Hi Siwei, We certainly appreciate seriousness and commitment. You can perhaps just give more details to make it clear to us that each task is given the appropriate amount of time.
With the vague expression "close the feedback loop" I was alluding to other e-mails I have exchanged publicly in the list, a demo paper that we've published [1], as well as fragments of text I've shared on the ideas page. Sorry for assuming that such overly vague citation would be immediately clear. 1. What's "feedback loop"? Do you mean that a user can submit > feedbacks again and again to make a loop? > By that I mean that: 0) Training Set = Wikipedia 1) the system learns from training set how to automatically annotate text 2) given input text, the system automatically annotates text 3) the user fixes the annotations that were wrong 4) the system stores the feedback 5) Training Set = Wikipedia + User Feedback; Go to 1. 2. What's "linker learning"? Do you mean the links among the > feedbacks, or among the feedbacks and the users? How to close a loop > by a linker learning? > I'm referring to the step 5 above. How to leverage feedback for enhancing the annotations next time around? Should you train a classifier for detecting wrong annotations? Which features would you use? Would it be better to use some rule-based approach? How? "Learning a linker" is a way to leverage feedback to filter out bad results in the next iteration. To close a loop by learning a linker is to be able to get better every time you loop back from step 5 back to step 1. I hope this clears up things better. Cheers, Pablo [1] http://www.wiwiss.fu-berlin.de/en/institute/pwo/bizer/research/publications/Mendes-Heder-SztakipediaSpotlight-WWW2012.pdf On Wed, Apr 4, 2012 at 2:21 PM, Siwei Yu <[email protected]> wrote: > Dear Pablo, > > Thanks for you advice. I'm composing the project proposal, especially > the project plan. I need to evaluate the work time for the tasks. I'm > not the guy who promises many but can not make it indeed. Once > promised, I'll stick to my words to complete them with responsibility. > I'll see what I can do. > > Could you tell me more about the idea of "closing the feedback loop by > having a linker learning from the feedback"? I'll try to enhance the > feedback service task. > 1. What's "feedback loop"? Do you mean that a user can submit > feedbacks again and again to make a loop? > 2. What's "linker learning"? Do you mean the links among the > feedbacks, or among the feedbacks and the users? How to close a loop > by a linker learning? > > It's greatly appreciated if you can show some examples. Thanks. > > Best regards, > Siwei Yu > > On Sun, Apr 1, 2012 at 5:50 PM, Pablo Mendes <[email protected]> > wrote: > > Hi Siwei, > > It is fine with me to push forward Chinese support as a task for you to > do > > as a member of the DBpedia Spotlight team instead of as a GSoC student. > We > > are always interested in good people joining the team. But I am afraid > that > > the effort needed for adding DBpedia Spotlight engines to Stanbol will > not > > be 1.5month full-time, neither will the /feedback service, therefore your > > project looks a bit smaller than the others. This could be a good thing, > if > > you manage to show in your application that you know exactly what needs > to > > be done ,and that it requires that much time. But it might also make your > > project weaker if you fail to convince us of that. > > > > Perhaps closing the feedback loop by having a linker learning from the > > feedback is what you need in addition. > > > > Cheers > > Pablo > > > > On Apr 1, 2012 7:28 AM, "Siwei Yu" <[email protected]> wrote: >
