713 Brand-Choice Analysis using Non-negative Tensor Factorization Tatsushi Matsubayashi Masahiro Kohjima Aki Hayashi Hiroshi Sawada NTT NTT Service Evolusion Laboratories matsubayashi.tatsushi@lab.ntt.co.jp, http://kecl.cslab.ntt.co.jp/ tatsushi/ kohjima.masahiro@lab.ntt.co.jp hayashi.aki@lab.ntt.co.jp hiroshi.sawada@lab.ntt.co.jp keywords: clustering, brand choice, non negative tensor factorization, NTF Summary In marketing science field, modeling of purchase behavior and analysis of brand choice are important research tasks. This paper presents a method that enables such analysis by time-series pattern extraction based on Non-negative Tensor Factorization (NTF). The development of the scanning devices and electronic payments (e.g. online shopping, mobile-phone wallet and electronic money) has led to the accumulation of more detailed POS data including the information about purchase shop, amount of payment, time, location and so on and it brings possibilities for more deep understanding of purchasing behaviors. On the other hand, due to the increase of the number of attributes, it is still difficult to effectively and efficiently handle large feature quantities. In this paper, we consider feature quantities as high-order tensor. Then, using NTF for simultaneous decomposition of multiple attributes, we show analytic effectiveness of pattern factorization for real Beer Item/Brand purchase data. By applying NTF considering three axes: USER-ID TIME-STAMP ITEM-ID, we find several temporal tendencies depending on the season. In addition, by focusing on the purchase-pattern correlations between beer items and brands, we find that the tendencies of brand choice strategies appear on the graph drawing results. 1. [Guadagni 83] [Winer 86] [hardie 93, Revelt 98, Andrews 02] [Allenby 98, Arora 98] [Erdem 96, 99, 05]. POS (Non-negative Matrix Factorization NMF) [Lee 01, 12] [Takeuchi 13a, 13]. NMF 1 K-means clustering
714 30 6 SP1-C 2015 2 1 NMF (Non-negative Tensor Factorization NTF) [Shashua 05] NTF NTF 2 3 NTF 4 NTF 5 6 2. 3 2 A F 6 2 3 A F 6 3 1 3 3 NMF SCI 2013 1 1 12 31 1 [ 06, 14] 84,587 9 77,926 10 10 1,763 ID JAN 1 4 JAN 8 JAN 13 ID ID 7 1,404 A H 8 2
715 1 2013/1 2013/12 10 1,763 10 77,926 44.2 27 1,404 55.5 5 4 2 A 250 B 200 C 180 D 30 E 200 F 200 G 200 H 100 1,404 6 4 (a) (b) 5 B GW 6 A B C E 4 (a) (b) i S i (t) μ i =1/T T t=1 S i(t) σ i = 1/T T t=1 (S i(t) μ i ) 2 L i (t) M i (t) L i (t) =(S i (t) μ i )/σ i (1) t M i (t) = L i (t ) (2) t =1 1 T =53 6(b) A B C 11 3 A 3 10 B 5 GW 11 C 3 11 E 3 5 3. NTF NTF
716 30 6 SP1-C 2015 7 NTF X A B C [Takeuchi 13b, 14] NTF [I] [J] [K] 3 X =[x ijk ] R+ I J K (J =53) I = 1763 J =53 K = 1404 NTF CANDE- COMP/PARAFAC decomposition (CP ) [Kolda 09, Cichocki 09] 3 X R 3 A =[a ir ] R I R + B =[b jr] R J R + C =[c kr ] R K R + A B C A B C ˆX =[ˆx ijk ]=A B C (3) ˆX D = I i J j K d(x ijk, ˆx ijk ) (4) k 4. 4 1 NTF 8 NTF Ā =[ā ir ]= B =[ b jr ]= C =[ c kr ]= A B C d(x ijk, ˆx ijk ) 10 KL 5 KL 3 5 1 2 4 11 c kr KL ˆX =[ˆx ijk ] R I J K + ˆx 12 c kr R ˆx ijk = a ir b jr c kr (5) E F 2 r NTF A B C D A B C 7 NTF R R =5 6 8 10 20 30 50 100 R =5 5 3 3 J j I i I i K k K k J j x ijk ˆx ijk t ijkr (6) x ijk ˆx ijk t jikr (7) x ijk ˆx ijk t ijkr (8) t ijkr = a ir b jr c kr B 9 10 (2) 9 10 1 11 2 2 8 9 10 3 5 7 8 3 9 11 4 1 5 GW
