Skip to main content



Epinions.com and the “Web of Trust”

     Epinions.com was created in 1999 as an online review site aimed towards providing its users with useful information about a wide variety of products. Like most websites that depend on algorithms to stay competitive, Epinions.com does not disclose its formula used to rank and order product suggestions to the public. Based on our knowledge of its operations, however, we are able to make assumptions as to what information is used in its algorithm.
     On Epinions.com, members can categorize others they encounter on the site as “trusted” or “blocked.” Of course, if a member is indifferent about a review he reads, he can choose to disregard that author and not place him in either of these two categories. Once a user decides to “trust” a review from another user, that user is added as a node to the “Web of Trust,” a directed graph that represents trust among users. This Web of Trust provides Epinions.com with very useful information about the relationships among its users.
     In order to rank reviews, Epinions.com can analyze the Web of Trust to find a solution. For example, say user A is searching for a review about a product. If user B is in his immediate Web of Trust, meaning user A has chosen to trust user B, then we can make further predictions based on user B’s immediate Web of Trust. If user B trusts user C, there is a strong possibility user A will also trust user C. Here is the Web or Trust that represents these relationships:
Web of Trust
     Similarly, if we know user A trusts user B, and user B does not trust (blocks) user C, we can predict that user A will also not trust user C. This creates another balanced graph:
Web of Trust
     From these predictions, Epinions.com will place higher value for reviews written by any user in the immediate Web of Trust of a trusted member. In accordance with this theory, the algorithm would not give precedence to a review if the author is blocked by a node included in the immediate Web of Trust of a member. Thus, the algorithm used by Epinions.com would display reviews for one member based on the trusted nodes of that member’s trusted nodes.
     While the Web of Trust provides very useful information, this algorithm stretches far beyond the limited information the Web of Trust can provide. In order to suggest a review that a certain user will find useful, Epinions.com and similar websites use recommender systems. This system takes into account the data provided by the information listed on the user profiles of the members. Recommender systems are made even more efficient when collaborative filtering is used in the algorithm. Collaborative filtering suggests reviews that are highly rated by users who tend to have similar opinions, even if they are disconnected in the Web of Trust.
We do know, however, that the Web of Trust plays a prominent role in the algorithm. Research finds that people are more likely to trust familiar users they have had positive experience with than foreign users who appear as a product of online recommender systems.

http://www.cwi.ugent.be/martine/papers/pvictor2011a.pdf

Comments

Leave a Reply

Blogging Calendar

September 2011
M T W T F S S
 1234
567891011
12131415161718
19202122232425
2627282930  

Archives