Improving Domain-Adapted Sentiment Classification by Deep Adversarial Mutual Learning
Authors: Qianming Xue, Wei Zhang, Hongyuan Zha9362-9369
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on multiple public datasets indicate our method obtains the state-of-theart performance, validating the effectiveness of mutual learning through label probers. In this section, we assess the effectiveness of our approach DAML by first clarifying the experimental setup and afterwards analyzing the experimental results. Table 2: Sentiment classification results in terms of Acc and RMSE. The best results are marked in bold. |
| Researcher Affiliation | Academia | Qianming Xue,1 Wei Zhang,1 Hongyuan Zha2 1School of Computer Science and Technology, KLATASDS-MOE, East China Normal University 2School of Computational Science and Engineering, Georgia Institute of Technology |
| Pseudocode | No | The paper describes the model architecture and training process in text and figures, but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | As a byproduct, we will release the source code of our approach1. 1https://github.com/Sleepy Bag/DAML |
| Open Datasets | Yes | We adopt multiple publicly available datasets with different orgins to evaluate DAML. The first pair of source and target datasets is Yelp and IMDB datasets built by (Tang, Qin, and Liu 2015b). Moreover, to investigate the performance in different domains with the same origin, we choose three domains, i.e., Electronics, CD, and Clothing, from the Amazon dataset (Mc Auley et al. 2015). |
| Dataset Splits | Yes | Domain Training Development Test. Yelp 62,522 7,773 8,671 IMDB 67,426 8,381 9,112 Electronics 79,942 9,994 9,994 CD 79,999 10,000 10,000 Clothing 79,990 10,000 10,000. For each domain, there are 80,000 pieces of documents in the training set and 10,000 in both the test set and development set. |
| Hardware Specification | No | The paper does not specify any particular hardware components (e.g., GPU models, CPU types, or memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions using HAN as feature extractors, Word2vec for embeddings, and Adam as the optimizer, along with Bidirectional GRU, but does not provide specific version numbers for any of these software components. |
| Experiment Setup | Yes | By default, η, λD, λM are set as 0.005, 1.0 and 1.0, respectively. Adam is adopted as the optimizer for all the experiments. For a fair comparison, we make other approaches to use HAN as well. Before the training process starts, 200-dimensional word embeddings are learned by Word2vec (Mikolov et al. 2013). The influence of η and λM on performance is also analyzed in detail. |