Inferring Trust Relationships in Web-based Social Networks Export

@article{citeulike:580963, author = {Golbeck, Jennifer and Hendler, James}, citeulike-article-id = {580963}, comment = { This paper describes trust and methods for inferring unknown trust relationships by propagating known trust values through a social network. The authors define trust in a person as ``a committment to an action based on a belief that the future actions of that person will lead to a good outcome''. The problem of the duality of trusting a person and trusting a person's judgement of other people is mentioned but handwaved away. Social networks are modeled as directed graphs where edges have an associated binary trust value with 1 representing trust and 0 representing a lack of trust (but not necessarily the presence of distrust). It seems a trinary scale would be better able to capture this difference, but this is not addressed. When a node in the network (the source) wants to know whether or not it should trust another node (the sink), it first checks to see whether or not it has explicitly stated a trust value for that node. If so, it is used. Otherwise, the trust values are averaged from each trusted node. Child nodes perform this exact same process recursively. Two algorithms are presented: one in which all intermediate values are rounded to the binary scale, and one in which intermediate values are continuous and only the source performs rounding. The version that performs rounding at each step is shown to be theoretically more accurate, although I could not follow this section and am not convinced. Experiments on synthetic networks were performed that confirmed this and showed that personalized accuracy was better than general accuracy if and only if general accuracy was at least 50\%. The simulations are interesting, but I am not sure they accurately reflect real social networks. Given the wealth of such networks that the authors claim exist, it would be much better if experiments were performed on real social data. A sample application of trust scores used for email filtering was provided but there is no evaluation of how effective this method is in practice. The paper provides a good background on trust in social networks, but does not clearly explain its theoretical contributions and does not contain any convincing experimental results.}, journal = {to appear in ACM Transactions on Internet Technology}, keywords = {communities, trust, web}, posted-at = {2006-04-10 04:10:05}, priority = {0}, title = {Inferring Trust Relationships in Web-based Social Networks} }

TY - JOUR ID - citeulike:580963 L3 - citeulike-article-id:580963 JF - to appear in ACM Transactions on Internet Technology TI - Inferring Trust Relationships in Web-based Social Networks KW - communities KW - trust KW - web N1 - This paper describes trust and methods for inferring unknown trust relationships by propagating known trust values through a social network. The authors define trust in a person as ``a committment to an action based on a belief that the future actions of that person will lead to a good outcome''. The problem of the duality of trusting a person and trusting a person's judgement of other people is mentioned but handwaved away. Social networks are modeled as directed graphs where edges have an associated binary trust value with 1 representing trust and 0 representing a lack of trust (but not necessarily the presence of distrust). It seems a trinary scale would be better able to capture this difference, but this is not addressed. When a node in the network (the source) wants to know whether or not it should trust another node (the sink), it first checks to see whether or not it has explicitly stated a trust value for that node. If so, it is used. Otherwise, the trust values are averaged from each trusted node. Child nodes perform this exact same process recursively. Two algorithms are presented: one in which all intermediate values are rounded to the binary scale, and one in which intermediate values are continuous and only the source performs rounding. The version that performs rounding at each step is shown to be theoretically more accurate, although I could not follow this section and am not convinced. Experiments on synthetic networks were performed that confirmed this and showed that personalized accuracy was better than general accuracy if and only if general accuracy was at least 50\%. The simulations are interesting, but I am not sure they accurately reflect real social networks. Given the wealth of such networks that the authors claim exist, it would be much better if experiments were performed on real social data. A sample application of trust scores used for email filtering was provided but there is no evaluation of how effective this method is in practice. The paper provides a good background on trust in social networks, but does not clearly explain its theoretical contributions and does not contain any convincing experimental results. AU - Golbeck, Jennifer AU - Hendler, James ER -