Yt

The idea was to make recommendation for required topics better than the YouTube recommendation. So to achieve this I started collecting different search results by using search().list() and using that data I retrieved the statistics(View Count, Like Count, Dislike Count, Favourite Count), Duration and Published At for the corresponding search results by passing the Video ID as argument to videos().list().

Approach

• The Approach for recommendation Technique was to implement a scoring function which gives scores based on the Duration of the Video, Like Count, View Count and Dislike Count of the video.
• Z-Score normalisation was used to normalize Duration, ViewCount, LikeCount, DislikeCount Columns.
• Scoring Function was created by giving weights to Duration_zscore, Views_zscore, Likes_zscore, Dislikes_zscore.
• Videos with very long duration tend to have lesser views compared to those with smaller duration which gives those videos a disadvantage so to even this out Duration was also given significant weight in the scoring function.
• Also, Dislike Count was given negative weight to make sure that only the best videos come at top during ranking of the videos based on score value.

APIs Used:

• Search: list
• Videos: list

Filters Used:

• The Videos which were longer than a minute (60 seconds) was chosen for analysis.

Challenges Faced:

• Main challenge was to get more rows of data while Search: list API has limit of 500 results.
• In some of the data the Like Count and Dislike Count was not available, so it led to removal of those data has they will affect the scoring function badly.
• In some cases, the search results contained YouTube Ads related to the topic itself, which are generally of 10-20 seconds. Those videos have enormous views and affected the analysis so these short videos were removed.

Key Learnings from Data Analysis:

• It was observed that most (almost 80%) of the videos were having less than 1 lakh views while some were having millions of views.
• Among my three topics (Data Structures, Fluid Mechanics and Data Science) I observed that Data Science was most viewed topic while Fluid Mechanics was the least viewed in YouTube.

References:

• https://matplotlib.org/tutorials/introductory/pyplot.html
• https://pandas.pydata.org/pandas-docs/stable/user_guide/merging.html