Download presentation
Presentation is loading. Please wait.
1
11/08/2009 Princeton University 1 Sharing Mart: An Experimental Platform for Socio-Technological Networks Research Dr. Hazer Inaltekin Department of Electrical Engineering Princeton University
2
11/08/2009 Princeton University 2 Agenda Today Introduction and Motivation Quick Introduction to the Sharing Mart System Multi-unit Sharing Mart Auctions Sharing Mart Experiments
3
11/08/2009 Princeton University 3 Introduction and Motivation Time Sociology Math Physics Computer Science Electrical Engineers Math Computer Science Electrical Engineers Physics Math Physics Computer Science Electrical Engineers Sociology 19501998 Milgram Granovetter 2006 Present WattsBarabasiKleinberg Lots of funding (NSF, DARPA, ARMY) on human-network interactions and joint socio-technological network characterization !!!
4
11/08/2009 Princeton University 4 Introduction and Motivation Two Modes of Connections: Physical connections via communication networks, virtual connections via social networks. Human Factor in Play: Needs of people, virtual ties among them, how they use communication nodes. P2P + Social Networks: More than 60% of total data traffic in Internet. Facebook: A Web Based Social Network. More than 150 million active users, has an average 250K registrants per day since January 2007, has an average weekly growth rate around 3%. Social Overlay - Communication Underlay Networks
5
11/08/2009 Princeton University 5 Introduction and Motivation An exponential increase in the number of social networking sites. An exponential increase in the number of active members. Bottom line: Understand human behavior and social ties among people better design next generation communication networks. Years Millions Social Overlay - Communication Underlay Networks
6
11/08/2009 Princeton University 6 Introduction and Motivation Social Overlay - Communication Underlay Networks
7
11/08/2009 Princeton University 7 Vision and Motivation Examples: Distributed decision making, trust formation, traffic engineering, network capacity and network connectivity calculations, network resource allocation, wireless social networking. Collaborators: Prof. Matthew Salganik from Sociology Department at Princeton University. Prof. Jacob Shapiro from Political Science Department at Princeton University/ Prof. Junshan Zhang from Electrical Engineering Department at Arizona State University. Social Overlay - Communication Underlay Networks
8
11/08/2009 Princeton University 8 Time/Resources v.s. Research Task Trade-off Time and Resources Required Mathematical Modeling / Analysis Agent Based Simulation and Empirical Data Analysis Testbed Design and Empirical Data Collection / Analysis 1-) Topology of Socio- technological Networks and Delay Characterization 1-) Delay Simulations 2-) Yahoo Communication Network Data. (500 Gbyte) 1-) Sharing Mart Project Introduction and Motivation
9
11/08/2009 Princeton University 9 Agenda Today Introduction and Motivation Quick Introduction to the Sharing Mart System Multi-unit Sharing Mart Auctions Sharing Mart Experiments
10
11/08/2009 Princeton University 10 Quick Introduction to Sharing Mart Sharing Mart, S-Mart, is a virtual money (token) based social file sharing platform for Web users to come together and exchange their files. Innovative research agenda with fun functional modules. Current Status: Active with 250 students in SEAS. User Behavior Characterization File Transactions Fixed Price or S-mart Auction Group Formation Content Request Content and User Rating User Generated Advertisements Digital Rights Management
11
11/08/2009 Princeton University 11 Quick Introduction to Sharing Mart Sharing Mart System http://sharingmart.princeton.edu S-Mart Main Web PageStudent Interests at SEAS
12
11/08/2009 Princeton University 12 Quick Introduction to Sharing Mart Personalized Home Pages
13
11/08/2009 Princeton University 13 Quick Introduction to Sharing Mart Personalized Web Stores
