topol05
TRANSCRIPT
Social Networks: Advertising, Pricing and All That
Zvi Topol & Itai Yarom
Agenda
• Introduction– Social Networks– E-Markets
• Motivation– Cellular market – Web-services
• Model
• Discussion
Social Networks
• Set of people or groups that are interconnected in some way
• Examples:– Friends – Business contacts – Co-authors of academic papers– Intermarriage connections– Protagonists in plays and comics – …
Social Networks (Continued)
Social Networks - Applications
• Information diffusion in social networks
• Epidemic spreading within different populations
• Virus spreading among infected computers
• WWW structure
• Linguistic and cultural evolution
• Dating, Jobs, Class reunions
• …
Social Network (continued)
• Popular books:
Properties of Networks
• Diameter of the network: – Average geodesic distance– Maximal geodesic distance
• Degree distributions– Regular graphs– Binomial/Poisson – Exponential
• Clustering/Transitivity/Network Density– If vertex A is connected to vertex B and vertex B is connected to
vertex C, higher prob. that vertex A is connected to vertex C– Presence of triangles in the graph– Clustering coefficient :
verticesof triplesconnected #
network in the triangles# x 3C
Properties of Networks (continued)
• Degree correlations – preferential attachment of high degree vertices/low degree vertices
• Network resilience/tolerance – effects on the network when nodes are removed in terms of– Connectivity and # of components– # of paths– Flow– …
• …
Small World Models
• Milgram conducted in the 60s a controversial experiment whose “conclusion” was 6 degrees of separation – “small world effect”
• In their study Watts and Strogatz validated the effect on datasets and showed that real world networks are a combination of random graphs and regular lattices (low dimensional lattices with some randomness)
• Barabasi et al showed that the degree distribution of many networks is exponential
E-Markets
• E-commerce opens up the opportunity to trade with information, e.g., single articles, customized news, music, video
• E-marketplaces enable users to buy/sell information commodities
• Information intermediaries can enrich the interactions and transactions implemented in such markets
E-Markets Examples
• Stock market (Continuous Double Auction)– Agents can outperform humans in unmixed markets and
have similar performance in mixed markets (of humans and agents) [1]
• Price posting markets– Cyclic price wars behavior occurs [2]
• What are the roles that agents can take in those markets?– Agent can handle large amount of information and never
get tired
[1] Agent-Human Interactions in the Continuous Double Auction, Das, Hanson, Kephart and Tesauro, IJCAI-01.
[2] The Role of Middle-Agents in Electronic Commerce, Itai Yarom, Claudia V. Goldman, and Jeffrey S. Rosenschein. IEEE Intelligent System special issue on Agents and Markets, Nov/Dec 2003, pp. 15-21.
Motivation
• Ubiquitous markets scenarios:– Cellular phones– Web services
• Applications:– Sale on demand– Advertising
Model
• Social Network where:– A is set of rational economic agents– E is set of edges connecting agents, representing
(close) social connections
• SN is weighted according to the function – Where T is a trust domain, usually T = [0, 1]– We look at trust as a partial binary relation, i.e.
– Let , then an edge e connecting both agents is in E iff
EASN ,
TEw :
),( ji aat
TAAt :Aaa ji ,
Model (continued)
• A seller s would like to use the Social Network to sell his product and bears a marginal cost function for production of
• We look at a repeated game, at the beginning of which he approaches a set of recommenders from SN and acts according to the following protocol:
C
Model(continued)
1. Seller: approaches potential recommenders
2. Recommender: sends list of recommended friends to seller
3. Seller: receives list of recommended customers (friends) and pays according to the function
4. Seller: approaches list of recommended friends
5. Customer (friend): decides whether to purchase the product
6. Recommenders: further remunerated according to
7. Seller: updates internal model of social network structure
)(1 rf
)(2 brf
Bootstrapping Details
• An initial scale-free network
• No prior knowledge of seller about the structure of the network
• Initial recommenders are picked randomly
Model (continued)
• The system updates the social network:– If a recommended agent buys the product, then the
recommender’s trustworthiness is increased by and the recommender is paid by the seller.
– If a recommended agent decides not to buy the product, then the recommender’s trustworthiness is decreased by
– Two not previously connected agents who both buy the product, have probability to be connected in the next time step.
1b
2b
Discussion
• Buyers want to identify the money maker recommenders
• Friend of a friend recommendation (different depths along the chain)
• Learning of Social Network behavior
• Relevant research