Thank you for being a valued part of the CNET community. As of December 1, 2020, the forums are in read-only format. In early 2021, CNET Forums will no longer be available. We are grateful for the participation and advice you have provided to one another over the years.

Thanks,

CNET Support

General discussion

recommender system

May 8, 2012 7:35AM PDT

im a phd student in the second years, i have to implement user-based collaborative filtering using traditionnal approach in matlab for producting recommendation and Mesuring the performance of the algorithm with MAE, i have some problems :when i use the whole matrix of the data base movielens, the time of computing is very long,mode busy
and the MAE is superieur than 1, what can i do, please?

Discussion is locked

- Collapse -
So let's start with what you use now.
May 8, 2012 7:41AM PDT

And what you feel would be better.

- Collapse -
PS. At the office others have this.
May 8, 2012 7:46AM PDT

The brutal truth is that you could find the fastest machine out there and still do something in Matlab that will take days. It's a balancing act of money, time and such.
Bob

- Collapse -
recommender system
May 9, 2012 7:46AM PDT

i want to discuss with people working in the field recommender system, exactely user-based filtering or item-based , because my professor is not someone from domaine.

for me ,i do this in matlab. i compute the MAE(mean absolute error) to measure the performance of the algorithm that produce the recommandations .i dont have the result, the program is basy, i dont have any idea , how much time to do this .

I dont know the good way to do this , how people of this field measure MAE ?if they use matlab or another things, if they use the whole matrix user_item matrix (943users, 1682 movies) (the matrix is very sparce) or there is a another dataset for test to compute MAE.
Can you please help me solving this problem if you are working in recommender system? Thanks in advance!

- Collapse -
So you are copying this paper?
May 9, 2012 8:04AM PDT
- Collapse -
yes
May 15, 2012 3:45AM PDT

i would how we do this ? not to copying the paper smart boy, this the traditional approch

- Collapse -
What is the coincidence of picking those numbers?
May 15, 2012 4:01AM PDT

It's so much a coincidence that either you are copying a paper or it's homework that you get to do on your own?

- Collapse -
RS
May 15, 2012 6:26AM PDT

all of people work with this database, those are not numbers , thanks for your help. no time to weast with you.perhaps that you manear is copying, by this way you think that all people do the same. you are terrible .

- Collapse -
Name some names? Such as which college and such?
May 15, 2012 6:45AM PDT

The coincidence of those numbers and that paper are too coincidental.

I agree that "all of peopler work with this database" since the course has to give everyone the same outdated data.

I'd like to discuss this further but it appears you are hiding something.
Bob

- Collapse -
So it seems ...
May 15, 2012 6:58AM PDT

your results show that the traditional approach takes a lot of time when doing it in Matlab.

So maybe you conclude:
a. The traditional approach is not the best one available
or
b. Matlab isn't the best tool available for the traditional approach
It's a nice subject for a phd course to find better approaches or better tools, isn't it?

Moreover, if your purpose is to "measure the performance of the algorithm", as you state in your first post, the time it takes is subject of the study. So why care if it takes long? That's a result also.

The interesting thing now is to determine the relationship between the size of the matrix (vary the number of users or number of movies) or the percentage of filled cells (it's very sparse now, you write) on one hand and the time needed for the algorithm on the other hand.

Kees

- Collapse -
Re: MATLAB
May 8, 2012 7:42AM PDT

This is both a very specific (Matlab, MAE) and very general (takes a long time) and very unclear (MAE > 1, no info at all about how you do it) question.

So it seems best to discuss this with your professor.

Kees

- Collapse -
recommender system
May 9, 2012 7:40AM PDT

yes very general, specific and not clear for all people.
i want to discuss with people working in the field recommender system, exactely user-based filtering or item-based , because my professor is not someone from domaine.

for me ,i do this in matlab. i compute the MAE(mean absolute error) to measure the performance of the algorithm that produce the recommandations .i dont have the result, the program is basy, i dont have any idea , how much time to do this .

I dont know the good way to do this , how people of this field measure MAE ?if they use matlab or another things, if they use the whole matrix user_item matrix (943users, 1682 movies) (the matrix is very sparce) or there is a another dataset for test to compute MAE.

- Collapse -
Re: recommender system
May 9, 2012 7:50AM PDT

As you can see, the Coding & Scripting forum is a forum in the Web Development section of these forums. And I'm quite sure that your issue has nothing to do with web programming. So this surely is quite the wrong forum for you needs. I don't expect any useful answers.

Of course, it doesn't help that it's still very unclear what the mean absolute error in (the performance of?) an algorithm that produces recommendations based on a matrix of users and movies exactly is and how you calculate it. It might be that Matlab isn't the right tool. Maybe using another tool or another way of calculating it in this tool would be more suitable.

Kees