Detecting Beneficial Feature Interactions for Recommender Systems
Authors: Yixin Su, Rui Zhang, Sarah Erfani, Zhenghua Xu4357-4365
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that our model (i) outperforms existing baselines in terms of accuracy, and (ii) automatically identifies beneficial feature interactions. |
| Researcher Affiliation | Academia | Yixin Su, 1 Rui Zhang, 2 Sarah Erfani, 1 Zhenghua Xu 3 1 University of Melbourne 2 Tsinghua University 3 Hebei University of Technology |
| Pseudocode | Yes | Section D of Appendix lists the pseudocodes of our model and the training procedures. |
| Open Source Code | Yes | Our implementation of the L0-SIGN model is available at https://github.com/suyixin12123/SIGN-pytorch. |
| Open Datasets | Yes | Frappe (Baltrunas et al. 2015) is a context-aware recommendation dataset... Movie Lens-tag (He and Chua 2017) focuses on the movie tag recommendation... Twitter (Pan, Wu, and Zhu 2015) is extracted from twitter sentiment classification... DBLP (Pan et al. 2013) consists of papers with labels... |
| Dataset Splits | Yes | Each dataset is randomly split into training, validation, and test datasets with a proportion of 70%, 15%, and 15%. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper mentions that the model is implemented and available on GitHub, implying Python and PyTorch due to the repository name, but it does not specify concrete version numbers for any software dependencies. |
| Experiment Setup | Yes | In the experiments, we use elementwise mean as both linear aggregation functions ψ( ) and φ( ). The linear function g( ) is a weighted sum function (i.e., g(u i) = w T g u i, where wg Rd 1 are the weight parameters). For the interaction modeling function h( ), we use a MLP with one hidden layer after element-wise product: h(ui, uj) = W h 2 σ(W h 1 (ui uj) + bh 1) + bh 2, where W h 1 , W h 2 , bh 1, bh 2 are parameters of MLP and σ( ) is a Relu activation function. We implement the edge prediction model based on the neural collaborative filtering framework (He et al. 2017), which has a similar form to h( ): fep(ve i , ve j) = W e 2 σ(W e 1 (ve i ve j)+be 1)+be 2. We set node embedding sizes for both interaction modeling and edge prediction to 8 (i.e., b, d = 8) and the sizes of hidden layer for both h and fep to 32. We choose the weighting factors λ1 and λ2 from [1 10 5, 1 10 1] that produce the best performance in each dataset. |