General Motors Company · NYSE:GM
General Motors — Technical Presentation (December 2005)
25 slides from General Motors's technical presentation, Dec 9, 2005
Slide 1: Attaining Fast and Successful Searches in E-Commerce Environments
Attaining Fast and Successful Searches in E-Commerce Environments Jeffrey Tew (GM) Raz Lin and Sarit Kraus (Bar-Ilan Univ.) GM ISL WORKSHOP ON FRONTIERS OF E-BUSINESS Bangalore 8-9 December 2005 OSGS: guided search in an online store is an online store for GM’s Auto Spare Parts. History Similar Info
Slide 2: Problems of current online stores
http://132.70.1.8:8080/gm India Science Lab GM ISL Workshop on Frontiers of E-Business Problems of current online stores: Information Overload • No Guidance • No Personalization (Homogeneous Search) • Bugs GM ISL Workshop on Frontiers of E-Business India Science Lab 2
Slide 3: System Overview OSGS
Long Term Goals Increasing enterprise’ enterprise’s long- long-term benefits Long-term relationships with customers Assist customers in rapidly Short Term attaining successful searches. Goals Make search user- user-friendly and flexible GM ISL Workshop on Frontiers of E-Business India Science Lab Sy
Slide 4: Core Modules' Description
Core Modules’ Description Error Module New Customer Module Keyword Searching Module Login Main Module Display Results Module Module Browse Searching Module Display Item Update Profile Module Module Calculate History Module * All modules have a flow of data to/from the Error Module. Purchase Item Mod
Slide 5: Bottom Up
Regular keyword Search: same keywords, different customers same items Example: If a Buick owner and a Chevrolet owner, both search for “motor”, they will both get the same results: 231 items. Buick Bottom Up Chevrolet Cooling System HVAC Wiper wiper motor Advanced motor motor motor blower GM ISL Wor
Slide 6: OSGS as a Plain Search Engine
OSGS as a Plain Search Engine Providing full search coverage - Keyword Browse Search Search Maximizing customer options of finding results. Need/Want to give More Help/Assistance GM ISL Workshop on Frontiers of E-Business India Science Lab OSGS as a Personalized Search Engine 1. Customer’s Preferenc
Slide 7: 1. Customer's Preferences
1. Customer’s Preferences Higher weight is given to items that pertain to the customer’s car: • Make •Model •Year Also applying preferences concerning the importance of: Quality, Price, Warranty GM ISL Workshop on Frontiers of E-Business India Science Lab GM ISL Workshop on Frontiers of E-Business I
Slide 8: 2. Customer's History
GM ISL Workshop on Frontiers of E-Business India Science Lab 2. Customer’s History Motivation: Assumption - Customers tend to buy items similar to items they have already purchased. History: a set of past events. An event is a search represented by the path to the item, time and whether the item was
Slide 9: Example
Formally, for each customer we maintain history H: Set of her most recent historical events. Historical event h is a triplet (<v1, v2, …, vk>, t, l): At time t the customer searched for item, whose path on the search tree consists of the vertices <v1, v2, …, vk> l is the likeability parameter: θ if
Slide 10: History Weight and Neighborhood Formation
History Weight: The weight of each historical event equals to the likeability value of the historical event, multiplied by exp( t/current_time ). The weight of an item or category equals to the sum of total weight of the historical events it appears in, divided by the number of the historical events
Slide 11: Proximity Measure
Proximity Measure: Using mean squared differences. Hx D xy = ( H x , H y ) 2 = ∑ 1 ( H xi − H yi ) 2 Hx The lower the mean squared difference, the greater the similarity GM ISL Workshop on Frontiers of E-Business India Science Lab 3. Neighborhood by Demographics- and Preference-Profile Motivation: A
Slide 12: 4. Neighborhood by History Profile
4. Neighborhood by History Profile Motivation: Assumption - Customers with similar history of purchases tend to buy items similar to items their “soul-mates” have already purchased. GM ISL Workshop on Frontiers of E-Business India Science Lab Want to consider also customers who purchased similar ite
Slide 13: Solution - Virtual Paths
