Because anyone can edit Wikipedia, the web encyclopedia's reliability varies wildly. Now a computer science professor hopes to give users a better baloney detector: software that flags questionable lines in Wikipedia entries.
Developed by Luca de Alfaro and colleagues at the University of California, Santa Cruz, the software will colour text some gradation of orange if there is reason to doubt its content. The deeper the orange, the more likely it is malarkey.
How does the software have any clue about that? Mainly by analyzing the reputations of the contributors responsible for each line.
By diving into Wikipedia's open volumes of edit histories, the software counts the degree to which any given contributor's work survives subsequent edits by other people. In general, the less tinkering your work on Wikipedia engenders, the more trustworthy you are deemed to be.
That system is not foolproof, as accurate contributions might get quickly overwritten in articles on contentious topics.
Even so, it does yield interesting insights. For example, in an extensive entry on the old Commodore 64 computers — the Santa Cruz software tags just three lines, each an unfootnoted statement of purported fact.
For now, the software — available at trust.cse.ucsc.edu — is in demonstration mode and operates on an older subset of Wikipedia entries. However, de Alfaro showed it to a receptive audience at last month's Wikimania convention in Taiwan, and he hopes to work with the Wikimedia Foundation, which oversees Wikipedia, to make it a real-time option on the site.
Foundation spokeswoman Sandra Ordonez said there have been no official decisions.
"I don't want to give the impression that Wikipedia is low-trust. It really works very well," de Alfaro said. "What I wanted to make sure is that nobody can single-handedly modify information without some trace of that being available for some time afterwards."