|
Wikipedia Suggest |
|
| Suggest: English | ||
| Suggest: Deutsch | ||
| Suggest: Français | ||
| Suggest: Polski | ||
| Suggest: 日本語 | ||
| Suggest: Nederlands | ||
| Suggest: Italiano | ||
| Suggest: Svenska | ||
| Suggest: Português | ||
| Suggest: Español | ||
| | ||
| Download | ||
| Project history | ||
| FAQ | ||
| Web statistics | ||
| | ||
| Personal Web Site | ||
History
July 2006
I started the development of wikipedia-suggest using XML articles produced in july 2006. The idea was to rank the list of results with a link analysis approach, each article votes for its favorite documents (using the link to other articles it contains and the frequency of this link). After this analysis, each article has a rank (number of link to them) and can be added by title in a trie (finite state machine). Each node of this trie contains a static table of best articles for this position.The figure below shows an example of trie with a static table associated to queries "a", "w", and "wi".
- The first one (DiskQuery) is a mono-thread implementation using very few RAM (only 16M of RAM for the whole automaton of english articles) but perform the query in a random access disk file. This solution is ideal if you have a server and very low RAM available for this service (this is the solution I use on my web site)
- The second version (MemoryQuery)is a multi-thread capable implementation allowing a lot of queries per second on a standard server. This version use about 200Mb of RAM for the whole automaton of english articles
All these programs are available under GPL licence in this [archive].