SICS-cite

This is a thesis project done at SICS. The overall goal of the project is to implement and evaluate a set of algorithms for citation analysis.

Dependencies

The following python modules are used in this project:

• NetworkX
• graph-tool
• pandas (only used for Logistic regression and correlation coefficients)
• Statsmodels (only used for Logistic regression)

Directories

• algorithms - Tools and algorithms used. Note: Additional library algorithms are used in the metrics directory.
• datasets - Contains rawdata, data, and parse files that create GraphML files from the rawdata. The data files are not available through GitHub due to space limitations.
• metrics - Parsing, calculation, and evaluation of metrics.
• arxivdownload - an arXiv crawler and parser, slow, not tested much
• boost - Some very basic tests using the C++ boost library

The algorithms directory is dependent on the python graph package NetworkX. The metrics directory uses the python graph package graph-tool for all graph processing and pandas and Statsmodels for statistical analysis.

Algorithms

Below is a list of algorithms used, along with source/package information:

• The Backbone algorithm as described in 'Tracing the Evolution of Physics on the Backbone of Citation Networks' by S. Gualdi, C. H. Yeung, Y.-C. Zhang - Found in algorithms/backbone.py, implemented with NetworkX.
• Co-citation graph generation - Found in algorithms/co_citation.py (graph-tool) and in algorithms/graphutils.py (#build_co_citation_graph (NetworkX)).
• Indegree and betweenness centralities - Done using graph-tool (indegree is a property for all graph-tool graphs)
• PageRank and HITS - Done using graph-tool in metrics/ and using NetworkX in the test file algorithms/pr_avg_age.py
• Burstiness - Done separately using the Sci2 software. Saved as csv files.

Workflow

The approximate workflow is as follows:

1. Collect raw citation data
2. Build GraphML files using the parse scripts available in each dataset's directory. (Currently available for AAN and APS).
3. Build co-citation graphs using algorithms/co_citation.py and backbone graphs using algorithms/backbone.py
4. Generate ranked lists for each metric in the metrics directory.
5. Evaluate the ranked lists using metrics/find_fellows.py.

Notes

The algorithms directory have been tested on both FreeBSD 9.3 and Windows 8. The metrics directory have been tested on Arch Linux 4.0.2-1. The main problem is to get graph-tool up and running, which requires boost. All OS's mentioned are x86-64 versions.

The test files related to fellows are not very modular, with parsing, checking and plotting done in one file.