Sequential Recommender System based on Hierarchical Attention Network (SHAN) implementation in Pytorch
Paper citation:
Ying, Haochao, et al. "Sequential recommender system based on hierarchical attention network." IJCAI International Joint Conference on Artificial Intelligence. 2018.
Changes from the original paper:
- The bootstrap iterations have been set as a parameter in the loss function.
- Regularisation for weights has been removed.
- Preprocessing: the data was generated using srdatasets, the prompts are present in the SHAN implementation notebook.
- Gowalla dataset
Latent Dimensions | Precision@1 | Precision@5 | Precision@10 | Recall@1 | Recall@5 | Recall@10 |
---|---|---|---|---|---|---|
5 | 0.00197 | 0.00079 | 0.00092 | 0.00036 | 0.00077 | 0.00155 |
10 | 0.01836 | 0.01023 | 0.00682 | 0.00348 | 0.00965 | 0.01306 |
20 | 0.1082 | 0.03948 | 0.02459 | 0.02327 | 0.04088 | 0.04984 |
50 | 0.44984 | 0.12472 | 0.06807 | 0.08858 | 0.12032 | 0.13117 |
75 | 0.44393 | 0.14911 | 0.08 | 0.08826 | 0.14209 | 0.15189 |
100 | 0.41639 | 0.14046 | 0.07482 | 0.08212 | 0.13366 | 0.14157 |
- Amazon dataset
Latent Dimensions | Precision@1 | Precision@5 | Precision@10 | Recall@1 | Recall@5 | Recall@10 |
---|---|---|---|---|---|---|
5 | 0 | 0 | 0 | 0 | 0 | 0 |
10 | 0 | 0 | 0 | 0 | 0 | 0 |
20 | 0 | 0 | 0.00061 | 0 | 0 | 0.00061 |
50 | 0 | 0 | 0 | 0 | 0 | 0 |
75 | 0 | 0 | 0 | 0 | 0 | 0 |
100 | 0 | 0 | 0 | 0 | 0 | 0 |
- MovieLens20M dataset
Latent Dimensions | Precision@1 | Precision@5 | Precision@10 | Recall@1 | Recall@5 | Recall@10 |
---|---|---|---|---|---|---|
5 | 0 | 0.00048 | 0.0006 | 0 | 0.00024 | 0.00024 |
10 | 0.00217 | 0.00135 | 0.00101 | 0.00022 | 0.00068 | 0.00101 |
20 | 0.00024 | 0.00039 | 0.00056 | 2.00E-05 | 0.00019 | 0.00056 |
50 | 0.00024 | 0.00092 | 0.00072 | 2.00E-05 | 0.00046 | 0.00072 |
75 | 0 | 0.00087 | 0.0008 | 0 | 0.00043 | 0.0008 |
100 | 0 | 0.00034 | 0.00056 | 0 | 0.00017 | 0.00056 |