Context

This project has been realized for the Network Science Analytics class (MSc AI 2019-2020 at CentraleSupelec) by :

• Gaël de Léséleuc
• Alexandre Duval

The aim of our project was to conceive a pipeline that would extract a dynamic narrative graph from a given novels and then used this graph to study the narrative story of the novel. A detailed report can be found in the Report folder.

Installation

To install and use our end-to-end processing pipeline : from novel to narrative graph, use the following command in a new environment.

git clone https://github.com/dldk-gael/novel_story_building.git
cd novel_story_building 
pip install -r requirements.txt

Our pipeline use BERT-NER in order to recognize the character names in the text. As a result, if you want to use it, you must download the weight of pre-trained BERT-NER model and save it in the models folder.

Note that we have upload some cache data in the data folder so that you can test our pipeline without having to preprocess raw data from scratch.

Pipeline structure

The folder has the following organization :

• models : to store the weight of BERT-NER model
• data: to store the data
  • raw_text : put here the novel.txt you want to analyse
  • book_by_chapter : will contain the novel split by chapter
  • coref_rules : contains database that are use by coreference rules module
  • entity_list : will contain the output of text processing as pickle file (ie: the occurence list)
  • graph : will contain the narrative graph
• report: contains the project assignment description, our project proposal, our project report
• src is divide in three modules :
  • third_party : contains source code of other frameworks we used such as BERT-NER and chapterize
  • text_processing : the module responsible of text processing
  • graph : the module responsible of graph creation and graph analysis

Utilisation

To process a book and obtain different narrative graph as .gexf files, store the file book_name.txt in data/raw_test/ and launch the command :

python main.py --book book_name 

By default, if you launch :

python main.py

it will re-run some graph analysis on HarryPotter book using cache data we left in the data folder.