ngs/middleman-blog-similar

[RFC] we need more debugging printouts

wkoszek opened this issue · 6 comments

wkoszek commented

Hello,

I enabled middleman-blog-similar on my website with this commit:

https://github.com/wkoszek/me/commit/2b8479586c3af100ea006b00f37a4a635f1d1658

But I don't have any indication why it picked certain pages. It somehow generated similar result for every page. You can observe it here:

http://www.koszek.com/

Click on any 3-4 links, scroll to "You may also like" section. Choices picked:

01001011, or on the art of snare drum patterns
How to write a good Google Summer of Code Proposals
Funny mistakes and The Toyota Way
Fixing Middleman-spellcheck
Why it's not about self-driving

and they're the same for every page.

ngs commented

The default algorithm looks up similar articles by strings from article body by spaces.

How about using TreeTagger instead?

wkoszek commented

I just tried TreeTagger. I have it enabled with:

+activate :similar, :algorithm => :'word_frequency/tree_tagger'

and I did the TreeTagger installation steps from the README.md. Results are similar. 1 thing I see:

  • blog entries tagged as articles all have the same "similar" articles
  • blog entries tagged as books all have the same "similar articles"

Is this normal? Do you have any other website where I could see middleman-blog-similar used?

wkoszek commented

@ngs ping. :-)

wkoszek commented

@ngs Ping. I didn't really get any better results with Tree Trigger. Do you think we could work on an improved documentation? I'd like to try 2 strategies:

  • random selection
  • semi-random based on tags of articles.
ngs commented

Refactored in #15 , This may not reproduce, please re-open if you see this error again. Thanks.