Suggestions feature

[Milan Dobrota <mi...@milandobrota.com>]

I am working on a feature similar to Youtube suggestions (where the videos
are suggested based on your viewing history).
What I do is parse the history and get the user's interests, in the form of
weighted topics. When I boost according to those interests, the dominant
ones take over the result list.

Is there any way to use the boost so that there is some variety in the
results?

Thanks
Milan

RE: Suggestions feature

[Steven A Rowe <sa...@syr.edu>]

Hi Milan,

I have three ideas:

1. Boost by log(weight) instead of just by weight.  This would reduce weight-to-weight ratios and so reduce the likelihood of hit list domination, while still retaining the user's relative preferences.  Multiple log applications will further decrease the weight-to-weight ratios and thus increase variety.

2. Take the top N topics by weight, thresholding either by some arbitrary weight or at some arbitrary N, and then boost those N topics equally.  This would increase variety at the expense of ignoring the user's minor interests.

3. Just apply a fixed boost to all of the user's interests and ignore their associated weights.  (This is equivalent to taking strategy #1 to the limit.)

Steve

> -----Original Message-----
> From: Milan Dobrota [mailto:milan@milandobrota.com]
> Sent: Tuesday, October 04, 2011 12:56 PM
> To: solr-user@lucene.apache.org
> Subject: Suggestions feature
> 
> I am working on a feature similar to Youtube suggestions (where the
> videos are suggested based on your viewing history).
> What I do is parse the history and get the user's interests, in the form
> of weighted topics. When I boost according to those interests, the dominant
> ones take over the result list.
> 
> Is there any way to use the boost so that there is some variety in the
> results?
> 
> Thanks
> Milan