HopRetriever: Retrieve Hops over Wikipedia to Answer Complex Questions

Authors: Shaobo Li, Xiaoguang Li, Lifeng Shang, Xin Jiang, Qun Liu, Chengjie Sun, Zhenzhou Ji, Bingquan Liu13279-13287

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on the Hotpot QA dataset demonstrate that Hop Retriever outperforms previously published evidence retrieval methods by large margins.
Researcher Affiliation Collaboration Shaobo Li,1 Xiaoguang Li,2 Lifeng Shang,2 Xin Jiang,2 Qun Liu,2 Chengjie Sun,1 Zhenzhou Ji,1 Bingquan Liu1 1Harbin Institute of Technology 2Huawei Noah s Ark Lab
Pseudocode No The paper describes its methods in narrative text and figures, but does not include structured pseudocode blocks or algorithms labeled as such.
Open Source Code No The paper does not provide a direct link to a source code repository, nor does it explicitly state that the code for the methodology is released or available.
Open Datasets Yes Hop Retriever is evaluated on the multi-hop question answering dataset Hotpot QA (Yang et al. 2018), which includes 90,564 question-answer pairs with annotated supporting documents and sentences for training, 7,405 question-answer pairs for development, and 7,405 questions for testing.
Dataset Splits Yes Hotpot QA (Yang et al. 2018), which includes 90,564 question-answer pairs with annotated supporting documents and sentences for training, 7,405 question-answer pairs for development, and 7,405 questions for testing.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions using BERT and BERT-base, but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes We restrict the maximum input sequence length of BERT to 384. In training, the batch size is set to 16, the learning rate is 3 × 10−5, and the number of training epochs is 3. We use beam search with beam size set to 8 at the inference time.