Date: 3 July 2014
Author: Alphan Kırayoğlu
Data: Keegan's analysis
Data: Willem Robert van Hage's analysis

Network of Turkish Cinema

In April, Walt Hickey shared an analysis of paintings of Bob Ross on FiveThirtyEight’s website. Brian Keegan approached the data from network analysis perspective.

Inspired by Brian Keegan’s work, I wanted to look at Turkish cinema through the lens of network analysis. After doing some research online, I came across a post by Willem Robert van Hage on visualizing a network of actors where directors or movies are viewed as the second layer of nodes in a bipartite network where movies constitute the second type of nodes. (Similar to a network of ingredients and recipes or employees and companies) I chose to include actors and directors together in the network as the first layer and use movie titles as the second layer.

I obtained the data from I would like to thank Ahmet S. Aktas and his colleague’s at for maintaining the website, gathering the data and sharing it with me. I converted the data into an incidence matrix, which is essentially a binary rectangular matrix where rows represent movies and columns represent actors, actresses and directors. i,jth element of the matrix is set to 1 if an actress (jth column) was cast for the movie represented by the ith row of the matrix. I have omitted the R code and the Gephi instructions since they were provided by Willem Robert van Hage.

When we look at the network of actors and movies, the biggest clusters are due to more recent movies with bigger cast and more detailed cast information on such as “Kurtlar Vadisi – Irak” and “Kanal-i-zasyon”.

When we look at the network of directors and actors, we see that the casts of directors of recent movies create distinct clusters. I should say directors of movies with bigger casts.

These results are due to the fact that majority of the actors in the data set has appeared only in a single movie (cast by a single director). I repeated the network analysis after removing actors with a single appearance. Then we don’t see distinct clusters, but we still get few nodes with more edges. These nodes are once again recent movies and directors such as Cagan Irmak, Murat Saracoglu and Serdar Akar. However, we do not see clusters that would suggest directors casting the same actors for different projects. Though this could be due to the limited number of movies by Turkish directors in our data set. We have 343 distinct directors for 543 titles. So it is hard to spot Woody Allen or Coppola like directors which would require a great number of titles directed by the same directors.

As mentioned above, most the results are due to the data set being skewed towards more recent movies and being limited. Kemal Sunal does not even appear in the data set. Yol is not mentioned. Hence, I would need to work on gathering a more complete data set.