Name
Abstract | ||
|---|---|---|
It is essential for the click-through rate prediction to learn informative representations. Some studies exploit user behavior data to learn user interest representation. These models usually integrate user historical data into the represented embedding without considering user relevance. Such an approach easily leads to suboptimal representations since it fails to capture the high-order collaboration signal. In this paper, we propose the Evolution Graph-based Interest Modeling (EGIM) to transform user behavior data into a dynamic structure. the user-item interaction is presented as an evolution graph. Graph Convolutional Network is applied as the interest extractor and the Long Short-Term Memory is adopted to learn the evolution of user interest. In the end, the relevance of the user interest sequence with the target item is introduced by the attention-based interest-aggregation layer. Extensive experimental results on three real-world datasets demonstrate that EGIM has significant improvements in terms of Recall@20 and NDCG@20 over several state-of-the-art models. |
Year | DOI | Venue |
|---|---|---|
2021 | 10.1007/978-3-030-82153-1_29 | KNOWLEDGE SCIENCE, ENGINEERING AND MANAGEMENT, PT III |
Keywords | DocType | Volume |
CTR Prediction, Interest representation, User-item interaction, Evolution graph, Graph convolutional network, Deep learning | Conference | 12817 |
ISSN | Citations | PageRank |
0302-9743 | 0 | 0.34 |
References | Authors | |
0 | 3 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Jian Hu | 1 | 0 | 1.69 |
| Qing Ding | 2 | 33 | 4.16 |
| Wenyu Zhang | 3 | 0 | 0.34 |