Project 5: a recommendation system for Netflix shows

Finally, I reached the final project for this self-paced data scientist program. As a final capstone project, I chose to develop a recommendation system on Netflix shows, which I have not conducted before.

In this project, I downloaded Netflix Moives and TV Shows dataset from: https://www.kaggle.com/datasets/shivamb/netflix-shows and developed a recommendation system with following two steps:

Comprehensive EDA was performed on different features of the data
Content-based recommendation system was developed and tested

EDA

The data I loaded from the website is a csv file: netflix_titles.csv

It has a total of 8807 entries and 12 columns.
There are some null values in 6 columns: 1) director:account for 1/3, this one is important 2) since people may like the movie/TV show from the same director. Therefore, I replace it with nodata 3) cast: same as director, replace it with nodata 4) country: this one is also important, instead of replacing with nodata, I will replace it with the most common country 5) data_added, rating, duration: just drop these entries

Go through some selected columns

The percentage of each type:

The directors who have the top 10 titles in the database:

The casts who have the top 10 titles:

figure of top10cast

The countries who have the top 10 titles:

figure of top10countries USA,India,and UK have the top 3 number of movies/TV show

Precentages of movies/tv shows with respect to adding years and release years

figure of noc20152021 In general, there are more movies than TV shows in each year

The rating

figure of rating TV-MA, TV-14, TV-PG are the three ratings with top three counts

The list_in

figure of list_in International,commedy,drama are the top three

The description

figure of description

Recommendation

I tried six combinations of features based on ‘director’,’cast’,’rating’,’listed_in’,’description’:

‘comb_features0’: (‘director’, ‘cast’, ‘rating’, ‘listed_in’),
‘comb_features1’: (‘director’, ‘cast’, ‘rating’, ‘description’),
‘comb_features2’: (‘director’, ‘cast’, ‘listed_in’, ‘description’),
‘comb_features3’: (‘director’, ‘rating’, ‘listed_in’, ‘description’),
‘comb_features4’: (‘cast’, ‘rating’, ‘listed_in’, ‘description’),
‘comb_features5’: (‘director’, ‘cast’, ‘rating’, ‘listed_in’, ‘description’)

and found they are generally similar, see an example: figure of difcombfeatures So I just used comb_feataures5, and two examples of TV show and Movie are shown below: TV show:’love on the spectrum’ figure of top10TVshow Movie: jeans

Conclusion

After a comprehensive EDA of netflix movie and TV shows, a content-based recommendation system was successfully built