近幾年電子商務公司在特殊節日的銷售額大於平日，然而從使用者瀏覽網頁的歷史紀錄中，我們發現僅有並非所有使用者在節日間皆比平時更常造訪電商網站，此現象表示銷售額的提升可能只是來自少數使用者所產生的影響，因此電子商務公司若能透過系統化的方式判別使用者行為，並給予不同行為表現使用者相對應的行銷手段，便能夠使市場行銷策略發揮更大的效能。

我們提出 Matrix Factorization based Multi-objective Method，當同時有多個待預測的目標時，本模型能有效利用多個預測目標之間可能存在互相影響的隱性因子。相較於為每個目標分別建立獨立的模型，本方法能有效減少模型需學習之參數，因此在訓練樣本數受限的情況，依然能達到有效的訓練結果。我們使用此方法同時預測使用者於特殊節日期間在不同類型網頁的行為變化，結果顯示：本方法在大多數時候能勝過單目標之訓練模型。

It is reported that sales by e-commerce companies were greater than usual during shopping holidays and festivals. However, based on users’ browsing logs, we found that not all users visit e-commerce websites more often than they normally do during holidays. Therefore, the increase in sales may come from the purchase behaviors of a small number of users. If the e-commerce companies can systematically assess and analyze user behaviors, they might be able to apply customized marketing method to maximize the effectiveness of their sales strategies.

This study proposes a matrix factorization based multi-objective method, which effectively uses the latent variables that represent possible interactions among multiple targets. Compared with establishing separate models for each target, this method can effectively reduce the parameters that the model needs to learn, and can, therefore, achieve an effective training outcome even when training samples are limited. We use this method to simultaneously predict users’ behaviors on different types of web pages during shopping holidays and festivals. The results show that this method can outperform the single target training model most of the time.

[1] “E-Commerce Continues To Be The Bright Spot For Holiday Sales,” https://www.forbes.com/sites/shoptalk/2016/12/27/ecommerce-continues-to-bethe- bright-spot-for-holiday-sales/28b1f6bb2780, accessed: 2017-06-15.
[2] “Alibaba’s Singles’ Day: What We Know About The World’s Biggest Shopping Event,” https://www.forbes.com/sites/franklavin/2016/11/06/alibabas-singles-daywhat- we-know-about-the-worlds-biggest-shopping-event/687e1e636da7, accessed:
2017-06-15.
[3] C. Lo, D. Frankowski, and J. Leskovec, “Understanding behaviors that lead to purchasing: A case study of pinterest,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. ACM, 2016, pp. 531–540.
[4] D. Kempe, J. Kleinberg, and É. Tardos, “Maximizing the spread of influence through a social network,” in Proceedings of the ninth ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2003, pp. 137–146.
[5] J. Leskovec, A. Singh, and J. Kleinberg, “Patterns of influence in a recommendation network,” in Pacific-Asia Conference on Knowledge Discovery and Data Mining. Springer, 2006, pp. 380–389. 32
[6] Y. Koren, R. Bell, and C. Volinsky, “Matrix factorization techniques for recommender systems,” Computer, vol. 42, no. 8, 2009.
[7] T. K. Landauer, P. W. Foltz, and D. Laham, “An introduction to latent semantic analysis,” Discourse processes, vol. 25, no. 2-3, pp. 259–284, 1998.
[8] S. Ruder, “An overview of multi-task learning in deep neural networks,” arXiv
preprint arXiv:1706.05098, 2017.