1
2

if we are going to use the relationship between people in social network, we need to know how well person A trust person B concerning some fields. what's the state of the art to get the trust metric between people in social network.

asked Jul 20 '10 at 02:38

charlie's gravatar image

charlie
140121417

Could you be more specific? What kinds of information do the nodes and links between people in this social network have?

(Jul 20 '10 at 14:54) aditi

Also, how would you even define trust? I've seen models of influence, but not trust.

(Jul 20 '10 at 15:14) Alexandre Passos ♦

models of influence is certainly helpful.

(Jul 22 '10 at 22:11) charlie

4 Answers:

Influence and trust should be regarded as closely related. If your network is, say, the blogosphere, then you can model trust by blogroll links and in-story hyperlinks (weighted by recency, and perhaps some sentiment analysis of the link's context, though that's not usually done). These are only explicit links, however. Deducing implicit links by expectation maximization on paths that form a given epidemic, a la this awesome paper Information Diffusion in Blogspace is hard, but perhaps rewarding. If your dataset is a cell phone contact network or a website, frequency of contact is probably your best bet, and perhaps some keyword mining (in wall post data, for instance). Unfortunately, it's likely that you won't have access to the most needed data due to the nature of these datasets and privacy concerns.

If you know what to measure, you can try to infer "trust links" as a sort of strong tie that tends to be associated with some characteristic - for instance, accepting a recommendation in a recommendation network. Network topology itself gives some clues and is the easiest data to obtain by far, e.g. Link Prediction Problem in Networks

More info on your problem and its context would be nice though.

answered Jul 22 '10 at 20:49

Jacob%20Jensen's gravatar image

Jacob Jensen
1644285360

1

influence and trust in social network may be some different. I think, we should take account of this problem in our work.

(Jul 22 '10 at 22:16) charlie
1

Unfortunately, however, you can't say that someone is well trusted unless there's evidence of influence. If someone is trusted but not influential, then there will tend to be very few direct references (hyperlinks in the context of blogs) to their opinions. If there are lots of references, you could measure similarity. If there are references that disagree with the source, you know that the poster doesn't trust the source. If the references agree with him, and particularly if they imitate him, you know the poster trusts the source. Consider an information epidemic - your best indication that a poster trusts a source is if he adopts that source's stance on a given issue, particularly if this happens very early in the outbreak of the information epidemic, when there's little precedent for the source's opinion. Unfortunately, data for this is extremely scarce. PageRank and authority measures will do a better job, probably, than crude measurements.

(Jul 22 '10 at 22:23) Jacob Jensen

This is relevant from WWW 2010

http://www2010.org/www/program/papers/social-networks-1/

Modeling Relationship Strength in Online Social Network

Rongjing Xiang, Jennifer Neville, Monica Rogati

Previous work analyzing social networks has mainly focused on binary friendship relations. However, in online social networks the low cost of link formation can lead to networks with heterogeneous relationship strengths (e.g., acquaintances and best friends mixed together). In this case, the binary friendship indicator provides only a coarse representation of relationship information. In this work, we develop an unsupervised model to estimate relationship strength from interaction activity (e.g., communication, tagging) and user similarity. More specifically, we formulate a link-based latent variable model, along with a coordinate ascent optimization procedure for the inference. We evaluate our approach on real-world data from Facebook and LinkedIn, showing that the estimated link weights result in higher autocorrelation and lead to improved classification accuracy.

This answer is marked "community wiki".

answered Jul 24 '10 at 15:51

DirectedGraph's gravatar image

DirectedGraph
54531422

From CHI 2009: Predicting tie strength with social media

Uses Facebook interaction characteristics to predict the level of friendship between users within 1-point accuracy on a 10-point scale.

answered Sep 22 '10 at 12:32

aditi's gravatar image

aditi
85072033

Your answer
toggle preview

Subscription:

Once you sign in you will be able to subscribe for any updates here

Tags:

×6

Asked: Jul 20 '10 at 02:38

Seen: 937 times

Last updated: Sep 22 '10 at 12:32

powered by OSQA

User submitted content is under Creative Commons: Attribution - Share Alike; Other things copyright (C) 2010, MetaOptimize LLC.