-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 25.04.2014 17:27, George Kadianakis wrote:
Juha Nurmi juha.nurmi@ahmia.fi writes:
On 22.04.2014 17:35, George Kadianakis wrote:
Enjoy GSoC :)
I will :)
BTW, looking again at your proposal, I see that you are going to do both popularity tracking and backlinks.
Yes, another crawler gathers backlinks from the public WWW and I will start gathering the URL clicks from the users.
How are these two technologies going to interact with each other? That is, how will the indexer consider the output of those two features?
Django front-end re-sorts the answers from YaCy back-end.
See https://ahmia.fi/static/gsoc/re_sort.jpg
I have this idea in mind: https://ahmia.fi/static/gsoc/sorter.py
The result is sorted according to YaCy result index, number of backlinks and clicks which are scaled.
Note the scaling: p_info.backlinks = 1 / (float(index) + 1) etc.
sum_function = 3.0*self.yacy + 2.0*self.backlinks + 1.0*self.clicks
where 3, 2 and 1 are test coefficients. I will optimize these and made a better model if necessary. However, clicks are easily spoofed and there have to be small coefficient for them.
That makes sense.
BTW, what is the 'yacy' score? Is it just the order that YaCy's indexer chose for each result? Or does YaCy actually expose a score for each result? How is the score derived? Or do you treat it as a blackbox and assume it's the most accurate of backlinks and popularity.
I am using only the order information.
BTW, we (Mikko installed new servers) are migrating YaCy servers and took down the old one system. There should be a working crawler + fresh full text search results soon :)
- -Juha