Generative language model for few-shot aspect-based sentiment analysis
Sentiment analysis is a task in natural language processing. The embodiments are directed to using a generative language model to extract an aspect term, aspect category and their corresponding polarities. The generative language model may be trained as a single, joint, and multi-task model. The single-task generative language model determines a term polarity from the aspect term in the sentence or a category polarity from an aspect category in the sentence. The joint-task generative language model determines both the aspect term and the term polarity or the aspect category and the category polarity. The multi-task generative language model determines the aspect term, term polarity, aspect category and category polarity of the sentence.
1. A method for generating a sentiment analysis, the method comprising:
receiving, at an aspect-based sentiment analysis (ABSA) generative language model stored in a memory, a sentence expressing a sentiment of a user; and
generating, using the ABSA generative language model, a plurality of sentiment pairs, wherein at least one pair includes an aspect term from the sentence and a polarity of a sentiment associated with the aspect term, and at least one pair includes an aspect category associated with the sentence and a polarity of a sentiment associated with the aspect category.
2. The method of claim 1 , wherein the ABSA generative language model is a GPT2 model.
3. The method of claim 1 , wherein the aspect term is at least one word or a span of text in the sentence that is associated with the sentiment of the term.
4. The method of claim 1 , further comprising:
training the ABSA generative language model to determine the aspect term using an aspect term extraction task.
5. The method of claim 1 , further comprising:
training the ABSA generative language model to determine the polarity of the aspect term using an aspect term polarity task.
6. The method of claim 1 , further comprising:
training the ABSA generative language model to determine the aspect category using is an aspect category detection task.
7. The method of claim 1 , further comprising:
training the ABSA generative language model to determine the polarity associated with the aspect category using an aspect category polarity task.
8. The method of claim 1 , wherein the aspect term or the aspect category are conditioned on other terms in the sentence.
9. The method of claim 1 , wherein the generating further comprises:
determining the aspect category from a set of predefined categories.
10. A method for generating a sentiment analysis, the method comprising:
receiving, at an aspect-based sentiment analysis (ABSA) generative language model stored in a memory, a sentence expressing a sentiment of a user; and
generating, using the ABSA generative language model, at least one pair, the at least one pair including an aspect term in the sentence and a term polarity associated with the aspect term.
11. The method of claim 10 , wherein the ABSA generative language model is a GPT2 model.
12. The method of claim 10 , wherein the aspect term is at least one word or a span of text in the sentence that is associated with the sentiment of the aspect term.
13. The method of claim 10 , further comprising:
training the ABSA generative language model to determine the aspect term using an aspect term extraction task.
14. The method of claim 10 , further comprising:
training the ABSA generative language model to determine the term polarity of the aspect term using an aspect term polarity task.
15. The method of claim 11 , wherein a training dataset includes the sentence appended with the aspect term and the term polarity.
16. A method for generating a sentiment analysis, the method comprising:
receiving, at an aspect-based sentiment analysis (ABSA) generative language model stored in a memory, a sentence expressing a sentiment of a user and an aspect category from a predefined set of categories; and
generating, using the ABSA generative language model, the sentence, and the aspect category, a category polarity associated with the aspect category, wherein the category polarity corresponds to the sentiment.
17. The method of claim 16 , further comprising:
training the ABSA generative language model to determine the category polarity associated with the aspect category using an aspect category polarity task.
18. The method of claim 17 , wherein a training dataset includes the sentence appended with the aspect category and the category polarity.
19. The method of claim 16 , further comprising:
wherein the receiving further comprises receiving, together with the aspect category an aspect term; and
wherein the generating further comprises, generating, using the ABSA generative language model, the sentence, and the aspect term, a term polarity associated with the aspect term, wherein the term polarity corresponds to a sentiment of the aspect term.
20. The method of claim 19 , further comprising:
training the ABSA generative language model to determine the term polarity associated with the aspect term using an aspect term polarity task.