Building models with SAS Enterprise Miner, SAS Factory Miner, SAS Visual Data Mining and Machine Learning or just with programming

How to build recommendation system in SAS E-Miner

Accepted Solution Solved
Reply
Contributor
Posts: 42
Accepted Solution

How to build recommendation system in SAS E-Miner

Dear Consultans,

 

I need to know how to build recommendation system in SAS E-Miner ?

 

Thanks


Accepted Solutions
Solution
‎07-11-2017 01:46 PM
Super Contributor
Posts: 336

Re: How to build recommendation system in SAS E-Miner

Hi Hussein,

Apologies in advance but the answer is: it depends.

 

Most people think the go-to algorithm for recommendation engines should always be collaborative filtering. This algorithm is very famous becuase it was used to solve the Netflix challenge. This means that if you are trying to solve a similar problem, you should probably start with collaborative filtering, which is not available in Enterprise Miner.

 

However, that does not mean that you cannot build a recommendation engine based on some other algorithm. Several customers build very successful recommendation engines based on measures such as simmilarity, nearest neighbor, or SVDs (who would think Text Mining nodes calculate those!?). If you need to update your model constantly, you may not want to use EM, unless you are realy good at programming and running Batch code.

 

The painful part is that you might have to try several algorithms and there is not a "Model Comparison" node for which one suits best your recommendation purposes. But as often happens in data mining, you learn as much from what does not work as you do from what actually works.

 

If I had to start somewhere, I would try the association node (if my items are in a transaction format), or the link analysis node. Here a paper that vouches for using other algorithms and other methods: www.springer.com/cda/content/document/cda_downloaddocument/9780387858197-c1.pdf

 

Good luck!

-Miguel

View solution in original post


All Replies
SAS Employee
Posts: 122

Re: How to build recommendation system in SAS E-Miner

husseinmazaar ,

As of now, there is no direct product feature or node in EM that does recommendation, say, collaborative filtering, although one may very well use Memory Base Reasoning (another name for KNN) and clustering capability to build out recommendations or next-best-offers. SAS in-memory product IMSTAT sports explicit Recommendation facility. Hope this helps. Best Regards. Jason Xin
Solution
‎07-11-2017 01:46 PM
Super Contributor
Posts: 336

Re: How to build recommendation system in SAS E-Miner

Hi Hussein,

Apologies in advance but the answer is: it depends.

 

Most people think the go-to algorithm for recommendation engines should always be collaborative filtering. This algorithm is very famous becuase it was used to solve the Netflix challenge. This means that if you are trying to solve a similar problem, you should probably start with collaborative filtering, which is not available in Enterprise Miner.

 

However, that does not mean that you cannot build a recommendation engine based on some other algorithm. Several customers build very successful recommendation engines based on measures such as simmilarity, nearest neighbor, or SVDs (who would think Text Mining nodes calculate those!?). If you need to update your model constantly, you may not want to use EM, unless you are realy good at programming and running Batch code.

 

The painful part is that you might have to try several algorithms and there is not a "Model Comparison" node for which one suits best your recommendation purposes. But as often happens in data mining, you learn as much from what does not work as you do from what actually works.

 

If I had to start somewhere, I would try the association node (if my items are in a transaction format), or the link analysis node. Here a paper that vouches for using other algorithms and other methods: www.springer.com/cda/content/document/cda_downloaddocument/9780387858197-c1.pdf

 

Good luck!

-Miguel

🔒 This topic is solved and locked.

Need further help from the community? Please ask a new question.

Discussion stats
  • 2 replies
  • 966 views
  • 0 likes
  • 3 in conversation