14
11/08/2009 Princeton University 14 Quick Introduction to Sharing Mart Content Upload
15
11/08/2009 Princeton University 15 Quick Introduction to Sharing Mart Ad Upload
16
11/08/2009 Princeton University 16 Quick Introduction to Sharing Mart Content Request
17
11/08/2009 Princeton University 17 Quick Introduction to Sharing Mart Group Formation
18
11/08/2009 Princeton University 18 Agenda Today Introduction and Motivation Quick Overview of the Sharing Mart System Multi-unit Sharing Mart Auctions Sharing Mart Experiments
19
11/08/2009 Princeton University 19 Multi-unit Sharing Mart Auctions Auctions: A simple solution used since 500 B.C. to discover market value of goods. Auctions Single UnitMulti-unit First PriceSecond Price Ascending Price English Auctions Descending Price Dutch Auctions Vickery Auctions Ebay Auctions DiscriminatoryUniformVickery Sharing Mart Auctions Multi-unit Descending Price Dutch Auction Ausubel Auctions
20
11/08/2009 Princeton University 20 Multi-unit Sharing Mart Auctions Sharing Mart Auction: is a uniform price, unit demand and multiple winner file auction. Parameters: Auction duration, minimum price, number of copies to be sold K. Can be specific to groups. Market Clearing Price in Sharing Mart Auctions: (K+1)st highest bid. Definition: An auction is efficient if it allocates the goods to highest bidders. Definition: An auction is incentive compatible if it induces a bidder to submit a bid that sincerely reflects her value for the item. Revenue Equivalence: A Sharing Mart auction is revenue equivalent to other two auction types (discriminatory and Vickery multi-unit auctions). Some Properties
21
11/08/2009 Princeton University 21 Multi-unit Sharing Mart Auctions Auction Interface
22
11/08/2009 Princeton University 22 Agenda Today Introduction and Motivation Quick Overview of the Sharing Mart System Multi-unit Sharing Mart Auctions Sharing Mart Experiments
23
11/08/2009 Princeton University 23 Sharing Mart Experiments Number of Participants: 19 undergraduate students. Number of Auctions: 4 Bidding Behavior: Manual bidding Bidding strategy can change over time No restriction Open Economy: Initial budget: 500 [Tokens] Can change over time by selling files and inviting friends in SEAS to buy their files Close to collecting plain data in eBay Experiment 1 - General Set-up
24
11/08/2009 Princeton University 24 Sharing Mart Experiments Auction Parameters: Number of students = 19 Minimum price = 20 [Tokens] Number of copies = 2 Start Date: 11/1/2008 11:30:00 AM End Date: 11/4/2008 11:59:00 PM Duration: 304,140 [Seconds] Auction Results: Final Price: 612 [Tokens] Number of unique bidders: 10 Total Income: 1224 [Tokens] Experiment 1 - Auction 1 Set-up
25
11/08/2009 Princeton University 25 Sharing Mart Experiments 65 69 62, bids 612. Token balance 612 626959 Experiment 1 - Auction 1 Results
26
11/08/2009 Princeton University 26 Sharing Mart Experiments Experiment 1 - Auction 1 Results
27
11/08/2009 Princeton University 27 Sharing Mart Experiments 6965 62 - Token balance 612 55 Winners Experiment 1 - Auction 1 Budget Distribution
28
11/08/2009 Princeton University 28 Sharing Mart Experiments Auction Parameters: Number of students = 19 Minimum price = 20 [Tokens] Number of copies = 5 Start Date: 11/5/2008 2:00:00 AM End Date: 11/8/2008 11:59:00 PM Duration: 338,340 [Seconds] Auction Results: Final Price: 551 [Tokens] Number of unique bidders: 14 Total Income: 2755 [Tokens] Experiment 1 - Auction 2 Set-up
29
11/08/2009 Princeton University 29 Sharing Mart Experiments 54 55 60 bid amount = 707 60 bid amount = 710 59 57 37 62, bids 616. Token balance 616 Experiment 1 - Auction 2 Results
30
11/08/2009 Princeton University 30 Sharing Mart Experiments Experiment 1 - Auction 2 Results
31
11/08/2009 Princeton University 31 Sharing Mart Experiments 5565 69 They do not bid 60, token balance = 710 57, 59 But they are not winners They are among winners 37, 62, 54 They are among winners Experiment 1 - Auction 2 Budget Distribution
