Social ranking and classification
Social ranking and classification means rating and classifying search results by users. This may help Wikia Search not only improve the quality of the results produced but also personalize them, based on how the user rated similar pages or those connected in other ways.
[edit] Recommendation systems
Algorithms used in recommendation systems by e-commerce Web sites might be successfully implemented in the engine. The most popular of these are:
- Collaborative filtering - predicting users' choices based on actions performed by similar users (those who rated similarly or purchased similar products).
- Item-to-item collaborative filtering also known as Item based filtering - designed by Amazon do deal with large datasets; it searches for similar products, not users.
- Cluster models - classifying users into groups; all users in the same group get the same recommendations; said to give recommendations of poor quality[1].
Cluster models, Item-based and topic-based filtering can be combined to detect topical or social foci. These foci can help to build recommendation systems, if the system is able to build bridges between existing focis and almost-new users [2].
Such systems have many disadvantages. Not being able to recommend any item to new users (since they haven't performed any action yet) and small chances of finding similar users are probably the most serious. Many of them might be overcome by social networks based on trust [3].
It is the art of social ranking and classification to circumvent these problems. In a search engine, we don't have "new" users, because the first action each user does, is entering a keyword. Thus we know already a bit of this user. We can not only use this information for finding results in the index, but also for suggesting alternatives and specifications the user might want to search for or many users searched/used before. (similar to the Item-to-Item collaborative filtering concept)
With social classification we can also allow users to add tags to links themselves, so the algorithm does not have to be so smart to really understand the website. To reduce spam, these tags can be ranked either be users or by the amount of similar tags in neighboring websites.
[edit] Notes and references
- ↑ Elizeu Santos-Neto: Content Reuse and Interest Sharing in Tagging Communities, Jan 2008
- ↑ Bielenberg, K. and Zacher, M., "Groups in Social Software: Utilizing Tagging to Integrate Individual Contexts for Social Navigation", Masters Thesis submitted to the Program of Digital Media, Unisersitat Bremen (2006)
- ↑ Massa, Paolo. "Using Trust in Recommender Systems: an Experimental Analysis", 2003
[edit] External links
- http://www.trustlet.org/wiki - wiki about trust metrics on social networks.