Aspect-augmented Adversarial Networks for Domain Adaptation
<jats:p> We introduce a neural method for transfer learning between two (source and target) classification tasks or aspects over the same domain. Rather than training on target labels, we use a few keywords pertaining to source and target aspects indicating sentence relevance instead of docume...
| Main Authors: | , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
MIT Press - Journals
2021
|
| Online Access: | https://hdl.handle.net/1721.1/135065 |
| _version_ | 1826192052888535040 |
|---|---|
| author | Zhang, Yuan Barzilay, Regina Jaakkola, Tommi |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Zhang, Yuan Barzilay, Regina Jaakkola, Tommi |
| author_sort | Zhang, Yuan |
| collection | MIT |
| description | <jats:p> We introduce a neural method for transfer learning between two (source and target) classification tasks or aspects over the same domain. Rather than training on target labels, we use a few keywords pertaining to source and target aspects indicating sentence relevance instead of document class labels. Documents are encoded by learning to embed and softly select relevant sentences in an aspect-dependent manner. A shared classifier is trained on the source encoded documents and labels, and applied to target encoded documents. We ensure transfer through aspect-adversarial training so that encoded documents are, as sets, aspect-invariant. Experimental results demonstrate that our approach outperforms different baselines and model variants on two datasets, yielding an improvement of 27% on a pathology dataset and 5% on a review dataset. </jats:p> |
| first_indexed | 2024-09-23T09:05:34Z |
| format | Article |
| id | mit-1721.1/135065 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T09:05:34Z |
| publishDate | 2021 |
| publisher | MIT Press - Journals |
| record_format | dspace |
| spelling | mit-1721.1/1350652023-02-24T21:29:25Z Aspect-augmented Adversarial Networks for Domain Adaptation Zhang, Yuan Barzilay, Regina Jaakkola, Tommi Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory <jats:p> We introduce a neural method for transfer learning between two (source and target) classification tasks or aspects over the same domain. Rather than training on target labels, we use a few keywords pertaining to source and target aspects indicating sentence relevance instead of document class labels. Documents are encoded by learning to embed and softly select relevant sentences in an aspect-dependent manner. A shared classifier is trained on the source encoded documents and labels, and applied to target encoded documents. We ensure transfer through aspect-adversarial training so that encoded documents are, as sets, aspect-invariant. Experimental results demonstrate that our approach outperforms different baselines and model variants on two datasets, yielding an improvement of 27% on a pathology dataset and 5% on a review dataset. </jats:p> 2021-10-27T20:10:34Z 2021-10-27T20:10:34Z 2017 2019-05-07T15:38:17Z Article http://purl.org/eprint/type/ConferencePaper https://hdl.handle.net/1721.1/135065 en 10.1162/TACL_A_00077 Transactions of the Association for Computational Linguistics Creative Commons Attribution 4.0 International license https://creativecommons.org/licenses/by/4.0/ application/pdf MIT Press - Journals MIT Press |
| spellingShingle | Zhang, Yuan Barzilay, Regina Jaakkola, Tommi Aspect-augmented Adversarial Networks for Domain Adaptation |
| title | Aspect-augmented Adversarial Networks for Domain Adaptation |
| title_full | Aspect-augmented Adversarial Networks for Domain Adaptation |
| title_fullStr | Aspect-augmented Adversarial Networks for Domain Adaptation |
| title_full_unstemmed | Aspect-augmented Adversarial Networks for Domain Adaptation |
| title_short | Aspect-augmented Adversarial Networks for Domain Adaptation |
| title_sort | aspect augmented adversarial networks for domain adaptation |
| url | https://hdl.handle.net/1721.1/135065 |
| work_keys_str_mv | AT zhangyuan aspectaugmentedadversarialnetworksfordomainadaptation AT barzilayregina aspectaugmentedadversarialnetworksfordomainadaptation AT jaakkolatommi aspectaugmentedadversarialnetworksfordomainadaptation |