Collaborative-based filtering is basically the process of recommending new content to users based on their shared interests and preferences. For eg:- The phrase “Customer who purchased this also bought” appears when new things are recommended whenever we shop on Amazon, as demonstrated below.
By using human interaction rather than content from the products that people are using, this gets around the drawback of content-based filtering. It merely requires the users’ past performance for this. Using historical data, it is assumed that users who have consented in the past are likely to do so again in the future.
Two varieties of collaborative filtering exist:
- User-Based Collaborative Filtering: Utilising the ratings of nearby users, the item is rated. It is founded, to put it simply, on the idea of user similarity.
Here’s an illustration. You can see an image on the left with three youngsters identified as A, B, and C, and four different fruits, namely grapes, strawberries, watermelons, and oranges.
Let’s imagine that, according to the image, A bought all four fruits, B bought just strawberries, and C bought both strawberries and watermelon. Because A and C are the same types of users in this case, C will be advised to eat grapes and oranges, as indicated by the dotted line.
- Item-Based Collaborative Filtering: Utilising the user’s own ratings on nearby products, the item’s rating is anticipated. It is founded on the idea of item similarity, to put it simply.
Let’s examine using the above-mentioned example of users and objects. Here, the only distinction is that we observe comparable goods rather than comparable people. For example, if you compare grapes and watermelon, you’ll see that everyone buys the latter, while only Children A and B buy the former. So grapes are advised for Children C.
After learning both, you could be unsure about when to employ which. The following is a solution if there are more items than users: employ user-based collaborative filtering, which will consume less computing power and result in fewer things. Use item-based collaborative filtering if there are more people than items. For instance, Amazon has trillions of customers but lakhs of products to sell. Because there are fewer products than customers, Amazon also uses item-based collaborative filtering.
Advantage
- Even with little data, it still functions well.
- This model aids consumers in discovering a fresh interest in a particular item, although the model may still suggest it because other users have that interest.
- not necessary
Disadvantage
- It is unable to handle new items since the model is not trained on the database’s most recent additions. Cold Start Problem is the name given to this issue.
- Doesn’t matter all that much, a side feature. In the scope of a movie suggestion, side features in this case could be an actor’s name or the year of release.
