Item Score

[sayanroyc]

I think discussing on here is better instead over texts, easier to keep records and go back to.

So, item score. This is open-ended but really important to our success because it allows users to (hopefully) find items of similar interest. Let's get a list of ideas: 1) Weight vector, where each element corresponds to the how often the user rents from a certain category. For examples, if the categories are ["Mens", "Electronics", "Books"], and a user tends to be renting a lot of books, few electronics, and no mens items, his weight vector would look something like: [0.01, 2, 25]. We can experiments with either addition or multiplication for promotion after each rent event.

• weight vector will be a property of the user class
• if a user is not logged in, set all category scores to 1.0 (aka just return x closest items)
• need to find a way to factor in location (should we, at all? How about return 100 closest items from the database, then calculate scores, and finally return a list of those 100 items ordered by score? )
• what happens when we add/delete a category? We will have to update weight vectors for EVERY user in the database, which is not good. On the other hand, how often will we be making new/deleting categories?

[nickgarfield]

Yeah, good ideas. I think we should probably expect that we will have to add new categories at some point in the future. Smart phones, tablets, and wearables could all be categories that couldn't have existed 8 years ago.

Theres definitely a lot of ways to go with score. I think what you defined is a really good. Basically, if a user has rented items in the Game Consoles section, show them more items in the Game Consoles section.

On top of this, we could look for purchasing patterns among all users and even get into demographics. If users that tend to rent from the Game Consoles section also tend to correlate with the users that rent from the Video Games, then we can recommended video games to someone who is looking at game consoles.

[sayanroyc]

DONEZO (for a while). score function returns a distance score, price score, and user_category_weight score. We can manually set the weighting of each for score and aggregate into one score.