I was planning to make an artwork recommendation system as a project by using the WikiArt open source dataset available on kaggle, I'm still looking for datasets which might already have user ratings but is it possible to make a recommender without any user ratings? The dataset currently has these features. Any help or insight is appreciated. Thanks. enter image description here

  • 1
    $\begingroup$ yes. You can recommend by similar artist, genre, or style; add in more data like subject matter and your recommendation engine gets even better $\endgroup$ – Mohammad Athar Dec 12 '18 at 18:46

Generally speaking, there are two types of recommender systems. One is community-based, the other is content-based.

In community-based, you are effectively telling a user, "you are a young female, here is content that other young females enjoyed".

In content-based, you are effectively telling a user, "you liked content A, here is content B that was liked by people who also enjoyed content A".

So, in your case, you would be looking to build a content-based system. That means that you will have to build a meta-data set based on user actions throughout your site. When someone watched content A, what did they watch after that? And then after that? What content made them leave? What content made them sign up for your site or buy something or do some other action on your site? What categories of content are likely to be watched together?

When you can gather that type of metadata then you will be in a better position to create a model. Note that I believe that you can extract this from the sample data you posted, it's just a matter of how you organize it.

  • $\begingroup$ But I don't have any user data to build a meta-data set, If I were to just use this dataset(what I have currently), any tips on how to organize it? $\endgroup$ – Kartikay Dec 13 '18 at 15:35

First you need to clarify what kind of recommendation you want to make.

The reason for the rating is to be able to associate an item to a user preference. You can avoid a rating and design the recommendation as a classification problem where you classify a product to some event (someone would like the artwork or not), but this still needs you to design a label based on some information.

In summary yes you can do without a rating but you still need some information on user preference or whatever the label will be.

  • $\begingroup$ I'm a noob, please bear with me. I was wondering what kind of recommendation can be made, since there is no user data, so user based collaborative filtering is out of charts, so is item-based collaborative filtering possible? if yes, then what should I take into consideration? Or is a content based recommendation possible as far as artworks are concerned? Because what I want is like when I want to get recomm for Starry Night, I get other works of Gogh or some other work of similar art style or movement. But I guess I can't get a recommendation on basis of popularity because no user data. $\endgroup$ – Kartikay Dec 12 '18 at 15:42
  • $\begingroup$ @Kartikay Sorry, but I disagree with this answer and have downvoted it. This answer only scratches the surface of recommender systems and shouldn't be relied on. Please see the answer I provided. $\endgroup$ – I_Play_With_Data Dec 12 '18 at 18:32
  • $\begingroup$ I really dont undesrstand why someone downgrades a correct answer just because he doesnt like it. Ofcourse it only scratches the surface since the question is vague. Your answer is also just scratches the surface and its incomplete. BTW my answer would have been a comment if I had the option. $\endgroup$ – saias Dec 12 '18 at 21:15
  • $\begingroup$ It's fine lol, it was a vague question, his answer did help. $\endgroup$ – Kartikay Dec 13 '18 at 15:31

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.