A game theoretic approach to class-wise selective rationalization
© 2019 Neural information processing systems foundation. All rights reserved. Selection of input features such as relevant pieces of text has become a common technique of highlighting how complex neural predictors operate. The selection can be optimized post-hoc for trained models or incorporated di...
| Main Authors: | , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
2021
|
| Online Access: | https://hdl.handle.net/1721.1/129379 |
| _version_ | 1826203195994537984 |
|---|---|
| author | Chang, Shiyu Zhang, Yang Jaakkola, Tommi S |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Chang, Shiyu Zhang, Yang Jaakkola, Tommi S |
| author_sort | Chang, Shiyu |
| collection | MIT |
| description | © 2019 Neural information processing systems foundation. All rights reserved. Selection of input features such as relevant pieces of text has become a common technique of highlighting how complex neural predictors operate. The selection can be optimized post-hoc for trained models or incorporated directly into the method itself (self-explaining). However, an overall selection does not properly capture the multi-faceted nature of useful rationales such as pros and cons for decisions. To this end, we propose a new game theoretic approach to class-dependent rationalization, where the method is specifically trained to highlight evidence supporting alternative conclusions. Each class involves three players set up competitively to find evidence for factual and counterfactual scenarios. We show theoretically in a simplified scenario how the game drives the solution towards meaningful class-dependent rationales. We evaluate the method in single- and multi-aspect sentiment classification tasks and demonstrate that the proposed method is able to identify both factual (justifying the ground truth label) and counterfactual (countering the ground truth label) rationales consistent with human rationalization. The code for our method is publicly available. |
| first_indexed | 2024-09-23T12:33:08Z |
| format | Article |
| id | mit-1721.1/129379 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T12:33:08Z |
| publishDate | 2021 |
| record_format | dspace |
| spelling | mit-1721.1/1293792021-01-13T03:19:25Z A game theoretic approach to class-wise selective rationalization Chang, Shiyu Zhang, Yang Jaakkola, Tommi S Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science © 2019 Neural information processing systems foundation. All rights reserved. Selection of input features such as relevant pieces of text has become a common technique of highlighting how complex neural predictors operate. The selection can be optimized post-hoc for trained models or incorporated directly into the method itself (self-explaining). However, an overall selection does not properly capture the multi-faceted nature of useful rationales such as pros and cons for decisions. To this end, we propose a new game theoretic approach to class-dependent rationalization, where the method is specifically trained to highlight evidence supporting alternative conclusions. Each class involves three players set up competitively to find evidence for factual and counterfactual scenarios. We show theoretically in a simplified scenario how the game drives the solution towards meaningful class-dependent rationales. We evaluate the method in single- and multi-aspect sentiment classification tasks and demonstrate that the proposed method is able to identify both factual (justifying the ground truth label) and counterfactual (countering the ground truth label) rationales consistent with human rationalization. The code for our method is publicly available. 2021-01-12T14:21:19Z 2021-01-12T14:21:19Z 2020-12-21T15:54:25Z Article http://purl.org/eprint/type/ConferencePaper 1049-5258 https://hdl.handle.net/1721.1/129379 Chang, Shiyu, Zhang, Yang and Jaakkola, Tommi S. "A game theoretic approach to class-wise selective rationalization." Advances in Neural Information Processing Systems, 32. en https://papers.nips.cc/paper/2019/hash/5ad742cd15633b26fdce1b80f7b39f7c-Abstract.html Advances in Neural Information Processing Systems Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use. application/pdf Neural Information Processing Systems (NIPS) |
| spellingShingle | Chang, Shiyu Zhang, Yang Jaakkola, Tommi S A game theoretic approach to class-wise selective rationalization |
| title | A game theoretic approach to class-wise selective rationalization |
| title_full | A game theoretic approach to class-wise selective rationalization |
| title_fullStr | A game theoretic approach to class-wise selective rationalization |
| title_full_unstemmed | A game theoretic approach to class-wise selective rationalization |
| title_short | A game theoretic approach to class-wise selective rationalization |
| title_sort | game theoretic approach to class wise selective rationalization |
| url | https://hdl.handle.net/1721.1/129379 |
| work_keys_str_mv | AT changshiyu agametheoreticapproachtoclasswiseselectiverationalization AT zhangyang agametheoreticapproachtoclasswiseselectiverationalization AT jaakkolatommis agametheoreticapproachtoclasswiseselectiverationalization AT changshiyu gametheoreticapproachtoclasswiseselectiverationalization AT zhangyang gametheoreticapproachtoclasswiseselectiverationalization AT jaakkolatommis gametheoreticapproachtoclasswiseselectiverationalization |