SeriesNetDynMatch

Dynamic community matching for TV series conversational networks

• Copyright 2016-17 Vincent Labatut

SeriesNetDynMatch is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation. For source availability and license information see licence.txt

• Lab site: http://lia.univ-avignon.fr/
• GitHub repo: https://github.com/CompNet/SeriesNetDynMatch
• Contact: Vincent Labatut vincent.labatut@univ-avignon.fr

---

Description

This set of R scripts was designed to perform community detection on dynamic networks corresponding to conversational interactions between characters of TV series. It also can generate alluvial diagrams representing the evolution of these interactions, as well as plots of the evolution of the character strength and the link weights.

These scripts were used in article [BLGL'17], which deals with TV series Breaking Bad, Game of Thrones, and House of Cards.

If you use this software, please cite reference [BLGL'17]:

@InCollection{Bost2016b,
  author    = {Bost, Xavier and Labatut, Vincent and Gueye, Serigne and Linarès, Georges},
  title     = {Extraction and Analysis of Dynamic Conversational Networks From {TV} Series},
  booktitle = {Social Network Analysis and Mining},
  publisher = {Springer},
  year      = {2018},
  series    = {Lecture Notes in Social Networks},
  chapter   = {3},
  pages     = {55-84},
  doi       = {10.1007/978-3-319-78196-9_3},
}

Data

The scripts have been written to be applied on the data previously extracted from the listed TV series. These data take the form of Graphml networks, which are available on Zenodo.

Download the data and unzip them in the data folder of this project, in order to match the existing folder structure.

Organization

Here are the folders composing the project:

• Folder src: contains the source code (R scripts).
• Folder data: contains the files used by our scripts, i.e. the inputs, as well as the folder created by our scripts, i.e. the outputs.
  • Folder xxx_dyn_ns: input files for series xxx, corresponding to graphs obtained through narrative smoothing.
  • Folder xxx_dyn_ns_alluv: alluvial diagrams generated for series xxx (only for the graphs obtained through narrative smoothing).
  • Folder xxx_dyn_ns_clstr: community detected for series xxx (only for narrative smoothing).
  • Folder xxx_dyn_ns_match: community matches obtained for series xxx (only for narrative smoothing).
  • Folder xxx_dyn_ts10: input files for series xxx, corresponding to graphs obtained through temporal integration using 10-scene windows.
  • Folder xxx_dyn_ts40: input files for series xxx, corresponding to graphs obtained through temporal integration using 40-scene windows.
  • Folder xxx_dyn_strength: strength and weight plots generated for all three types of graphs (narrative smoothing, 10- and 40-scene windows).

Installation

1. Install the R language
2. Install the following R packages:
   • igraph (tested with version 1.0.1).
   • alluvial (tested with version 0.1-2)
3. Download this project from GitHub and unzip the archive.
4. Download the data from FigShare (see Section Data) and unzip in the data folder, so as to match the existing folder structure.

Use

In order to replicate the experiments from the article, perform the following operations:

1. Open the R console.
2. Set the project root directory as the working directory, using setwd("<my directory>").
3. Using the command source("<script.R>"), run:
   • The script src/custom-match.R to perform community detection and matching on the narrative smoothing graphs;
   • Or the script src/strength-evol.R to generate comparison plots from the three types of graphs.

The generated files will be placed in the data folder, consistently with the description given in Section Organization.

Dependencies

• igraph package: used to build and handle graphs.
• alluvial package: used to generate the alluvial diagrams.

Note: the src/_archive.zip file contains various programs and scripts which were tried in order to perform community detection and matching, but were not finally kept. There are provided here just for information.

References

• [BLGL'17] X. Bost, V. Labatut, S. Gueye and G. Linarès. Extraction and Analysis of Dynamic Conversational Networks from TV Series, in: Social Network Analysis and Mining, Lecture Notes in Social Networks, 3:55-84, Springer, 2017. DOI: 10.1007/978-3-319-78196-9_3 ⟨hal-01543938⟩