Research map / Architectures and efficiency / Attention and transformers
Bridging Towers of Multi-task Learning with a Gating Mechanism for Aspect-based Sentiment Analysis and Sequential Metaphor Identification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use GBM-MTL-BERT instead of Chen and Qian (2020)-BERT | ABSA · Laptop | ABSA-F1: 63.40 -> 65.61 | Helped |
| Use GBM-MTL-BERT instead of Chen and Qian (2020)-BERT | ABSA · Restaurant | ABSA-F1: 75.42 -> 75.73 | Helped |
| Use GBM-MTL-BERT instead of the strongest reported baseline | SMI · VUA | SMI-F1: 76.8 -> 77.6 | Helped |
| Add GBM to the hard-parameter-sharing MTL model (ablation) | MTL on Laptop, Restaurant, and VUA · Laptop, Restaurant, VUA validation sets (average) | ABSA-F1/F1: 67.44 -> 68.39 | Helped |
| Add weighted sum pooling to hard-parameter-sharing MTL (ablation) | MTL on Laptop, Restaurant, and VUA · Laptop, Restaurant, VUA validation sets (average) | ABSA-F1/F1: 67.44 -> 67.96 | Helped |
| Add both GBM and weighted sum pooling (ablation) | MTL on Laptop, Restaurant, and VUA · Laptop, Restaurant, VUA validation sets (average) | ABSA-F1/F1: 67.44 -> 68.53 | Helped |
| Replace weighted sum pooling with average pooling (ablation) | Full MTL model · Laptop | ABSA-F1: 0.29% | Hurt |
| Replace weighted sum pooling with average pooling (ablation) | Full MTL model · Restaurant | ABSA-F1: 0.18% | Hurt |
| Replace weighted sum pooling with average pooling (ablation) | Full MTL model · VUA | SMI-F1: 0.20% | Hurt |
| Replace weighted sum pooling with average pooling (ablation) | Full MTL model · VUA | PoS tagging F1: 0.16% | Hurt |
| Increase the number of blocks from three to four (ablation) | Full MTL model · Laptop, Restaurant, VUA validation sets (average) | ABSA-F1/F1: 0.07% | Helped |
| Use GBM instead of cross-stitch bridging (ablation) | MTL · Laptop, Restaurant, VUA validation sets (average) | ABSA-F1/F1: 67.48 -> 68.39 | Helped |
Closest papers by meaning
- GIRNet: Interleaved Multi-Task Recurrent State Sequence Models (AAAI 2019 · similarity 0.85)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.85)
- Conditionally Adaptive Multi-Task Learning: Improving Transfer Learning in NLP Using Fewer Parameters & Less Data (ICLR 2021 · similarity 0.85)
- An Information-theoretic Multi-task Representation Learning Framework for Natural Language Understanding (AAAI 2025 · similarity 0.84)
- Emergence and Effectiveness of Task Vectors in In-Context Learning: An Encoder Decoder Perspective (ICML 2025 · similarity 0.83)
- Distribution Matching for Multi-Task Learning of Classification Tasks: A Large-Scale Study on Faces & Beyond (AAAI 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.