search

ynuwm / spam-detection

Implement of paper Learning to Represent Review with Tensor Decomposition for Spam Detection
9 stars 4 forks source link

Have you run this codes on the restaurant data ? #1

Open Jack-Ray opened 5 years ago

Jack-Ray commented 5 years ago

I have been interest in the spam detection project recently, so I used your code to implement this paper "Learning to Represent Review with Tensor Decomposition for Spam Detection". And I found the speed was too slow and it would take a long time to get the 11 relationship under the restaurant data. And we should use RESCAL-ALS method in order to get the matrix B and this step takes a long time (more than 5 hours under the restaurant data). So I want to know whether there is a faster method?

ynuwm commented 5 years ago

@Jack-Ray First of all, thank you very much for your interest in my spam detection project. It’s been a long time. I remember running this on my desktop and Nvidia GTX 1060 for a long time, so you can see that I have kept a lot of intermediate results in my code. If you want to understand the matrix decomposition, you can directly look at the RESCAL-ALS source code, because I first touched the matrix decomposition and so I quoted this source code. Of course, you can also write a matrix decomposition method by yourself, which may be much faster. You can also see some things about how to quickly create and store numpy arrays so you can get eleven relational matrices quickly. Finally, if you find a better way, you can also tell me, thank you.

Jack-Ray commented 5 years ago

Thanks for replying! I think I should read the source code of RESCAL-ALS and try to rewrite it. Maybe it can work.