32
11/08/2009 Princeton University 32 Sharing Mart Experiments Auction 3: Number of copies = 7 Total Income: 3500 [Tokens] Auction 4: Number of copies = 9 Total Income: 180 [Tokens] Experiment 1 - Auction 3 and 4
33
11/08/2009 Princeton University 33 Sharing Mart Experiments Seller Revenue Optimization
34
11/08/2009 Princeton University 34 Sharing Mart Experiments ELE 382 SEAS Intra-community Links Inter-community Links Community Structure
35
11/08/2009 Princeton University 35 Sharing Mart Experiments Number of Participants: 7 graduate students. Bidding Behavior: Automated bidding. Bidding strategy does not change over time. Closed Economy: Initial budget: 1800 [Tokens]. Can change over time by selling/buying files from other 6 graduate students. 96-hours Auction Competition: Each student sets 3 auctions to sell - Seller Strategy: Free Parameters: Number of items, duration and minimum price. Each student submits an automated bidding agent to bid 18 auctions - Buyer Strategy. Total Number of Points Collected: Number of Auctions Won * 100 + Total Revenue in 3 Auctions. Experiment 2 collects auction data in a controlled manner. Experiment 2 - General Set-up
36
11/08/2009 Princeton University 36 Sharing Mart Experiments Experiment 2 - General Set-up
37
11/08/2009 Princeton University 37 Sharing Mart Experiments Points of Student i: is the number of items sold in auction k from student i. is the final sale price in auction k from student i. is equal to 1 if student i wins auction k from student j. Score Function to Optimize: Experiment 2 - Nash Equilibrium : Minimum sale price in auction k from student i : Duration of auction k from student i : Bidding strategy of student j
38
11/08/2009 Princeton University 38 Sharing Mart Experiments Theorem: The above experiment has a symmetric and socially optimal Nash equilibrium at which: Note: At this equilibrium, all students earn 3600 points, therefore a score of 100. Experiment 2 - Nash Equilibrium
39
11/08/2009 Princeton University 39 Sharing Mart Experiments Average Number of Copies (Averaged over 3 Best Seller Strategies) is 5.33. Nash equilibrium predicts this number to be 6. Experiment 2 Results - Revenue v.s. Number of Copies All students benefit from setting number of copies to 6. Conclusion 1 Average revenue increase is 20%. Conclusion 2
40
11/08/2009 Princeton University 40 Sharing Mart Experiments Experiment 2 Results - Revenue v.s. Number of Copies Minimum Price = 97 Empirical revenue versus number of copies curve peaks around 5-6 copies. Conclusion 3
41
11/08/2009 Princeton University 41 Sharing Mart Experiments Minimum Price: Average minimum price over all auctions is 97 [Tokens]. Average minimum price over three best sellers is 95 [Tokens]. High minimum price decreases the percentage of copies sold. All copies are sold if the minimum price is below 80 but none of the all copies sold if minimum price is above 100. Conclusion 4: Nash equilibrium strategy minimum price matches with the empirically observed minimum price. Bidding Strategy: Three most successful bidders with success rate 97% snipe within the last 60 seconds. Average number of bids per auction from the three most successful bidders is 1.3. Continuous bidding with 8.6 bids per auction reduces to the success rate to 62%. Conclusion 5: Small average number of bids per auction is in accordance with the efficiency and incentive compatibility property of Sharing Mart auctions. Experiment 2 Results - Effects of Other Parameters
42
11/08/2009 Princeton University 42
43
11/08/2009 Princeton University 43 Summary and Conclusions Boundaries between sociology, physics, computer science and electrical engineering are disappearing. New Networking Paradigm: Social Overlay - Communication Underlay Sharing Mart: Collect data in
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.