Structural Deep Embedding for Hyper-Networks
Authors: Ke Tu, Peng Cui, Xiao Wang, Fei Wang, Wenwu Zhu
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct extensive experiments on four different types of hyper-networks, including a GPS network, an online social network, a drug network and a semantic network. The empirical results demonstrate that our method can significantly and consistently outperform the state-of-the-art algorithms. |
| Researcher Affiliation | Academia | Ke Tu,1 Peng Cui,1 Xiao Wang,1 Fei Wang,2 Wenwu Zhu1 1 Department of Computer Science and Technology, Tsinghua University 2 Department of Healthcare Policy and Research, Cornell University |
| Pseudocode | Yes | Algorithm 1 The Deep Hyper-Network Embedding (DHNE) |
| Open Source Code | No | The paper does not include an explicit statement about the release of source code or a link to a code repository for the described methodology. |
| Open Datasets | Yes | GPS (Zheng et al. 2010): The dataset describes a user joins in an activity in certain location. The (user, location, activity) relations are used for building the hypernetwork. Movie Lens (Harper and Konstan 2016): This dataset describes personal tagging activity from Movie Lens1. Each movie is labeled by at least one genres. The (user, movie, tag) relations are considered as the hyperedges to form a hyper-network. 1https://movielens.org/ drug2: This dataset is obtained from FDA Adverse Event Reporting System (FAERS). It contains information on adverse event and medication error reports submitted to FDA. We construct hyper-network by (user, drug, reaction) relationships, i.e., a user who has certain reaction and takes some drugs will lead to adverse event. 2http://www.fda.gov/Drugs/ wordnet (Bordes et al. 2013): This dataset consists of a collection of triplets (synset, relation type, synset) extracted from Word Net 3.0. |
| Dataset Splits | No | The paper mentions training and testing splits (e.g., 'randomly hide 20 percentage of existing edges and use the left network to train', 'randomly sample 10% to 90% of the vertexes as the training samples and use the left vertexes to test the performance'), but does not explicitly specify a separate 'validation' split with percentages or counts. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, processor types, memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details, such as library names with version numbers, needed to replicate the experiment. |
| Experiment Setup | Yes | For our model, we use one-layer autoencoder to preserve hyper-network structure and one-layer fully connected layer to learn tuplewise similarity function. The size of hidden layer of autoencoder is set as 64 which is also the representation size. The size of fully connect layer is set as sum of the embedding length from all types, 192. We do grid search from {0.01, 0.1, 1, 2, 5, 10} to tune the parameter α which is shown in Parameter Sensitivity section. Similar to LINE (Tang et al. 2015), the learning rate is set with the starting value ρ0 = 0.025 and decreased linearly with the times of iterations. |