CA-RNN: Using Context-Aligned Recurrent Neural Networks for Modeling Sentence Similarity
Authors: Qin Chen, Qinmin Hu, Jimmy Xiangji Huang, Liang He
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on three benchmark datasets, namely TREC-QA and Wiki QA for answer selection and MSRP for paraphrase identification, show the great advantages of our proposed model. |
| Researcher Affiliation | Academia | 1Department of Computer Science & Technology, East China Normal University, China 2Information Retrieval & Knowledge Management Research Lab, York University, Canada |
| Pseudocode | No | The paper provides architectural diagrams (Figure 1, Figure 2) and mathematical formulas for its model, but it does not include any explicit 'Pseudocode' or 'Algorithm' blocks or structured steps. |
| Open Source Code | No | The paper provides links to third-party tools used (e.g., 'https://github.com/ma-sultan/monolingual-word-aligner') and word embedding datasets, but it does not include any explicit statement or link indicating that the source code for their proposed CA-RNN model is openly available. |
| Open Datasets | Yes | Two widely used datasets, namely TREC-QA and Wiki QA, are adopted in our experiments. TREC-QA was created by Wang et al. (Wang, Smith, and Mitamura 2007)... Wiki QA (Yang, Yih, and Meek 2015)... We utilize the Microsoft Research Paraphrase corpus (MSRP) (Dolan, Quirk, and Brockett 2004)... For the answer selection task, we use the 100-dimensional Glo Ve word vectors3, which are trained based on the global word co-occurrence (Pennington, Socher, and Manning 2014). For the paraphrase identification task, we concatenate the Glo Ve vectors with the 25-dimensional PARAGRAM vectors4 that are developed for paraphrase tasks (Wieting et al. 2015). |
| Dataset Splits | Yes | Both TREC-QA and Wiki QA have the train, development and test sets, and each sample is labeled as 1 or 0... The MSRP dataset contains 4,076 sentence pairs in the training set, and 1,725 ones in the test set. Since no development set is provided, we randomly select 100 positive pairs (labeled as 1) and 100 negative pairs (labeled as 0) from the training set as the development set. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper mentions algorithms used (e.g., 'bidirectional LSTM', 'Ada Delta algorithm') and references general tools ('Stanford Core NLP tools'), but it does not specify any software dependencies with version numbers (e.g., Python version, library versions like PyTorch or TensorFlow). |
| Experiment Setup | Yes | The dimension of the hidden state is set to 50. We use the Ada Delta (Zeiler 2012) algorithm for parameter update when training. The optimal parameters are obtained based on the best performance on the development set, and then used for evaluation on the test set. For the answer selection task, we use the 100-dimensional Glo Ve word vectors... For the paraphrase identification task, we concatenate the Glo Ve vectors with the 25-dimensional PARAGRAM vectors... |