Most recent edit on 2004-07-27 19:39:33 by DarTar []

Additions:

Deletions:

---

Edited on 2004-07-27 19:38:39 by DarTar []

Additions:

Deletions:

---

Edited on 2004-07-27 18:49:46 by DarTar []

Deletions:

---

Oldest known version of this page was edited on 2004-07-27 18:31:22 by DarTar []
HomePage

---

Authority

1. From relevance to reliability

The web can be described as a complex system of distributed knowledge, whose main actors are knowledge providers (or sources), knowledge mediators and knowledge users.

The main action a user performs for retrieving information in the case in which direct access to sources is not possible is to query a knowledge mediator. By querying a knowledge mediator (e.g. a search engine, a directory, or any other infrastructure offering services for mediating access to information sources), a user is generally trying to apply the most accurate strategy she knows for rapidly retrieving qualitative information relevant to her needs. Consequently, there are at least three criteria that an optimal strategy must satisfy:

1) Costs required to have access to sources (e.g. time)
2) Relevance of results
3) Reliability of results

Beside costs, the first problem that is raised by this epistemic relation between a user and a knowledge mediator is thus the way in which relevance is assessed, i.e. the way in which the user selects among possible queries and among available sources the most relevant or the ones that match closely her informational needs.

But as soon as more than one relevant source of information is available, a much harder problem arises: the problem of selection of reliable content. A user can ideally have perfect knowledge of how to maximise relevance of a query or how to discard non relevant content provided by a knowledge mediator. But in most cases, unless she has previous acquaintance with the sources, the user has no a priori clue on how to select the most reliable information provider.

2. Authority and deference: reliability as a social construction

Assessing reliability of information is much more complicated than assessing relevance, in that in general the user cannot establish reliability on her own: in most cases reliability judgments require deference to the epistemic authority of thirds.
Epistemic authority is the property a subject, an institution or another source of knowledge has to orient an agent's trust on specific beliefs.

If a doctor diagnoses the presence of a disease and is contradicted by another doctor, the patient's choice of the more reliable diagnosis will probably depend on a different deferential attitude towards them, namely on a comparison of their respective epistemic authority. It is virtually impossible for the patient to have an exact measure of either's diagnosis reliability, since this would require knowing the scientific titles of both doctors, the percentage of false and true diagnoses they respectively formulated, the success of therapies they previously prescribed to other patients etc.

Hence, factors affecting one's epistemic authority are very complex and must be studied as globally emergent phenomena of local interactions among users: my act of trusting a doctor is a local phenomenon that is modulated by and that can virtually modulate a complex network of other social relationships.

We assume that reliability of information is a function of the epistemic authority of its provider and that authority is the result of a complex network of social interactions.

In the web, considered as a system of distributed knowledge, authority becomes even more crucial, in that unmediated access to sources is in most cases impossible. There are of course also in the case of the web many external factors affecting authority of sources of knowledge, but, as many have argued, a fundamental factor is hidden in the topological structure of the web.
A site receiving a large number of links can be considered, under certain assumptions, as an authoritative source of information. What are the reasons for considering a large number of inbound links as a reliable cue for establishing a site's authority?

3. Linking as trust giving

There are of course many reasons why a site may link to another. A personal homepage can contain links to homepages of friends, as well as links to pages of interest for the site's owner. An academic homepage usually contains links to affiliated institutions and scientific partners as well as links to research projects or resources related to the site's owner's work. A commercial portal contains advertisement that usually consists in images or messages linking to the advertiser's site (like banners). A site advocating against a particular political campaign can contain links to texts of reference of this campaign. The fine-grained analysis of the social relationships hidden behind the linking practice shows that linking cannot be straightforwardly identified with a particular function.

Nevertheless, it is undeniable that linking relationships determine clusters of sites that contain valuable information about epistemic deference. In most cases links can be compared to quotations of authoritative content: the fact of adding a link from site A to site B often corresponds to an implicit declaration that content of the linked page (B) is a reliable source, or reliable integration for content of the original site (A).

If locally, the consequences of linking as trust giving are not important, at a more global level they become much more salient. Consider the example of two distinct sites on the same topic referring to a certain number of external sites. The fact that an external site may be jointly referred to by two distinct sites dealing with the same topic can be taken as an example of joint epistemic deference. In general, the larger the number of referrers, one might be tempted to say, the highest the alleged epistemic authority of the referred site.

It is not enough to consider the number of incoming links.
- spam
- self-quotation
- St. Matthew effect


So, why is the system fair enough and refractory against manipulations?

- distribution and recursive definition

4. Categories of trustees

Hubs and authorities

5. Linking: consequences on users behavior and on web structure

Linking per se directly increases the probability a user has to access a site. But at the same time linking, by modifying the topology of the web, also affects indirectly the way in which the user selects information.

6. Bibliometrics and the web

Differences between bibliometrics and cybermetrics.


Self-quotations


St Matthew effect


Spam


Attrition, inercy and average life of websites


Incremental authority in bibliometrics


 



Current projects:


E - Authorities
(Roberto Casati, Christophe Heintz, Gloria Origgi, Barry Smith, Dario Taraborelli)