24. Text: More Recommendation Technniques

Three Types of Recommendations

There are three methods that you have now implemented for making recommendations. These are the three most recognized methods in industry:

1. Knowledge Based Recommendations

Knowledge based recommendations frequently are implemented using filters, and are extremely common amongst luxury based goods. Filters that you might see when purchasing items like cars or homes are examples of knowledge based recommendations. In knowledge based recommendations, users provide information about the types of recommendations they would like back.

2. Collaborative Filtering Based Recommendations

Collaborative filtering uses the connections between users and items to make recommendations. Even the content based recommendation you just implemented used some collaborative filtering techniques, as you were not treating items and users independent from one another. In this lesson, you used neighborhood based collaborative filtering to find users who were alike and then recommend new movies based on these similar users.

Even in the content based recommendation, you were using collaborative filtering. You were finding items that were similar and making recommendations of new items based on the highest ratings of a user. Because you were still using the user ratings of an item, this was an example of a blend between content and collaborative filtering based techniques.

3. Content Based Recommendations

In the previous notebook, you created a matrix of similarities between items (movies) based only on the content related to those movies (year and genre). The similarity matrix that was used, was completely created using only the items (movies). There was no information used about the users implemented. For any movie, you would be able to determine the most related additional movies based only on the genre and the year of the movie. This is the premise of how a completely content based recommendation would be made.

Often blended techniques of all three types are used in practice to provide the the best recommendation for a particular circumstance.

There are still more advanced techniques that are related to the methods that you learned about here, and they will most likely fall in one of the three buckets below.