Using Time-SVD++ in Dato Graphlab create

User 1798 | 4/22/2015, 5:59:24 PM

Hi Danny, For one of my current project I was panning to use time-SVD++ implemented in GraphLab. Mentioned in: 1." 2.

I was wondering if you can direct me to proper material to instruct how to select time-SVD++ in graphlab recommender create function. I could not find a reference to time-SVD++ in the API documentation in ""

I really appreciate your help regarding this.

Best Mahmudur


User 4 | 4/22/2015, 7:57:34 PM

Hi @mmrahman, the time-SVD++ factorization method from PowerGraph is not currently implemented in GraphLab Create. You can choose another method from the available methods in <code>graphlab.recommender</code>, or using <code>graphlab.recommender.create</code>, allow the toolkit to choose a method for you based on the characteristics of the training data.

User 1592 | 4/24/2015, 7:26:08 PM

Hi Mahmudur, The improvement in time-svd++ over SGD is that the time bin is taking into account to improve the quality of the recommendations. In GraphLab Create we have incorporated a much more advanced support of side features like recommendation time, user features like zip code age and item features like category price. Those capabilities are available in our factorization recommenders.

In other words, time-svd++ is only 3 dimensional (user/ item/ time) while we support way more general models.

I recommend trying it out and let us know the results you are getting.

User 1798 | 4/24/2015, 8:13:32 PM

Hi Danny, Thanks for the suggestions. I will try them in my project.

Thanks Mahmudur