Solution - Virtual Paths: When purchasing items,OSGS also considers other paths of the items. Giving (lower) weights to virtual paths: Not losing customer’s intentions. Optimizing neighborhood search. GM ISL Workshop on Frontiers of E-Business India Science Lab GM ISL Workshop on Frontiers of E-Busi
Slide 14: GM ISL Workshop on Frontiers of E-Business India Science Lab GM ISL Workshop on
GM ISL Workshop on Frontiers of E-Business India Science Lab GM ISL Workshop on Frontiers of E-Business India Science Lab 14
Slide 15: GM ISL Workshop on Frontiers of E-Business India Science Lab GM ISL Workshop on
GM ISL Workshop on Frontiers of E-Business India Science Lab GM ISL Workshop on Frontiers of E-Business India Science Lab 15
Slide 16: Experiments Scenario example
Experiments 3 subjects: • Customer’s History • Neighborhood by History Profile • Neighborhood by Demographics- and Preference-Profile. • The subjects had to look for items without the help of our algorithms and with the help without knowing the exact settings. • 60 subjects GM ISL Workshop on Fronti
Slide 17: Measured
We compared with ACDelco’s online store GM ISL Workshop on Frontiers of E-Business India Science Lab Measured: 1. Satisfaction; 2. Time; 3. Number of Clicks. Main Results: the subjects were significantly more satisfied when getting our algorithms’ help than without it or when using ACDelco. GM ISL W
Slide 18: Significantly higher satisfaction from keyword search with our algorithms than ACDelco
Significantly higher satisfaction from keyword search with our algorithms than ACDelco. Keyword Search with Algorithm vs. ACDelco (All Experiments) Keyword Search ACDelco 90% 80% Percentage of # 70% of 60% Experimenters 50% 40% 30% 20% 10% 0% Preference of all subjects concerning keyword search with
Slide 19: Significantly higher satisfaction from using our system than ACDelco
Significantly higher satisfaction from using our system than ACDelco. GM S IM vs. ACDe lco (All Ex pe rim e nts) GM SIM A CDelc o 80% 70% P e rce nta ge of # of 60% Ex pe rim e nte rs 50% 40% 30% 20% 10% 0% Preference of all subjects concerning GMSIM vs. ACDelco. GM ISL Workshop on Frontiers of E-Bu
Slide 20: Searches with History Algorithm vs. Searches without Algorithm
Searches with History Algorithm vs. Searches without Algorithm With Algorithm Without Algorithm 80% 70% 60% 50% Percentage of 40% # of 30% Experimenters 20% 10% 0% Browse Search Keyword Search Preference of subjects concerning keyword search and browse search with history algorithm vs. keyword searc
Slide 21: Subjects rate significantly higher than ACDelco
Subjects rate significantly higher than ACDelco Keyword Search with Algorithm vs. ACDelco Keyword Search with Algorithm ACDelco 7 Neighborhood by Demographics- 6 and Preference-Profile Neighborhood by History 5 Customer’s History General 4 Satisfaction 3 2 1 0 General satisfaction of subjects using
Slide 22: Adapting GMSIM to any Domain
Subjects were significantly more satisfied with respect to ease and flexibility from using Keyword Search with Algorithm vs. ACDelco Keyword Search with Algorithm ACDelco 7 Neighborhood by Demographics- 6 and Preference-Profile Neighborhood by History 5 Customer’s History Ease and 4 Flexibility 3 2
Slide 23: Conclusions
Conclusions • OSGS : personalized, adaptive on-line store •Algorithms: based on collaborative filtering and information retrieval; help customers in search for products. •Experiments:OSGS outperforms other non-personalized search systems. GM ISL Workshop on Frontiers of E-Business India Science Lab
Slide 24: Benefits of Personalized Search
Benefits of Personalized Search 90% of online consumers use the Web to research products before buying: • 37% use a manufacturer’s site before buying at a physical store; • 61% said it strongly influenced which product to buy; and, •65% said it influenced their decision to visit the company’s physic
Slide 25: Benefits: Customer Interactions
Benefits: Customer Interactions Direct Sales: No middleman needed. Availability: 24 hours a day, 7 days a week, 365 days per year from anywhere. Targeted Customer Base: Access to established community of online shoppers for special deals, product announcements, other advertising as well as sales. GM
General Motors Company — Technical Presentation (December 2005) on 6ix