Graph-Based Tri-Attention Network for Answer Ranking in CQA
Authors: Wei Zhang, Zeyuan Chen, Chao Dong, Wen Wang, Hongyuan Zha, Jianyong Wang14463-14471
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on three real-world CQA datasets demonstrate GTAN significantly outperforms state-of-the-art answer ranking methods, validating the rationality of the network architecture.In this section, we elaborate the experimental setup and analyze the experimental results, aiming to answer: RQ1: Can GTAN achieve better answer ranking performance than the state-of-the-art methods for answer ranking? RQ2: How do the key model components and information types used in GTAN contribute to the overall performance?Table 2 shows the overall comparison of GTAN with different baselines. |
| Researcher Affiliation | Academia | 1School of Computer Science and Technology, Shanghai Institute for AI Education, East China Normal University 2School of Data Science, Shenzhen Research Institute of Big Data, The Chinese University of Hong Kong, Shenzhen 3Department of Computer Science and Technology, Tsinghua University |
| Pseudocode | No | The paper describes the model components and equations but does not include any pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any statement about releasing source code or a link to a code repository. |
| Open Datasets | Yes | We conduct experiments on three real-world datasets, i.e., Stack Overflow1, Zhihu (collected from its website), and Quora (Lyu et al. 2019). 1https://archive.org/download/stackexchange/stackoverflow.com-Posts.7z |
| Dataset Splits | Yes | We split the datasets into training sets, validation sets, and test sets, according to the ratios of about 8 to 1 to 1. |
| Hardware Specification | Yes | We implement the models by Tensorflow and run the experiments on a GPU (Nvidia Ge Force GTX 1080 Ti) with 11GB memory. |
| Software Dependencies | No | The paper mentions 'Tensorflow' and 'Jieba' but does not provide specific version numbers for these software components. |
| Experiment Setup | Yes | Based on the performance of GTAN on validation datasets, we set the number of propagation layers T = 2. The dimension of representations for words, questions, answers, and respondents are all set to 64. The number of FC layers K is set to 2 for a non-linear transformation. Adam (Kingma and Ba 2015) is adopted for model optimization, with the initial learning rate of 0.0005. |