Abstract:Aspect-based sentiment classification (ABSC) is a crucial subtask of fine-grained sentiment analysis (SA), which aims to predict the sentiment polarity of the given aspects in a sentence as positive, negative, or neutral. Most existing ABSC methods based on supervised learning. However, these methods rely heavily on fine-grained labeled training data, which can be scarce in low-resource domains, limiting their effectiveness. To overcome this challenge, we propose a low-resource cross-domain aspect-based sentiment classification (CDABSC) approach based on a pre-training and fine-tuning strategy. This approach applies the pre-training and fine-tuning strategy to an advanced deep learning method designed for ABSC, namely the attention-based encoding graph convolutional network (AEGCN) model. Specifically, a high-resource domain is selected as the source domain, and the AEGCN model is pre-trained using a large amount of fine-grained annotated data from the source domain. The optimal parameters of the model are preserved. Subsequently, a low-resource domain is used as the target domain, and the pre-trained model parameters are used as the initial parameters of the target domain model. The target domain is fine-tuned using a small amount of annotated data to adapt the parameters to the target domain model, improving the accuracy of sentiment classification in the low-resource domain. Finally, experimental validation on two domain benchmark datasets, restaurant and laptop, demonstrates that significant outperformance of our approach over the baselines in CDABSC Micro-F1.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is cross - domain aspect - based sentiment classification (CDABSC) in low - resource domains. Specifically, most existing sentiment classification methods based on supervised learning rely on a large amount of fine - grained labeled data, and in low - resource domains, such data are often scarce, which limits the effectiveness of these methods. To overcome this challenge, the paper proposes a low - resource cross - domain aspect - based sentiment classification method based on pre - training and fine - tuning strategies.
### Main problems:
1. **Domain differences**: The data distributions and features in different domains are significantly different, including differences in vocabulary use, syntactic structures, semantic inconsistencies, and sentiment expression methods. Such differences make it difficult to maintain the features of the source domain and accurately predict sentiment in the target domain in cross - domain sentiment classification tasks.
2. **Sparse domain - specific labeled data**: The target domain may lack sufficient labeled data to train a good model, or there may be a significant gap in the amount of training data between the source domain and the target domain.
### Solutions:
The paper proposes a method based on pre - training and fine - tuning strategies. It uses the labeled data in high - resource domains to pre - train the model, and then uses a small amount of labeled data in the low - resource target domain to fine - tune the model. The specific steps are as follows:
1. **Pre - training stage**: Select a high - resource domain as the source domain, and use a large amount of fine - grained labeled data to pre - train the AEGCN model. Save the best parameters of the pre - trained model.
2. **Fine - tuning stage**: Select a low - resource domain as the target domain, use the parameters of the pre - trained model as the initial parameters, and use a small amount of labeled data to fine - tune the model in the target domain to adapt to the specific features of the target domain and improve the accuracy of sentiment classification.
### Model framework:
- **AEGCN model**: Combine the multi - head self - attention mechanism (MHA) and the enhanced graph convolutional network (GCN), and construct graph convolutional layers through sentence dependency trees to effectively capture the semantic information of aspects and contexts.
- **Pre - training and fine - tuning strategies**: In the pre - training stage, the model is trained with a large amount of labeled data in the source domain; in the fine - tuning stage, a small amount of labeled data in the target domain is used to adjust the model to adapt to the specific features of the target domain.
### Experimental verification:
The paper conducts experimental verification on two benchmark datasets (restaurant and laptop), and the results show that the proposed method significantly outperforms the baseline methods in the cross - domain aspect - based sentiment classification tasks in the low - resource target domain and achieves state - of - the - art performance.
### Contributions:
1. Propose a low - resource cross - domain aspect - based sentiment classification method based on pre - training and fine - tuning strategies.
2. Through experimental verification, it is shown that the strategy of freezing the parameters of the first few layers performs best in the fine - tuning process.
3. Verify the effectiveness of the proposed method on two benchmark datasets, demonstrating its competitiveness in the low - resource target domain.
Through this method, the paper effectively solves the cross - domain aspect - based sentiment classification problem in low - resource domains, providing new ideas and methods for research in related fields.