A survey of trust use and modeling in current real systems
From TrustLet, a free, collaborative project for collecting and analyzing information about trust metrics.
A survey of trust use and modeling in current real systems [1] is a paper by Paolo Massa.
[edit] Abstract
This chapter discusses the concept of trust and how trust is used and modeled in online systems currently available on the Web or on the Internet. It starts by describing the concept of information overload and introducing trust as a possible and powerful way to deal with it. It then provides a classification of the systems that currently use trust and, for each category, presents the most representative examples. In these systems, trust is considered as the judgment expressed by one user about another user, often directly and explicitly, sometime indirectly through an evaluation of the artifacts produced by that user or her activity on the system. We hence use the term “trust” to indicate different types of social relationships between two users, such as friendship, appreciation and interest. These trust relationships are used by the systems in order to infer some measure of importance about the different users and influence their visibility on the system. We conclude with an overview of the open and interesting challenges for online systems that use and model trust information.
[edit] Excerpts
We have identified few different categories in which the systems can be grouped based on the common features and properties they share. The categories we define are:
- E-marketplaces
- Opinions and activity sharing sites
- Business/job networking sites
- Social/entertainment sites
- News sites
- The Web, the Blogosphere, and the Semantic Web
- Peer-to-Peer (P2P) networks
E-marketplaces are online systems in which a user can sell items owned and can buy offered items. In such a context, typically the buyer does not know the seller, and vice versa. So, in order to decide about a possible commercial exchange that involves the risk of not being paid or of not receiving the products already paid for, it is very important to be able to decide in a quick, reliable, and easy to understand way about the trustworthiness of the possible commercial partner. The success of eBay (http://ebay.com) is largely due to the fact it provides an easy way to do this.
Opinions and activity sharing sites on the other hand are Web sites where users can share with the other users their opinions on items and, in general, make their activities and preferences visible to and usable by other users. The best example of an opinions site is Epinions, in which users can write reviews about products. In activity sharing sites, the user activity is made visible to the other users who can in some way take advantage of it. Two examples of activity sharing sites are Del.icio.us, in which users can bookmark URLs they consider interesting, and Last.fm, in which users make visible which songs they listen to. Bookmarking a URL and listening to a song can be considered as the elicitation of a positive opinion about the considered item. However, a user might be more interested in following the reviews and activity of a certain other user, and trust statements can be used exactly for this purpose.
Business/job networking sites are Web sites where users post information about their job skills and ambitions so that other people can find them when they are looking for someone to hire for a specific job. Lately, many systems started to exploit the social (trust) network between users: users can explicitly state their connections, that is, professionals they have already worked with and found reliable and trustworthy. In this way, using the system, a user can enter in contact with the connections of his/her connections and discover potentially interesting new business partners. Linkedin.com and Ryze.com are two examples of such sites.
The idea behind the social/entertainment sites is similar to business/job networking sites. However, in this case, the context is more relaxed and informal, and sometimes involves dating and partner search. Here users are, in general, requested to list their friends so that, by browsing the social networks of them, it is possible to discover some friends of friends that might become friends. The first successful example was Friendster.com, soon followed by many other attempts.
News sites are centralized Web sites where users can submit news and stories and comment on them freely. The challenge is to keep the signal noise ratio high. Usually, the users can rate other users’ activities (posted news and comments) and these ratings are used to give more visibility to posts and comments the other users appreciate and value. Slashdot.org and Kuro5hin.org are two examples of this category.
The Web, the Blogosphere, and the Semantic Web can be viewed as decentralized news sites. They are systems in which anyone is free to publish whatever content at whatever time in whatever form. Different from the previous examples, in these systems, there is not a single, central point where content is submitted and stored, but the content is published in a decentralized way; for example, it is stored on different Web servers. The challenge in this case is to design a system able to collect all this content and find a suitable algorithm to quickly decide about its importance and value. Google.com was the first company able to achieve this and in fact, a large part of its initial success over search engines of the time is due to the PageRank algorithm (Brin & Page, 1998). PageRank’s assumption is to consider a link from page A to page B as a vote of A to B, or, in our jargon, a trust statement. The number of received trust statements influences the authority value of every Web page. In the following, we will review also how the concept of trust can be used in the Blogosphere (the collections of all the Web logs) and research efforts for introducing and exploiting trust in the Semantic Web.
The last category of systems that use and model trust is Peer-to-Peer (P2P) networks. P2P networks can be used to share files. P2P networks are, in fact, a controversial technology: large copyright holders claim they are used mainly to violate the copyright of works they own and are fighting to shut down this technology all together. We will not comment on this issue, but present the technological challenges faced by P2P networks. The open, autonomous, and uncontrollable nature of P2P networks in fact opens new challenges: for example, there are peers that enter poisoned content (e.g., corrupted files, songs with annoying noise in the middle) into the network. It has been suggested that a possible way to spot these malicious peers is to let every peer client express their opinions about other peers and share this information using the P2P network in order to isolate them. On a more positive take, a peer can mark as interesting (i.e., trust) another peer when it makes available for download many files that are considered interesting by its human user, or P2P networks can be used to share files only with a controlled and limited community of trusted friends.
Before going on with the discussion, it is worth mentioning that one of the first uses of the concept of trust in computational settings was in PGP (Pretty Good Privacy), a public key encryption program originally written by Phil Zimmermann in 1991. In fact, in order to communicate securely with someone using PGP, the sender needs to be sure that a certain cryptographic key really belongs to the human being he/she wants to communicate with. The sender can verify this by receiving it physically from that person’s hands but sometimes this is hard, for example if they live in different continents. The idea of PGP for overcoming this problem was to build a “Web of Trust”: the sender can ask someone whose key she already knows to send her a certificate confirming that the signed key belongs to that person. In this way, it is possible to validate keys based on the Web of Trust. The Web of Trust of course can be longer than two hops in the sense the sender can rely on the certificate received by someone who received it as well as from someone else, and so on. However, in this chapter, we are interested in the concept of trust from a more sociological point of view: trust here represents a social relationship between two entities, usually two users of an online system.
In the following, we present in more details different examples of online systems and, for each of them, what are the entities of the system and which social and trust relationships they can express in the system. We also analyze how trust is used by the system for providing a better experience to the user.
