Hi,

For example, if you're dealing with document classification, apply cross-validation normally once converting your data into a bag-of-words.

HTH.

Samer

Hi Samer Sarsam,

Thank you for your answer.

I have a small dataset (650 items, 20 users, 700 ratings).

Each item is represented by features. I performed a TF-IDF in order to calculate the weight of each feature in the item.

To learn the user profil, I performed a simple average weights of the features belonging to list of items liked by the user in the past.

I want to know if the weights learned, for each user, gives a good results.

Can I apply a cross-validation in this case ?

Thank you

Cross-validation is an evaluation approach that can be used to assess training process. Hence, you can use it in your task to accurately evaluate your classifier model.

HTH. 

Samer 

From what I know, the cross-validation is generally used to tune a classifier, by executing it with different parameters. The results of each execution are compared to the validation-set. At the end, we hold the parameters which gives good results.

But in my case, I'm little confused about the parameters. What are the parameters in my algorithm ?

As I mentioned, cross-validation itself is an evaluation method that can be applied in general learning process. However, some schemes adopt this method for "Parameter Selection" (which is different) as it can optimize over an arbitrary number of parameters, with only one drawback. 

In regards to your case, it depends on the algorithm you're using and whether such optimisation is really required (it's our task to know that). 

HTH. 

Samer 

Actually, I just start with simple averaging of the liked items by the user.

But it seems to be too simple to produce good result.

Is the Rocchio's classifier good for this type of task ? I don't know if I can use it by choosing "Positive Rated Class" and "Negative Rated Class"  to perform the clissification.

No one can say what is the best classifier, you need to try several classifiers in order to discover the one that is really reasonable for your task.

Cheers, 

Samer 

Can you please suggest me a good materials on applying rocchio's to content-based recommender ?

Thank you for your yelp.

Please have a look at the following articles:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.724.1014&rep=rep1&type=pdf

http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0158423

You could also try those two classifiers for document classification: 

Multinomial naive bayes and stochastic gradient descent.  

Cheers,

Samer

Thank you for your help !

Hi Samer Sarsam,

In the first article: http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0158423

They used Rocchio. But I don't undestand the following statement:

"Once the query vector q has been defined for a user, a nearest neighbor search will be performed to retrieve suggested elements."

Why they have used nearest neighbor ?

Thank you.