I remember the time when I first stepped into the world of Wikipedia, I didn't where to start. There are too many policies, too many rules, too many pages to scroll through, to even get a handle on what to do. This puts a lot of new editors off. The welcome templates, even with cookies and kittens, are often insufficient in retaining new editors. Furthermore, the IRC help, Help Desk, and help pages of different projects (such as the AfC), are often flooded with questions. This takes out a lot of time from active editors.
At the moment, this system is OK. But it can be improved upon. Like how an increasing number of services online have an 'online chat' function, where if there's something you don't quite get or have a question, you will be allocated a person to talk to. I think this is a much more personal way that is attractive to new editors.
This can be done through a group of volunteers, like it is at the moment with the IRC, but in a way which is much more accessible (so will be available in every page, unless a user ops-out). However, this again will take up the valuable time of editors. What I propose then, is to create an AI bot, which will take the questions asked, decipher them, and give a reply - not link them to a policy page, but a reply that is 'human' and conversational. I propose a tab, such as the 'feedback' tab in this app, to be used on every page.
- To create an AI bot that will help answer questions editors will have, in a conversational and friendly interface
- To increase the retaining rate of new editors to the Wikimedia projects
This sounds like a query answering or interrogating dialog system, very similar to "Betty", but for the project and help pages and not for the command line. Betty is written in Ruby, but it does not seem to be very difficult to rewrite this in PHP. The problem description is although a little wider than what Betty is made to do. The definitions could be embedded in the pages themselves, but be hidden during normal surfing. A more general tool could try to find snippets on non-tagged pages, but then the problem would be to limit the output so the extraction would still be useful comparing to go to the actual page.
As a search construction engine it would be pretty simple. If any interrogating word is found, then a search is done in the meta-pages of the site. Pages found will then be summarized on content found on the previous search. The summarizer will mostly just try to exclude non-relevant content. See also Question Answering and Summarization in Speech and Language Processing, ISBN 978-1-292-02543-8.
A variant of a dialog system could use a dialog marker in the text that list topics for a summarizer. Topics together with a pseudonym list could increase the recall a lot. Pseudonyms could also be bound to pages, but this could also be done by redirects.
I think this is doable and could be a nice addition to Wikipedia. — Jeblad 16:33, 2 November 2014 (UTC)
Welcome, brainstormers! Your feedback on this idea is welcome. Please click the "discussion" link at the top of the page to start the conversation and share your thoughts.