Takeaways

Text cleaning was really challenging due to conversational nature of the text and dissimilarity between structure of episode transcripts 

Increase in episodes featured in Artificial Intelligence category

Negative sentiments per episode are higher in Artificial intelligence, Machine learning and Deep learning categories as compared to others

Each dialogue in episode needs to be individually evaluated for sentiments associated with a technology as sentiments can be influenced by speaker background

Unique keywords extracted fairly represent the categories episodes are classified in.  Overlap between episode categories complicates this process

Topic modeling nicely represented text into different topics and how they have evolved over time