深度學習在電商平臺中用戶行為預測之運用

隨著網路基礎設施和資訊科技的發展，日常生活和網路越來越息息相關，各企業為了強化競爭優勢與增加對網站使用者的了解，更加重視使用者在網路上的瀏覽足跡，在資料時代的現今社會，能從資料中萃取出有用的資訊，預測未來者，才得以掌握市場。本研究將重心放在使用者點擊行為上，過去文獻指出，有著類似點擊紀錄的使用者，會有相似的購買行為，在以進入官網的管道以分析重點下，本研究提出兩階段使用者行為分析模型，先將使用者分群，後預測該群體之行為，使得企業在分配資源時，能夠根據此模型，進而在進行資源預算分配時，得以最大化資源效益，找出獲利最高之分配策略。分析這些數據有一個終極的目標，就是提高使用者在網路上行為的轉換率，也就是增加使用者購買的機率。在實際有購買之前，使用者會在不同裝置上，和企業有各式各樣的接觸點，例如關鍵字廣告、社群廣告轉換等等為了要最佳化企業的資源分配，我們利用多點擊歸因模型的概念，分析各個接觸點的發生順序的影響，探索使用者行為，將此轉換成有商業價值的洞見。分析使用者的歷史點擊路徑，找出使用者的行為模式，用以預測使用者的下一個點擊路徑，接著將資源根據預測的機率去做分配，最大化企業的收益。

關鍵字

使用者行為；行為預測；多管道點擊； LSTM ；線上客戶旅程

並列摘要

The application of deep learning on E-commerce user behavior prediction has been widely used in the business world. Nowadays, customers can interact with firms through miscellaneous online ads on different channels easily. In other words, the customer now has innumerable options and limitless time to accomplish their commercial activities with firms, individualizing their own online customer journey. This kind of convenience emphasizes the importance of online advertisement allocation on different channels. Therefore, a profound understanding of customer behavior can make considerable benefit from optimizing fund allocation on diverse ad channels. To achieve this objective, multiple firms utilize a numerical methodology to create a data-driven advertisement policy. In our research, we aim to exploit online customer click data to discover the correlations between each channel and their sequential relations. We use LSTM to deal with the sequential property of our data and compare its accuracy with other non-sequential methods, such as decision tree, logistic regression, etc. Besides, we also classify our customers into several groups by their behavioral characteristics to perceive the differences between all groups as customer portrait. As a result, we discover distinct customer journey under each customer portrait. Our article provides some insights into marketing research and can help the firm to formulate online advertising criteria.

並列關鍵字

online customer journey ； behavior prediction ； LSTM ； K-Prototype

參考文獻

Asteriou, D. Hall, SG., (2011) ARIMA models and the Box–Jenkins methodology.Applied Econometrics,2(2),265-286

Google Scholar

Bishop, C., (1995) Neural networks for pattern recognition. United States: Oxford university press.

Google Scholar

Chan, G., Cheung, C., Kwong, T., Limayem, M. Zhu, L., (2003) Online Consumer Behavior: A Review and Agenda for Future Research.Paper presented at the 16th Bled eCommerce Conference eTransformation, Slovenia.

Google Scholar

Chen, J. Stallaert, J., (2014) An economic analysis of online advertising using behavioral targeting.MIS Quarterly,38(2),429-449

Google Scholar

Chen L., Tao Z. Yuan C., (2019) Customer Purchase Intent Prediction Under Online Multi-Channel Promotion: A Feature-Combined Deep Learning Framework. IEEE Access,7,112963-112976

Google Scholar

國際替代計量

深度學習在電商平臺中用戶行為預測之運用

未授權

主題瀏覽