Figure 1.
Excerpt from the film citation network extracted from the Internet Movie Database (IMDb).
The network contains six different types of citations: The reference is the most common and generic type of citation. In the broadest sense, it can refer to visual or narrative elements that are incorporated into a newer film. For example, the main character in ‘Mulholland Drive’ (2001) consciously assumes the name Rita after seeing the poster of the film ‘Gilda’ (1946), starring Rita Hayworth. A film is said to feature another work if it shows a characteristic sequence from it. For instance, ‘Black Cat, White Cat’ (1998) features ‘Casablanca’ (1942) when its main character watches the ending scene of the latter on TV. Films in a series follow each other. The ‘Godfather II’ (1974) is an example of both sequel of (it picks up the story of the original film) and prequel to (narrates the antecedents) the ‘Godfather’ (1972). A remake is a newer version of an old film, such as ‘Nosferatu the Vampyre’ (1979), which is a stylistic retake on the silent film ‘Nosferatu’ (1922). The spoof relation indicates the ironic imitation of a film, such as the leading character of ‘Gilda’ (1946) that served as an inspiration for the animated and exaggerated character Jessica in ‘Who Framed Roger Rabbit’ (1988). Finally, archive footage from the burlesque ‘Sherlock Jr.’ (1924) is edited into the documentary ‘When Comedy Was King’ (1960), thus exemplifying the most straightforward form of film citation: the direct inclusion of long sequences of the original film.
Table 1.
Statistics for the subnetworks corresponding to the individual citation types and the aggregated network containing all six types.
Figure 2.
The inverse of the cumulative distribution function for the in- and out-degrees.
Plots are shown for the subnetworks constructed from the different citation types and the aggregated network. The P-values that result from a Kolmogorov-Smirnov test confirm for all but the follows subnetwork the significant difference between the in- and out-degree distributions.
Table 2.
Assortativities for the subnetworks corresponding to the individual citation types and the aggregated network containing all six types.
Figure 3.
Parallel edges and shared nodes between the subnetworks constructed from the different citation types.
Both the edge and node overlap are defined as the percentage of edges/nodes from the network indicated on the y-axis that are also present in the network specified on the x-axis.
Table 3.
Centrality indices that take into account the time-ordering of the nodes.
Figure 4.
The impact of the highest-ranked films we extracted from the film citation network as expressed by the number of awards and nominations received within two years of their appearance, as well as their average rating on IMDb.
The films are sorted from top to bottom by decreasing ranking according to the film citation network. Ratings are based on a 10 star system with 10 indicating the highest possible rating. Decimal values are possible due to the averaging of individual user ratings by IMDb.
Figure 5.
Transforming a film top list into a network based on shared film attributes.
(a) Subgraph of the film–attribute bipartite graph containing film–director, film–writer, and film–composer relations. (b) Exemplary subgraphs of the corresponding multiplex film network.
Figure 6.
Comparison of various top lists: the personal favourite lists of four randomly chosen IMDb users; the IMDb Top 100, which is based on the votes of its regular users; two lists compiled by the film experts Roger Ebert and Tim Dirks; the lists of the American and British Film Institutes; and our list deduced from the film citation network.
(left and center) Overlap between the edges in the multiplex film networks, indicating the ratio of shared attributes. (right) Rankings of all films in the given top lists as compared to the ranking based on the film citation network. For each such list, the ranks of all it's films according to the film citation network ranking are summed and divided by the sum of optimal ranks (i.e. for a list of
films). The average result for a randomly selected list of 100 films from the data set is
(not shown).