Ask HN: Recommend a recommendation system

[muchbetterguy]

I'm looking into the viability of using a recommendation system to add additional search results to a users' results list in cases where the search criteria may exclude results because that result does not have up to date data.

For example, searching for hotel rooms within a certain price bracket for a certain period and we don't have price and availability data for the period in question.

It would use previous searches by all users to show that if your results contain X,Y and Z, it's likely that result A and B would also be of interest. e.g. Previous searches where price and availability data was present.

I would think it'd work in a similar fashion to a regular recommendation engine for a shop or news site - if you read / bought these, try these.

Currently a Django / Postgresql / Elasticsearch based site

[nattaylor]

Maybe give PredictionIO a shot? There is a vagrant image that makes it relatively easy to give it a try. It wouldn't help with injecting the recs into search results.

[jvvlimme]

Predictionio is the glue between, amongst other things, mahout and elasticsearch if I'm not mistaking.

[infomofo]

You may also want to mention in your post what languages and frameworks you're already using.

If you have a java, scala, or clojure backend, using Mahout to make item to item recommendations is pretty simple to do. If you have a different language, I would just look into libraries for collaborative filtering in that language.

[muchbetterguy]

collaborative filtering + python leads me to graphlab - will investigate - Thanks.

[elchief]

smlar extension for pg

"Simple Recommender System" section: http://www.sai.msu.su/~megera/postgres/talks/pgcon-2012.pdf

http://railsware.com/blog/2012/05/10/effective-similarity-se...

http://blog.databasepatterns.com/2014/07/postgresql-install-...

[bra-ket]

directededge