In this Projekt I used the YouTube API to collect data and did some EDA.
Founded in 2005, Youtube has grown to become the second largest search engine in the world (behind Google) that processes more than 3 billion searches per month. [1]. It is, however, generally a myth how the Youtube algorithm works, what makes a video get views and be recommended over another. In fact, YouTube has one of the largest scale and most sophisticated industrial recommendation systems in existence [2]. For new content creators, it is a challenge to understand why a video gets video and others do not. There are many "myths" around the success of a Youtube video [3], for example if the video has more likes or comments, or if the video is of a certain duration. It is also worth experimenting and looking for "trends" in the topics that Youtube channels are covering in a certain niche.
Having recently stepping into the content creation world with a new Youtube channel on data analytics and data science, I decided to gain some insights on this topic which might be useful for other new content creators. The scope of this small project is limited to data science channels and I will not consider other niches (that might have a different characteristics and audience base). Therefore, in this project will explore the statistics of around 10 most successful data science Youtube channel.
Within this project, I would like to explore the following:
- Getting to know Youtube API and how to obtain video data.
- Analyzing video data and verify different common "myths" about what makes a video do well on Youtube, for example:
- Does the number of likes and comments matter for a video to get more views?
- Does the video duration matter for views and interaction (likes/ comments)?
- Does title length matter for views?
- How many tags do good performing videos have? What are the common tags among these videos?
- Across all the creators I take into consideration, how often do they upload new videos? On which days in the week?
- Explore the trending topics using NLP techniques
- Which popular topics are being covered in the videos (e.g. using wordcloud for video titles)?
- Which questions are being asked in the comment sections in the videos
- Obtain video meta data via Youtube API for the top 10-15 channels in the data science niche (this includes several small steps: create a developer key, request data and transform the responses into a usable data format)
- Prepocess data and engineer additional features for analysis
- Exploratory data analysis
- Conclusions
