I am building a recommendation system that recommends relevant articles to the user. I am doing this using simple similarity-based techniques (with the Jaccard similarity) using as features the page title, the tags, and the article content.

Now my problem is I have different "adult articles" and some are articles that expire (for example, an article about a movie in Jan 2019 would not be relevant in Dec 2019).

I want to keep these adult articles separate, as a person who is reading about history does not want to be led to an adult article and not recommend articles that have expired or would not be relevant in the present moment.

Should I just improve the quality of my features or tags? Or is there any other way to achieve this?

  • $\begingroup$ wouldn't it better to have two systems, one for adult and one else if you don't want them to connect? Considering the fact that you will add items to your system you could also remove th ones that are more ephemeral or find a way to use date as a distance metric between two items. Though I'm not familiar with recommendation systems. $\endgroup$ – Al rl Sep 11 '20 at 13:43
  • $\begingroup$ Is there some reason you could not hardcode this?? $\endgroup$ – hisairnessag3 Sep 11 '20 at 23:48
  • $\begingroup$ @hisairnessag3 it would not be scalable right? $\endgroup$ – m2rik Sep 13 '20 at 21:54
  • $\begingroup$ @Alrl I thought of doing that in the pre-processing step but the problem is just identifying the tags with sex etc in it does not help me get an exact match for an adult article. I also get different articles which talk about sex but are not adult articles. $\endgroup$ – m2rik Sep 13 '20 at 21:56
  • $\begingroup$ @m2rik i fail to see why it wouldnt be $\endgroup$ – hisairnessag3 Sep 15 '20 at 4:41

Your Answer

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

Browse other questions tagged or ask your own question.