Description
In this project, we will create and explore a network from the IMDb
movie data. The data is available at IMDb Interface. We are going to
study the properties of actor/actress network and movies. You can also
also find a cleaned up version of the data at this Box link.
1. Download actors.list.gz, actresses.list.gz (or use the
actor_movies.txt and actress_movies.txt files from the
cleaned up data), merge those 2 lists into one file, and remove
all actors/actresses with less than 5 (so actors who have acted in
four or fewer number of movies) movies; Note that you will have
to parse the data in these lists as accurately as possible to extract the entities consistently and create the network. So plan on
spending some time in cleaning the data set.
2. Construct a weighted directed graph G(V, E) from the list, while
V = {all actors/actressess in list}
Si = {m|i ∈ V, m is a movie in which i has acted}
E = {(i, j)|i, j ∈ V, Si ∩ Sj 6= φ}
and for each directed Edge i → j, a weight is assigned as |Si∩Sj |
|Si|
.
3. Run pagerank algorithm on the actor/actress network, look into
those who are among top 10, do you know their names? List the
top 10 famous movie celebrities in your opinion, what are their
pagerank scores? Do you see any significant pairings among actors? Any major surprises, in the sense that well-known actors do
not show up in the high pagerank list?
1
4. Similarly, remove all movies with less than 5 actors/actresses on
list, construct a movie network according to the set of actors/actresses,
with weight assigned as the jaccard index of the actor sets of 2
movies. Now we have an undirected network instead.
5. Do a community finding on the movie network; use the Fast Greedy
Newman algorithm. Tag each community with the genres that
appear in 20% or more of the movies in the community. Are these
tags meaningful?
6. Add the following nodes into the network, For each of them, return the top 5 nearest neighbors. Which communities does each
of them belong to?
Batman v Superman: Dawn of Justice (2016)
Mission: Impossible – Rogue Nation (2015)
Minions (2015)
7. Download the ratings list, derive a function to predict the ratings
of the above 3 movies using the movie network. (hint: try to use
the ratings of neighbor movies and movies in the same community.)
8. Using a set of features that include the following:
• top 5 pageranks of the actors (five floating point values) in
each movie.
• if the director is one of the top 100 directors or not (101
boolean values). These are directors of the top 100 movies
from the “IMDb top 250”. You can also find a list of these
movies in the ratings.list.gz file.
train a regression model and predict the ratings of the 3 movies
mentioned above. Specify the exact feature set you use and how
you compute the numerical values for these features. Compute
and state the goodness of fit for your regression model.
9. (Bonus) Try predicting the ratings of these movies with a different approach. Construct a bipartite graph, with actors/actresses
representing the vertices of one part and movies representing the
2
vertices of the other part. An actor/actress is connected to all the
movies he has played in. Assign a score to each actor based on the
ratings of the movies he has played in (it’s on you to define an appropriate metric here; natural choices are the average among all
ratings, the highest, or average among the top ones). Then predict the ratings of the new movies based on the scores you have
assigned to the actors.
3