Feature Reweighting-Based Factorization Machine for Effective Learning Latent Representation
Factorization machines (FMs) are widely employed as supervised predictors in collaborative recommendation. FMs can efficiently model second-order feature interactions through inner products, which is beneficial for mitigating the negative effects of data sparsity. However, existing research has larg...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10978851/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Factorization machines (FMs) are widely employed as supervised predictors in collaborative recommendation. FMs can efficiently model second-order feature interactions through inner products, which is beneficial for mitigating the negative effects of data sparsity. However, existing research has largely overlooked the potential correlations and attributes among features in FMs. These inherent relationships between features can enhance our understanding and facilitate the modeling of meaningful feature representations. To address this gap and capture intrinsic correlation information in the data, we propose a novel model named Feature Reweighting-based Factorization Machine (FRFM) in this paper. Specifically, we incorporate similarity into FM and quantify the strength of interactions between features using a similarity calculation method based on mutual information. We then introduce a feature reweighting strategy to effectively learn latent representations, ensuring that similar features exhibit comparable first-order weights and second-order embedding vectors based on their similarity. By assigning different weights to different feature pairs, FRFM adeptly captures the potential correlations and attributes among features within the model. Furthermore, FRFM can be seamlessly integrated into other models to enhance their performance. Extensive experiments conducted on six real-world datasets demonstrate the advantages of our proposed FRFM compared to the state-of-the-art methods. |
|---|---|
| ISSN: | 2169-3536 |