDM, one under review on the ACM TIST,two on the AAAI 2020. In terms of benchmark of quantitative metrics, our behavioral analysisrecommendation models outperform random forestother base models by 40. The increase in the proportion of customer spending increased by 2.87 times. The conversion rate of promotion considering consumer behavior reached 48.95. The TemPEST model developed by us outperforms the BiSET model, which is also for the title generation, by 16. Our PORL-HG is considered to be more attractive by 63.1.
Industrial Applicability The development of our technologies is through cooperation with KKDAYE-Sun Bank,the use of its spatio-temporal database, clickscomments database, social interaction database, etc., to explore the technologies required for e-commerce engagement modules. By analyzing user browsing consumption recordsother available datasets to explore important features like customer preferencesbehavior patterns, our model is able to generate EDM content,to deliver to target customers at the right time,hence it allows to materialize the actual occurrence of purchase behavior.