We employ a nationwide phone call dataset from Jan. 2015 to Dec. 2016. The log interaction duration and log interaction frequency in each phase (intermediate results) are both provided. Currently, we upload the Results folder to Google Drive. (https://drive.google.com/drive/folders/1h4rHZvzzQO7niYMelbzToJZernOij1dv?usp=sharing)
Please download the files from google drive for replication purposes.
In each file, we list tie ranges and interactions in all phases. For example, in 'Results/Graph_season_TR_Duration.txt', the former eight columns are tie range and the latter eight columns are log interaction duration. Tie range is calculated by the length of the second shortest path of two nodes. '-1' means that one node of this connection has no interaction with others in this phase. '100' means that there is no second path between two nodes, indicating that the tie range is infinite. '101' means that the degree of one node is 1, indicating that the tie range is infinite.
Differential privacy is applied to protect the privacy of users. Concretely, we add a Gaussian noise with μ=0, σ=5 to log interactions. When reproducing the results, please remove all numpy.log in the codes, and minus a σ for the calculation of error bars.
Paper | Code | Results | Date | Stars |
---|