Apache Mahout as Recommendation Engine

Question

I want to use Apache Mahout as Recommendation Engine; but over here I found that it force us to use its own table called taste_preferences with only 3-4 columns and data type as number(Long/big int). Is it mandatory to use this table and store data in number format only.

Answer 1

That is one way to build a recommendation engine, but there are simpler ways as well.

There is a small book available for free from

http://www.mapr.com/practical-machine-learning

which explains a way to deploy recommendation engines on top of a search engine. This requires an off-line analysis to build the data that gets put into the search engine, but once you have the indicator data in the search engine, you can do recommendations using search queries. These queries are not textual queries, but instead use past behavior as a query.

You can also see slides describing the approach here:

http://www.slideshare.net/tdunning/building-multimodal-recommendation-engines-using-search-engines

and here:

http://www.slideshare.net/tdunning/using-mahout-and-a-search-engine-for-recommendation

The book is easier to understand than the slides without the narrative, but both are likely useful since the slides have more details.