New contact recommendation filtering

[letnotimitateothers]

From now, recommended new contacts to a user are
---> the contacts of its contacts
---> we use (FOAF('knows'), FOAF('knows)) in the function followPath in the lib rdflib-stample-pg...

Some questions arises :
---> what new parameters the algorithm of recommendation should consider ?
---> what is the best number of jumps to do down the graph to discover new contacts for recommendation ?
---> how to filter the contacts in the recommendation ? Not showing the same contacts again and again; not showing new contacts with wrong URIs
---> how to paginate the contacts for recommendation ?
---> filter new contact without URIs or with wrong URIs
(issue description not finished yet)

[bblfish]

Recommending users requires some process the social graph, which is expensive. So there should be a fixed number of visible recommendations, but they should as much as possible always be different. This implies searching the graph randomly, up to a certain depth, and until a certain number of people have been found. The criteria for proposing people, should be that the contact information is of some form of minimal good quality:

• BNode Agents should probably be given second place ( unless they contain a lot of good information such as phone number, email or home page)
• the jump for WebID agents should work ( so that one can recuperate picture and other information )
• One should show how many people in one's network are connected to someone, double connections count as verifications ( x knows y said by x. and y knows x said by y )

[slorber]

Bnodes are already ignored and we ignore unjumpable profiles.

It is not really easy in RxJs to stop the processing of a stream I think so
even if you only take the first 5 results the background computation may
continue, even if you unsubscribe the stream.
I think the concept of stream backpressure is relative to that, and it is
said that Rx* is not good to handle backpressure (this is why some people
prefer Iteratees in Scala for exemple)

2014-02-13 15:27 GMT+01:00 Henry Story notifications@github.com:

Recommending users requires some process the social graph, which is
expensive. So there should be a fixed number of visible recommendations,
but they should as much as possible always be different. This implies
searching the graph randomly, up to a certain depth, and until a certain
number of people have been found. The criteria for proposing people, should
be that the contact information is of some form of minimal good quality:

• BNode Agents should probably be given second place ( unless they
  contain a lot of good information such as phone number, email or home page)
• the jump for WebID agents should work ( so that one can recuperate
  picture and other information )
• One should show how many people in one's network are connected to
  someone, double connections count as verifications ( x knows y said by x.
  and y knows x said by y )

Reply to this email directly or view it on GitHubhttps://github.com//issues/15#issuecomment-34982453
.