User Profile Preserving Social Network Embedding

Authors: Daokun Zhang, Jie Yin, Xingquan Zhu, Chengqi Zhang

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on four real-world social networks show that compared to state-of-the-art baselines, our method learns better social network representations and achieves substantial performance gains in node classification and clustering tasks.
Researcher Affiliation Collaboration a Centre for Artificial Intelligence, FEIT, University of Technology Sydney, Australia b Data61, CSIRO, Australia c Dept. of Computer & Electrical Engineering and Computer Science, Florida Atlantic University, USA d School of Computer Science, Fudan University, China
Pseudocode Yes Algorithm 1 The UPP-SNE Algorithm
Open Source Code No The paper does not provide an explicit statement or a link to the open-source code for the methodology described.
Open Datasets Yes We perform experiments on four real-world social networks listed as follows: Google+1 is an ego-network of a Google+ user and nodes in this network represent the user s friends. There are 1206 nodes and 66918 links in this network. We use people s gender as class label. Each node is described by a 940-dimensional bag-of-words vector constructed from treestructured user profiles [Leskovec and Mcauley, 2012]. Ego-Facebook1 is an ego-network of a Facebook user. This network consists of 755 nodes and 60050 links. People s education type is used as class label. Each node is described by a 477-dimensional vector [Leskovec and Mcauley, 2012]. Hamilton and Rochester2 are two of the collection of 100 US university Facebook networks [Traud et al., 2012]. The two networks contain 2118 nodes, 87486 edges, and 4145 nodes, 145305 edges respectively. Each node s profile is described by 7 user attributes: student/faculty status flag, gender, major, second major/minor, dorm/house, year, and high school. We select student/faculty status flag as class label and construct a 144-dimensional, and a 235-dimensional binary feature vector for Hamilton and Rochester, respectively.
Dataset Splits Yes To make fair comparisons, we vary the training ratio from 1% to 10% by an increment of 1%. For each training ratio, we randomly split the training set and test set for 10 times and report the averaged accuracy.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper does not provide specific version numbers for any software dependencies.
Experiment Setup Yes For UPP-SNE, we set the number of random walks per node γ as 40, the walk length l as 40, the window size t as 10, and the number of iterations as 40. All baselines use default parameters as reported. The dimension of learned node representations m is set as 128 for all the algorithms.