Asynchronous Multi-grained Graph Network For Interpretable Multi-hop Reading Comprehension

Authors: Ronghan Li, Lifang Wang, Shengli Wang, Zejun Jiang

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the proposed model on the Hotpot QA dataset and achieve top competitive performance in distractor setting compared with other published models. Extensive experiments show that the asynchronous update mechanism can effectively form interpretable reasoning chains at different granularity levels. Table 1 shows published and unpublished models on both development and blind test set3 of Hotpot QA.
Researcher Affiliation Academia Ronghan Li , Lifang Wang , Shengli Wang and Zejun Jiang School of Computer Science and Engineering, Northwestern Polytechnical University, Xi an, China {lrh000, wsljsj}@mail.nwpu.edu.cn, {wanglf,claud}@nwpu.edu.cn
Pseudocode No The paper describes an algorithm for asynchronous message propagation and provides mathematical formulations, but it does not include a clearly labeled pseudocode block or algorithm figure.
Open Source Code No The paper mentions implementing experiments based on Huggingface and DFGN's open-source implementation with footnotes linking to their repositories, but it does not provide an explicit statement or link for the open-source code of their proposed AMGN model.
Open Datasets Yes We evaluate our method on Hotpot QA [Yang et al., 2018], which is a prevalent benchmark for multi-hop MRC.
Dataset Splits Yes For each question, the Distractor setting contains two gold paragraphs with ground-truth answers and supporting facts and eight negative paragraphs as distractors... We finetune on the training set for 8 epochs, with batch size as 32. Table 1 shows published and unpublished models on both development and blind test set3 of Hotpot QA.
Hardware Specification No The paper does not specify any particular hardware components (e.g., GPU models, CPU types, memory) used for running the experiments.
Software Dependencies No The paper mentions using "Huggingface" and the "official open-source implementation of DFGN" and "Roberta-large" but does not provide specific version numbers for these software components.
Experiment Setup Yes We use Roberta-large for paragraph selection and set K = 3. Since Hotpot QA only requires two-hop reasoning, a two-step graph update is conducted thus T is 2. We finetune on the training set for 8 epochs, with batch size as 32. For optimization, We use BERTAdam with an initial learning rate of 2e 5.