Inferring win–lose product network from user behavior
0 likes559 views
1) The document proposes a new method to analyze relationships between substitute products using user browsing and purchase behavior data from e-commerce sites. It examines which products are superior to others in attractiveness. 2) The method was tested on wedding venue data from a Japanese wedding planning site. It accurately identified competitive and win-lose relationships between venues based on correlations with user survey data. 3) The method also extracts keywords explaining why one product is superior by analyzing reviews from users who chose that product over others. This provided more accurate superiority factors than a simple baseline method.
More Related Content
Similar to Inferring win–lose product network from user behavior (20)
More from Shuhei Iitsuka (20)
Recently uploaded (20)
Inferring win–lose product network from user behavior
- 1. Inferring Win–Lose Product Network from User Behavior Shuhei Iitsuka? , Kazuya Kawakami? , Seigen Hagiwara* , Takayoshi Kawakami** , Takayuki Hamada*** , Yutaka Matsuo? 1 ? The University of Tokyo, Japan ? University of Oxford, UK * Recruit Marketing Partners Co., Ltd., Japan ** Industrial Growth Platform, Inc., Japan *** IGPI Business Analytics & Intelligence, Inc., Japan
- 2. Background ● E-commerce is expanding, and various data mining methods have been proposed. ● However, few data mining techniques have been proposed to provide: ○ superiority relations of product attractiveness. ○ why that superiority is formed. → Understanding competitive advantages is important for product marketers. INTRODUCTION 2 Data mining is playing an important role in e-commerce marketing. https://www.amazon.com/dp/B005EOWBHC/
- 3. Objective ● We propose a new method to examine win–lose relation. ○ Superiority relation among substitute products in terms of attractiveness. ● We also propose a review mining method to extract why that superiority is formed. ● Our method uses the difference between users’ browsing and purchasing behavior. 3 INTRODUCTION browse browse purchasepurchase SUPERIORITY SUPERIORITY ??? USER #1 USER #N stylish, modern compact, light win–lose network
- 4. Product Relation Substitute or Complementary ● Common means of perceiving product relations in consumer theory. ● Have been used in e-commerce mining to understand product relations. Network Analysis ● Network analysis methods have been imported to e-commerce marketing. → Few studies have examined directed relation of products. 4 RELATED WORKS SUBSTITUTE COMPLEMENTARY
- 5. Win–Lose Product Relation ● Competitive relation Item A and B are browsed by the same user. = Item A and B are in competition. (A ? B) ● Win–lose relation Item A is purchased after item A and B are browsed. = Item A is superior to item B. (A ← B) 5 PROPOSED METHOD User Browsed (Purchased) 1 B, C 2 A, B, C 3 B, C Competitive network Win–lose network Access Log
- 6. Superiority Factor Analysis ● Examines why the superiority is formed in a form of keywords (superiority factors). ● Item A’s superiority factors to item B comes from the reviews of products purchased by patrons who prefer item A to B. 6 PROPOSED METHOD Users who supports A > B superiority
- 7. Zexy http://zexy.net/ ● Japanese largest wedding portal website. ● Browse = browse a venue page Purchase = reserve a venue tour ● Used log data ○ Jan 1, 2012 — Oct 31, 2012 ○ User ID, URL, Flag for tour reservation 7 ANALYSIS RESULTS Tour reservation made! BROWSE PURCHASE VENUE PAGE LIST PAGE
- 8. Product Network Competitive network of wedding venues in Japan The color segments match well with the competition cluster. → Competition takes place per region. 8 ANALYSIS RESULTS Win–lose network of selected venues in Tokyo Directed relation of attractiveness is shown. → E-commerce owners can expect users’ tendency to make conversion actions.
- 9. Superiority Factor Analysis 9 ANALYSIS RESULTS ● ceremony ● garden ● banquet ● solemnity ● photograph ● Japanese dish ● Japanese-style room ● ceremony ● garden ● bus Venue H Venue J Venue A
- 10. Experimental Setup ● Evaluate how much our method can estimate the actual product relations. ● Conducted a user survey of couples to observe actual user perceptions. ○ Jan 23, 2012 — Dec 14, 2013 ○ Couples who used Zexy for tour reservation and held a ceremony ○ Selected 10 venues in Tokyo 10 EVALUATION EXPERIMENT Log data User survey USER (N=202) PRODUCT USER (N=173) PRODUCT BROWSE A VENUE RESERVE A TOUR ATTEND A TOUR HOLD A CEREMONY
- 11. Results Experiment #1: Correlation of network weights ● Correlation between the weights of the product network: user survey VS log data. ● Significant correlation was found between them for both of competition and win–lose network. → Log data can be a good alternative of the user survey. 11 EVALUATION EXPERIMENT Correlation of competitive relation 0.685 (p < 0.01) Correlation of win–lose relation 0.648 (p < 0.01)
- 12. Results Experiment #2: Superiority factor analysis ● Actual factor: Responses to the question “Reason for selection”. ● Baseline method: Regards the winner product’s review as the superiority factors. ● Across all venues, proposed method estimated more actual factor words significantly (p < 0.05). ● Proposed method shows interesting findings while Baseline method only shows general and well-known property of the product. EVALUATION EXPERIMENT ATTENDED A TOUR HELD A CEREMONY Reason for selection Actual factor words Method Factor Words of product D against G Proposed (13/20) chapel, hospitality, ceremony, guest, feeling, day, impression, staff, stained glass, banquet, dish, atmosphere, church, location, San, photograph, lovely, venue, Omotesando, weddings Baseline (3/20) map, cathedral, forbidden, she, Akka, order, impression, problem, standard, cloud, stained glass, church, European, minute, overall, exchange, movie, ring, Omotesando, bringing 12
- 13. Discussion & Conclusion ● Our proposed method is useful to estimate the superiority relation of products and why that superiority is formed. ● Our text mining method does not consider polarity of the sentence. → Our method captures aspects which users care. ● Analysis needs to be done in the same category of products and only between substitutes. Contribution ● Proposed a new data mining method to analyze superiority product relation. ● Proposed a text mining method to analyze superiority factors. → E-commerce owners can plan effective marketing or promotion strategies. ● Evaluated if log data can be a good alternative of user survey. → Huge costs (distribution, data input etc.) can be saved. 13
- 14. Thank you for listening. 14 https://tushuhei.com iitsuka@weblab.t.u-tokyo.ac.jp