Erste Seite Zurück Weiter Letzte Seite Übersicht Grafik

Rating systems


Notizen:

There has already been a lot of research into systems which are able to make ratings with the goal to guide the user. Content based filters examine the content of a message and often have rules upon which their ratings are based. Such systems can only be used with textual content and not with multimedia data in general.
Collaborative filtering overcomes this limitation by letting the users decide which information is interesting to them. Through a connexion of many users, the burden of manual filtering is divided among the community. With a large number of users, the opinions of the users might drift apart, rendering some ratings useless or even deceptive to others. This problem gets worse once people consciously enter false ratings or try to influence it to their own advantage. We might run into problems of social scalability.
For this reason, reputation systems introduce the notion of trust into large groups of people. They usually collect, distribute and aggregate feedback about the past behaviour of the users. An example is eBay, where the buyer and the seller can rate each other after a transaction. Because everybody will only make deals with people with good reputation, most users behave honestly.
Thus, if a Freeporter had a good reputation, I would probably be interested in his or her next reports. The reports of a Freeporter with a bad reputation would be of little interest to me.
But there is a difference between finding an honest seller and finding interesting Freeporters. Especially, since eBay is collecting global reputation. The Freeporters would need local reputation which is dependent on everybody's individual points of view. The users would have a different reputation, depending on who is asked.
Such a system can be facilitated with a web of trust.