I am creating a recommendation engine to recommend events (similar to meetup.com). Each event has 1 Interest and 0..N Tags (the tags are similar to SO tags). Should I enter these as a single text entry that is embedded? Or have a bool[] with it set to true for the matching entries?

The bool[] gives the AI services a cleaner set of data. But this means for an event it will have 55 features, 40 of which are the interests & tags. Will that dominate all NN comparisons - that would be bad.

Update: To explain more without violating what's allowed in a question. My approach for a recommendation engine is to create a vector of every user and of every event. The event vectors will include the Interest & Tags for the event, along with some other bools, numbers, and some text that is converted into embeddings.

Them to get recommendations for a user, I do two things: First, I do a nearest neighbor lookup of the users closest to their user vector. And then get the events that user has signed up for.

Second, if the user has signed up for events, I do a nearest neighbor of the events they signed up for and get those events.

I then take the 30 best matches across both lists of events the user has not signed up for yet, and return that.

  • $\begingroup$ Hello. Welcome! You're not explaining which algorithm or model you're trying to use. It seems you're using some AI service. It might be a good idea to mention which one and how you're calling this service, etc., but bear in mind that asking about how to use specific software is off-topic. More details here: ai.stackexchange.com/help/on-topic $\endgroup$
    – nbro
    Commented May 21 at 14:16
  • $\begingroup$ @nbro Added more detail, without specifying the software used. If you need more, please ask - it's difficult figuring out how to ask without listing the services used. $\endgroup$ Commented May 21 at 15:26


You must log in to answer this question.

Browse other questions tagged .