Deep Neural Networks for Learning Graph Representations
Authors: Shaosheng Cao, Wei Lu, Qiongkai Xu
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To demonstrate the effectiveness of our model, we conduct experiments on clustering and visualization tasks, employing the learned vertex representations as features. Empirical results on datasets of varying sizes show that our model outperforms other stat-of-the-art models in such tasks. |
| Researcher Affiliation | Academia | Shaosheng Cao Xidian University shelsoncao@gmail.com Wei Lu Singapore University of Technology and Design luwei@sutd.edu.sg Qiongkai Xu Australian National University National ICT Australia CRL Qiongkai.Xu@nicta.com.au |
| Pseudocode | Yes | Table 1: SDAE description Input: PPMI matrix X, Number of SDAE layers Γ 1. Initialize SDAE Set number of nodes nj in layer j, X(1) = X 2. Greedy layer-wised training For j = 2 to Γ 2.1 Construct one hidden layer SDAE with input of X(j) 2.2 Learn hidden layer representation h(j) 2.3 X(j) = h(j) X(j) Rn nj Output: Matrix of vertices representations R |
| Open Source Code | No | Our source code will be released at http://shelson.top. |
| Open Datasets | Yes | 2. Wine, a dataset from UCI Machine Learning Repository (Lichman 2013), is the the results of a chemical analysis of wines grown in the same region in Italy but derived from three different cultivars. ... 3. Wikipedia Corpus is a free and open content online encyclopedia. We picked the April 2010 snapshot from (Shaoul 2010) used in (Huang et al. 2012) as our training corpus for learning word representations. |
| Dataset Splits | No | The paper mentions datasets used for training and testing, but it does not specify explicit train/validation/test splits with percentages or counts. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as CPU/GPU models, memory, or cloud instance types used for running experiments. |
| Software Dependencies | No | The paper mentions algorithms like K-means and t-SNE, but it does not specify any software dependencies with version numbers for reproducibility (e.g., Python, TensorFlow, PyTorch, or specific library versions). |
| Experiment Setup | Yes | In baseline systems, we set walk length η = 40, total walks γ = 80 |V |, where |V | is the number of graph vertices as suggested in (Perozzi, Al-Rfou, and Skiena 2014). To reach optimal performance for each baseline algorithm, we set negative samples λ = 5, context window size K = 10 for Deep Walk and SGNS, as stated in (Levy, Goldberg, and Dagan 2015). For our model, we tuned the dropout ratio which reduces error in the optimization process. As listed in Table 2, the neural networks for 3NG and 6NG have 3 layers and 9NG and Blogcatalog have 4 layers. All neurons were activated by the sigmoid function. |