Decoding the Popularity of TV Series: A Network Analysis Perspective: Discussion

Table of Links

• Abstract and Introduction
• Dataset
• Methods
• Results
• Discussion
• Conclusion, References, and Appendix

DISCUSSION

The results of our research suggest that different character network structures may have different impacts on the quality of TV episodes. For example, the negative correlation between the number of active nodes and episode review scores in Game of Thrones may indicate that having too many characters in an episode can be overwhelming for viewers. On the other hand, for Breaking Bad, the positive correlation between the standard deviation of eigenvector centrality and episode review scores suggests that focusing on a smaller number of main characters is preferred over a wider focus on many different characters in a single episode.

House of Cards had four correlations. The negative correlation between network efficiency and episode review implies that nodes are not in tight-knit groups - for example, character A might talk to character B who might talk to character C, but they might not talk in a group together. It also found a negative correlation between maximum harmonic centrality and episode review, implying again that characters have isolated conversations with one another rather than engaging in group discussions.

Breaking Bad had a single positive correlation between network transitivity and episode review. This implies that episodes with tightly connected groups of characters tend to be preferred by viewers.

It is important to note that a limitation of our study is that we only have the reviews for the episode as a whole, and not just for the character networks/dynamics. This means that many other factors such as cinematography, script writing, and plot, are all factored into a single review. Additionally, the placement and release of the episode are factors. For example, the finale of a series might be highly anticipated and receive a higher rating, or a guest star might appear in a particular episode and contribute to its score. Because the overall score of an episode represents all of these details together, we do not have the most accurate data to find a correlation between the character network and the metrics themselves.

Our study is also limited to the types of metrics we tested. Though we tested many types, there are still many more network metrics that could potentially find greater correlations with the data.