717 8 NTF Ā B C ( 0) D 3 0 5. k k Sim kk = cos( c kr, c k r)= R r c kr c k r R R r c2 kr r c2 k r (9) 9 10 B r B Sim kk 0.92 1,305 38,025 13 14 [ 07] c kr 13 c kr / R r c kr 0.3 r r r cos( c kr, c kr ) 3 2 3 2 3 1 5 13 2 3
718 30 6 SP1-C 2015 11 5 1 13 14 13 14 3 10 3 10 A B C 3 2 NTF 12 3 r=1 r=2 r=3 r=4 r=5 r=1 0.754 0.762 0.802 0.608 r=2 0.754 0.890 0.692 0.713 r=3 0.762 0.890 0.801 0.803 r=4 0.802 0.692 0.801 0.781 r=5 0.608 0.713 0.803 0.781 6. (NTF) NTF NTF NTF
非負値テンソル因子分解を用いた購買行動におけるブランド選択分析 図 13 アイテムのランク類似度を元にグラフ可視化結果を行った結果1 各ノードに対してランク r ごとに色 分けを行い 各ランクへの寄与度が高いノードのみを表示し 購買数に比例してノードサイズで描画を 行っている 図 14 アイテムのランク類似度を元にグラフ可視化結果を行った結果2 各ノードに対してブランドごとに色 分けを行い 購買数に比例したノードサイズで描画を行っている 719
720 30 6 SP1-C 2015 NTF R NTF [Guadagni 83] Guadagni, Peter M., and John DC Little. A logit model of brand choice calibrated on scanner data. Marketing Science 2.3 (1983): 203-238. [Winer 86] Winer, Russell S. A reference price model of brand choice for frequently purchased products. Journal of Consumer Research (1986): 250-256. [hardie 93] Hardie, Bruce GS, Eric J. Johnson, and Peter S. Fader. Modeling loss aversion and reference dependence effects on brand choice. Marketing Science 12.4 (1993): 378-394. [Revelt 98] Revelt, David, and Kenneth Train. Mixed logit with repeated choices: households choices of appliance efficiency level. Review of Economics and Statistics 80.4 (1998): 647-657. [Andrews 02] Andrews, Rick L., Andrew Ainslie, and Imran S. Currim. An empirical comparison of logit choice models with discrete versus continuous representations of heterogeneity. Journal of Marketing Research 39.4 (2002): 479-487. [Allenby 98] Allenby, Greg M., and Peter E. Rossi. Marketing models of consumer heterogeneity. Journal of Econometrics 89.1 (1998): 57-78. [Arora 98] Arora, Neeraj, Greg M. Allenby, and James L. Ginter. A hierarchical Bayes model of primary and secondary demand. Marketing Science 17.1 (1998): 29-44. [Erdem 96] Erdem, Tülin. A dynamic analysis of market structure based on panel data. Marketing Science 15.4 (1996): 359-378. [ 99]. POS. : 44.3 (1999): 154-163. [ 05],. POS., (2005) [Lee 01] Lee, Daniel D., and H. Sebastian Seung. Algorithms for non-negative matrix factorization. Advances in Neural Information Processing Systems (2001) [ 12]. NMF, 95(9) (2012): 829-833 [Takeuchi 13a] Takeuchi K., Ishiguro K., Kimura A., and Sawada H. Non-negative Multiple Matrix Factorization. Proceedings of 23rd International Joint Conference on Artificial Intelligence (IJ- CAI2013), pp.1713-1720, 2013. [ 13],,,., 16 (IBIS2013), 2013. [Shashua 05] Shashua A., and Tamir H. Non-negative tensor factorization with applications to statistics and computer vision. Proceedings of the 22nd International Conference on Machine Learning (ICML2005), ACM, 2005. [ 06]. POS. 34 (2006): 64-71. [ 14]. DEIM Forum 2014, F3-5 (2014) [Takeuchi 13b] Takeuchi, Koh, et al. Non-negative Multiple Tensor Factorization. Data Mining (ICDM), 2013 IEEE 13th International Conference on. IEEE (2013) [ 14],,,.. 11 (Neteco2014) (2014) [Kolda 09] Kolda, Tamara G., and Brett W. Bader. Tensor decompositions and applications. SIAM review 51.3 (2009): 455-500. [Cichocki 09] Cichocki, Andrzej, et al. Nonnegative matrix and tensor factorizations: applications to exploratory multi-way data analysis and blind source separation. John Wiley & Sons (2009) [ 07], etal... MPS, 2007.128 (2007): 65-68. 2014 10 7 2000. 2002 10 2. 2005.. NTT NTT. ( ) 2009,. 2012,.,. NTT,,. 2010 2012 NTT 1991 1993 (NTT) VLSI CAD 2000 2009 2013 2001 ( ) IEEE