Modeling Citation Trajectories of Scientific Papers

• This repository contains the code and data required to reproduce the results reported in the paper.
• The following sections contain instructions on running the models.
• An additional data file node_edge_dict is required for models run on the APS dataset. Here is the link to it, since it exceeded the 100MB file size limit set by git.

Evaluation Model

• javac CitationPattern.java
• java CitationPattern.class ACTIVATION_PERIOD PEAK_THRESHOLD to classify the nodes, according to the specified parameters.

Ground-truth network

• python generate_gt_net.py

Barabasi-Albert, Additive-fitness, Multiplicative-fitness models

• python main_<ba,add,mult>.py --out_suffix OUT_SUFFIX --track_till GROW_TILL --num_threads 0
• The argument GROW_TILL should have the year till which we want to grow the network. This means, only the nodes published till the year GROW_TILL - 10 will be classified by the evaluation model.

LBM

• python main_lbm.py --out_suffix OUT_SUFFIX --track_till GROW_TILL --num_threads 0 --gamma GAMMA
• The argument GROW_TILL should have the year till which we want to grow the network. This means, only the nodes published till the year GROW_TILL - 10 will be classified by the evaluation model.
• The argument GAMMA represents the type of function mapped from the number of nodes in the network. The options are constant, linear, sqrt and log.

LBM

• python main_lbm.py --out_suffix OUT_SUFFIX --track_till GROW_TILL --num_threads 0 --skip SKIP --stdev STDEV
• The argument GROW_TILL should have the year till which we want to grow the network. This means, only the nodes published till the year GROW_TILL - 10 will be classified by the evaluation model.
• The argument SKIP represents the frequency at which the active subspace is shifted. A negative value means the frequency is in terms of months, while a positive value means the frequency is in terms of number of new nodes entering the network.
• The argument STDEV represents the standard deviation of the Gaussian distribution over each active subspace, from which location vectors are sampled and assigned to nodes.