I have a data set with 100,000 instances and about 40 features. Each instance is a customer and each feature is a property of the customer. The first column is binary 0/1 which indices whether the customer click the ads. The task is fitting the data with models or one model and predicting if a new customer will click the ads or not.

I want to start with only one kind of model to do this but don't know which model is suitable. I can think svm (with libsvm), logistic regression. I don't think matrix factorization methods can help because the columns are features of the customer, not items.

  • $\begingroup$ Can you provider more informations about the items in this recommendation system? Do you have only one Ads or multiples? $\endgroup$ – imanis_tn Nov 20 '15 at 6:07
  • $\begingroup$ @imanis_tn Just only one ads. And I agree with Adrien Renaud. May be I need a classifier, not a recommender system. $\endgroup$ – Xiang Zhang Nov 20 '15 at 8:19

Given the nature of the data shown in your question, it seems that your are looking for a classifier algorithm and not a recommender system.

A recommender system would answer the question: "What particular ads to show to this particular user". Your question is really different: "Will a new customer click the ads or not".

To build a recommender system, you need to record interactions between users and items. In your case, it seems that you have only one item. Intuitively it seems clear that you can't build a recommender with one item.

With only one item, you are good to use classifier algorithm. Choosing a model for a binary classification problem is a very broad subject. You mentioned two very good examples, svm and logistic regression. I would add a third one that is very often used: boosted decision trees like in adaboost or xgboost.

| cite | improve this answer | |
  • 1
    $\begingroup$ ...beyond classification some kind of probability estimate of the customers clicking the add or not may be useful to optimize the gains of a given strategy. That should be attainable by all recommended models. $\endgroup$ – Soren Havelund Welling Nov 19 '15 at 21:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.