依據中華民國經濟部統計處的資料顯示，近十年來消費者習慣的購物渠道已與往年不同，而企業必須與時俱進才能在市場上生存，因此如何有效分析消費者購買行為成為提高市場競爭力的重要課題。
　　本研究運用非監督式表徵學習方法，對臺灣零售業消費者購買序列資料進行深度分析。通過非負矩陣分解法（Non-negative Matrix Factorization, NMF）分析消費者購買行為，識別出不同的消費者群體及其購買偏好；進而應用霍爾特-溫特方法進行時間序列分析，預測商品分類的銷售趨勢。
　　經過實驗證明，藉由非負矩陣分解法（NMF）可成功劃分不同消費者群體，以深入分析其購買特性和行為模式。除此之外，霍爾特-溫特方法在本研究中展現其學習季節性和趨勢性商品銷售量的成效，對於優化庫存管理和制定行銷策略具重要意義，但在實務應用上仍需要考量到市場環境等實際狀況。
　　整體而言，研究證明非監督式表徵學習方法能夠有效分析零售業消費者的購買序列資料，為學術研究和產業應用提供實證基礎。非負矩陣分解法（NMF）結合霍爾特-溫特方法能深入洞察消費者行為，預測市場趨勢，為零售業者提供數據驅動決策的參考準據。

This study utilizes statistical data from the Ministry of Economic Affairs, R.O.C., to address the shift in consumer shopping channels over the past decade. It highlights the necessity for businesses to evolve to survive in the market. The focus is on effectively analyzing consumer purchase behavior, aiming to enhance market competitiveness through unsupervised representation learning methods.
The research comprises two parts: first, identifying diverse consumer groups and their preferences using Non-negative Matrix Factorization (NMF); second, forecasting sales trends of product categories using the Holt-Winters time series analysis method.
Results show that NMF successfully segments ten consumer groups, offering insights into their purchase patterns. Additionally, the Holt-Winters method predicts sales trends and seasonal changes in product categories, which are crucial for inventory management and marketing strategies. However, its practical application requires consideration of the market environment and other real-world factors.
Overall, the study demonstrates the effectiveness of unsupervised representation learning in analyzing retail consumer purchase sequences, providing valuable empirical evidence for academic and industrial applications. The integration of NMF and the Holt-Winters methods offers deep insights into consumer behavior and market trends, supporting data-driven decisions in retail.

論文審定書 i
誌謝 ii
摘要 iii
Abstract iv
目錄 v
圖次 vii
表次 vii
第一章 緒論 1
1.1 研究背景 1
1.2 研究目的 3
第二章 文獻探討 3
2.1 零售業普遍性使用的消費者購買行為分析法 4
2.1.1 顧客終身價值 4
2.1.2 RFM模型 4
2.1.3 LRFM模型 5
2.2 非監督式表徵學習之消費者行為分析 6
2.2.1 主成分分析法 6
2.2.2 非負矩陣分解法 7
2.3 時間序列分析與商品銷售數量預測 8
2.3.1 霍爾特-溫特方法（Holt-Winters method） 8
第三章 研究方法與步驟 9
3.1 研究對象 9
3.2 資料說明 10
3.3 研究流程與方法 11
3.4 模型評估標準 12
第四章 實證結果與分析 12
4.1 非負矩陣分解法（NMF）之應用 12
4.1.1 W矩陣（基矩陣）實證結果 13
4.1.2 H矩陣（編碼矩陣）實證結果 13
4.1.3 非負矩陣分解法（NMF）之模型評估 14
4.2 霍爾特-溫特方法（Holt-Winters method）之應用 15
4.2.1 資料前處理 16
4.2.2 訓練集、測試集及驗證集資料拆分說明 16
4.2.3 霍爾特-溫特方法實證結果 16
4.2.4 霍爾特-溫特方法之模型評估 18
4.3 實證結果 19
第五章 結論 20
第六章 參考文獻 21

