Fairwalk: Towards Fair Graph Embedding
Authors: Tahleen Rahman, Bartlomiej Surma, Michael Backes, Yang Zhang
IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that Fairwalk reduces bias under multiple fairness metrics while still preserving the utility.In this section, we first describe our dataset, followed by the set-up of the machine learning model for our experiments. We then present the evaluation results for our Fairwalk using the fairness metrics defined in Section 3 and finally show the recommendation utility in terms of precision and recall. |
| Researcher Affiliation | Academia | CISPA Helmholtz Center for Information Security {tahleen.rahman, bartlomiej.surma, backes, yang.zhang}@cispa.saarland |
| Pseudocode | Yes | Algorithm 1 Fair random walk trace generation |
| Open Source Code | No | The paper does not include an unambiguous statement about releasing the source code for the described methodology or provide a direct link to a code repository. |
| Open Datasets | No | We use Instagram data collected from two of the biggest cities in different English speaking countries, namely London and Los Angeles (LA). The data was collected in 2016 using the Instagram API. The paper does not provide any link, DOI, or specific citation with authors and year to indicate public availability. |
| Dataset Splits | Yes | We iterate our experiments 5 times. To this end we divide our dataset into 5 equal slices. We train a random forest with 100 trees using 4 out of 5 slices, leaving out a different slice each time. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU models, CPU models, memory) used to run the experiments, only general statements about running experiments. |
| Software Dependencies | No | The paper mentions using a 'random forest classifier' and refers to a 'neural network' but does not specify any software libraries or frameworks with version numbers (e.g., scikit-learn version, TensorFlow version, PyTorch version). |
| Experiment Setup | Yes | We use the following hyper-parameters (following [Grover and Leskovec, 2016]), i.e., length of each walk: walk len = 80, number of walks starting from each node in the graph: walk num = 20, number of dimensions of the resulting vector space: d = 128. We train a random forest with 100 trees |