Scalable Multiplex Network Embedding

Authors: Hongming Zhang, Liwei Qiu, Lingling Yi, Yangqiu Song

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct experiments on two tasks: link prediction and node classification using six different multiplex networks. On both tasks, our model achieved better or comparable performance compared to current state-of-the-art models with less memory use.
Researcher Affiliation Collaboration Hongming Zhang1, Liwei Qiu2, Lingling Yi2, Yangqiu Song1 1Department of CSE, The Hong Kong University of Science and Technology 2Tencent Technology (SZ) Co., Ltd., China hzhangal@cse.ust.hk, deolcaqiu@tencent.com, chrisyi@tencent.com, yqsong@cse.ust.hk
Pseudocode Yes Algorithm 1 Multiplex Network Embedding
Open Source Code Yes The source code of our model is available at: https://github.com/HKUST-KnowComp/MNE.
Open Datasets Yes We use open multiplex networks from Manlio De Domenico project1 as our experimental datasets.1http://deim.urv.cat/~manlio.dedomenico/data.php and Twitter: The Twitter data used in our experiments is the higgs-twitter dataset2. This dataset was used to analyze the spreading of news [De Domenico et al., 2013]. 2https://snap.stanford.edu/data/higgs-twitter.html
Dataset Splits Yes we evaluate AUC values of different models on all of the datasets with five-fold cross-validation setting. Considering the size of datasets being small, we conduct 2-fold cross-validation experiment on this task.
Hardware Specification No The paper discusses memory usage and scalability, but it does not specify any particular CPU, GPU, or other hardware used for running the experiments.
Software Dependencies No The paper mentions algorithms like Skip-gram and SGD, but it does not provide specific software dependencies with version numbers (e.g., specific libraries, frameworks, or their versions).
Experiment Setup Yes For all the embedding based methods, we set their embedding dimensions and the dimension of common embedding in our model to be 200. For all the random walk based methods, we set the width of the window to be ten and select five negative samples. As LINE has two kinds of embedding (the so-called first-order proximity and second-order proximity), we set the embedding dimension for both embeddings to be 100 and concatenate them together. For Node2Vec, we empirically use the best hyper-parameter for training, which is p = 2 and q = 0.5. For the three PMNE models, we will use the hyperparameters given by their original paper. For our algorithm multiplex network embedding (MNE), we simply set the dimension of additional vectors to be 10.