疾病管制署. (2021, 五月 19). 因應本土疫情持續嚴峻，指揮中心自110年5月19日至5月28日止提升全國疫情警戒至第三級，各地同步加嚴、加大防疫限制，嚴守社區防線 [文字]. 疾病管制署; 疾病管制署. https://covid19.mohw.gov.tw/ch/cp-4822-61184-205.html
經濟部統計處. (2023). 中華民國經濟部統計處 [資料庫]. 經濟部統計處; 經濟部統計處. https://www.moea.gov.tw/Mns/dos/home/Home.aspx
林靈宏, & 魁峰. (2006). 消費者行為學. 五南圖書出版股份有限公司.
林珮君上下游記者 林怡均. (2022, 一月 26). 全台缺蛋》不明疫情侵襲蛋雞場，成壓垮產業鏈最後一根稻草，4月前難解蛋荒 | 上下游. 上下游News&Market. https://www.newsmarket.com.tw/blog/164772/
欒美雲. (2015). 超市通路轉型策略之研究－全聯福利中心為例.
蕭正和. (2020). 全國連鎖性超級市場產業之經營發展策略分析—以惠康百貨為例.
Alizadeh Zoeram, A., & Karimi Mazidi, A. (2018). A new approach for customer clustering by integrating the LRFM model and fuzzy inference system. Iranian Journal of Management Studies, Online First. https://doi.org/10.22059/ijms.2018.242528.672839
Christy, A. J., Umamakeswari, A., Priyatharsini, L., & Neyaa, A. (2021). RFM ranking – An effective approach to customer segmentation. Journal of King Saud University - Computer and Information Sciences, 33(10), 1251–1257. https://doi.org/10.1016/j.jksuci.2018.09.004
Erevelles, S., Fukawa, N., & Swayne, L. (2016). Big Data consumer analytics and the transformation of marketing. Journal of Business Research, 69(2), 897–904. https://doi.org/10.1016/j.jbusres.2015.07.001
Gupta, S., Hanssens, D., Hardie, B., Kahn, W., Kumar, V., Lin, N., Ravishanker, N., & Sriram, S. (2006). Modeling Customer Lifetime Value.
Lee, D. D., & Seung, H. S. (2000). Algorithms for Non-negative Matrix Factorization.
Marisa, F., Syed Ahmad, S. S., Mohd Yusof, Z. I., Hunaini, F., & Akhriza Aziz, T. M. (2019). Segmentation Model of Customer Lifetime Value in Small and Medium Enterprise (SMEs) using K-Means Clustering and LRFM Model. International Journal of Integrated Engineering, 11(3). https://doi.org/10.30880/ijie.2019.11.03.018
Mutinda, F., Nakashima, A., Takeuchi, K., Sasaki, Y., & Onizuka, M. (2019). Time Series Link Prediction Using NMF. Journal of Information Processing, 27(0), 752–761. https://doi.org/10.2197/ipsjjip.27.752
Nikmah, T. L., Harahap, N. H. S., Utami, G. C., & Razzaq, M. M. (2023). Customer Segmentation Based on Loyalty Level Using K-Means and LRFM Feature Selection in Retail Online Store. Jurnal ELTIKOM, 7(1), 21–28. https://doi.org/10.31961/eltikom.v7i1.648
Prasad, S., & Bruce, L. M. (2008). Limitations of Principal Components Analysis for Hyperspectral Target Recognition. IEEE Geoscience and Remote Sensing Letters, 5(4), 625–629. https://doi.org/10.1109/LGRS.2008.2001282
Shlens, J. (2014). A Tutorial on Principal Component Analysis (arXiv:1404.1100). arXiv. http://arxiv.org/abs/1404.1100
Sisodia, A., Burnap, A., & Kumar, V. (2022). Automatically Discovering Unknown Product Attributes Impacting Consumer Preferences.
Wang, X.-X., Zheng, Z.-H., Tsao, Y., & Hong, J.-W. (2016). 基於非負矩陣分解法之雜訊抑制於語音強化之研究.
Wang, Y.-X., & Zhang, Y.-J. (2013). Nonnegative Matrix Factorization: A Comprehensive Review. IEEE Transactions on Knowledge and Data Engineering, 25(6), 1336–1353. https://doi.org/10.1109/TKDE.2012.51
Wei, J.-T., Lin, S.-Y., & Wu, H.-H. (2010). A review of the application of RFM model. Afr. J. Bus. Manage.
Zalaghi, Z., & Abbasnejad Varzi, Y. (2014). Measuring customer loyalty using an extended RFM and clustering technique. Management Science Letters, 4(5), 905–912. https://doi.org/10.5267/j.msl.2014.3.026