Get startedGet started for free

Item-based or user-based

1. Item-based or user-based

Throughout this chapter, we have worked with item-based recommendations and user-based recommendations.

2. Item-based filtering

As a reminder, item-based recommendations are where you use the average of the k most similar items that a user has rated to suggest a rating for an item they haven't yet seen. If we want to see what this user would rate the purple book, we take the 3 most similar books they have read and reviewed, and average their scores.

3. User-based filtering

Alternatively, user-based recommendations are when you use the average of the ratings the k most similar users gave an item, to suggest what rating the target user would give it. So in the same example as before we would find the 3 most similar users that have reviewed the purple book, and take that average as our user's score. Both provide useful results, but you, of course, will want to know when to use item-based, and when to use user-based recommendations. Often it will very much depend on the data, but there are some standard pros and cons of both which we will discuss.

4. Why use item-based filtering?

First of all, item-based recommendations are more consistent over time. Users' preferences change, for example, you might enjoy animated movies when you are younger, but change your preferences to action movies later in life. Items on the other hand do not usually change, a movie that was a horror movie when it came out is still a horror movie years later. Item-based recommendations can be easier to explain Telling a user that they were recommended a book because they liked a similar one (item-based collaborating) can make more sense than persuading them they might like a book because a user they have never met liked it (user-based collaborating). Item-based recommendations can have more precalculations. Any online store generally has a finite known inventory. Its owner can calculate what items in the inventory are similar to each other and which ones are not offline and use it on their site. New users, on the other hand, appear every day, so cannot benefit from as much precalculation. One negative is that item-based recommendations are often very obvious, for example just suggesting the next movie in a series which might not be much of a value add.

5. Why use user-based filtering?

In fact although item-based recommendations appear to be preferable to users based on most accounts, one area that user-based can win out is that user-based recommendations can be a lot more interesting, and unexpected than item-based. It can be particularly useful at finding less popular items that the user would like. For this reason, while item-based recommendations are preferable in use-cases that conservative suggestions are encouraged, such as an e-commerce store, user-based recommendations can add value for more subjective items such as movies, books, or other entertainment.

6. Let's practice!

Now you can not only generate item-based and user-based collaborative recommendations you should also be able to recognize the value of them both. Let's see if you can compare them and understand their benefits.

Create Your Free Account

or

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.