goweiting / project-infnet

UG4 Project

Informatics Collaboration Network & Topic Network

Focusing on the School of Informatics, University of Edinburgh, a collaboration network was created using informtion from the University's collection of research publications Edinburgh Research Explorer. More details in infnet-scrapper.

Using the publications scrapped from the research explorer, topic models were inferred, and a topic-similarity networks[1] were generated. A collaboration network was also created, visualised and analysed.

---

Directory

1. Data

   • bin
     • scrapy : scripts for scraping using scrapy
     • pdfminer: contains binary from pdfminer.six
     • scripts used to process PDFs using pdfminer
   • data_dblp : dblp dataset, but metadata of publications are not stored due to the size of the dataset. We only store tokenised pickled files and dictionary in it.
   • data_schoolofinf : Informatics dataset retrieved in Jan 2018
   • notebooks : corresponds to steps taken to process and generate lookup tables for the remaining steps.

2. infnet-analysis

   • notebooks : contain the jupyter notebook used to generate each informatics network.
     • community detection and homophily test is carried out in analysis.ipynb

3. embedding

   • notebooks : creation of topic-similarity networks

4. topicModel

   • notebooks : generate topic models using Gensim's implementation of LDA; also explore the performance of each model
   • src : contain scripts to generate each topic model

Setting up

The project is still in development. To use the datasets and run the notebooks on your system, follow the following instruction:

1. The project is developed in python3.6. Using anaconda to setup the virtual environment will be the easiest. You can get a copy of miniconda by issuing the following command:

$ curl -O https://repo.continuum.io/miniconda/Miniconda3-latest-MacOSX-x86_64.sh # For MacOSX
$ curl -O https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86.sh # For linux/ubuntu
$ bash Miniconda3-latest-MacOSX-x86_64.sh # Install miniconda onto your system
$ echo "export PATH=\""\$PATH":$HOME/miniconda3/bin\"" >> ~/.benv
$ source ~/.benv

*** NOTE the use of python3 instead ***

# Create conda environment (name infnet3) for project:
# Also install essential packages across all modules:
$ conda create -n infnet3 python=3 pandas matplotlib jupyter ipython ipykernel
$ source activate infnet3 # Activates the environment
(infnet3) $ <--- this shows the successfull acitvation of the environment.

Now, we have to install required python packages. This list is updated as the project progresses:

1. For data pre-processing, additional packages are installed:

(infnet3) $ conda install scrapy # for scrapping the research explorer
(infnet3) $ conda install nltk # this is used for creating tokens for topic modelling

1a. To configure NLTK, executing the following in a new terminal with infnet3 activated :

(infnet3) $ python to launch a python3 shell
> import nltk
> nltk.download('stopwords') # select `yes` when prompt.
> nltk.download('WordNet')

2. For infnet-analysis:

(infnet3) $ conda install networkx numpy
(infnet3) $ pip install python-louvain # community detection package

3. For topic modelling:

For topic modelling using latent diriclet allocation

$ conda install gensim # to generate LDA
$ pip install pyldavis # for visualisation of the LDA

For data exploration, visualisation of data and clustering:

$ conda install scikit-learn # for k-means, manifold, dbscan...
$ conda install -c conda-forge hdbscan

---