Multi-hop Reading Comprehension across Documents with Path-based Graph Convolutional Network

Tang, Zihua; Shen, Yongliang; Ma, Xinyin; Xu, Wei; Yu, Jiale; Lü, Weiming

doi:10.24963/ijcai.2020/540

Cited by 13 publications

(12 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In Table II we present the performances of ClueReader in the development and test sets of WIKIHOP and MEDHOP, and compare them with previously published models mainly basing on GNNs. Our model has improved the accuracy of the HDE based on the heterogeneous GCNs in test set from 70.9% to 72.0% and Path-based GCN (with GloVe word embedding setting) in dev set from 64.5% to 66.9%, while Path-based GCN using GloVe and ELMo surpassed our model by 0.5% in the test set, which confirms the conclusion that the initial representations of nodes is extremely critical [25]. However, limited by the architecture and computing resources, we have not used the powerful contextual word embedding like the ELMo and the BERT in our model which can be further addressed.…”

Section: Results and Analysessupporting

confidence: 78%

“…And on account of full usage of the question's contextual information, it applied the bi-directional attention mechanism, both node2query and query2node, which aimed to obtain the query-aware nodes' representations in the reasoning graph for better predictions. And Path-based GCN [25] introduced related entities in graph more than the nodes merely match to the candidates to enhance the performance of the model. Furthermore, the HDE [26] introduced the heterogeneous nodes into GCNs, which contains different granularity levels of information.…”

Section: B Graph Neural Network For Multi-hop Mrcmentioning

confidence: 99%

“…To echo the strategy of the grandmother cells, the multi-angle representations of entities and documents have been constructed previously in the reasoning graph as heterogeneous nodes, then we use attention to activate or deactivate them to select some key node pairs in our reasoning clues, and empirically we regard this process as the reading reasoning in graph. Gating Mechanism Previous study [18] has shown the GNNs are suffering from the smoothing problem when they are calculated by stacking many layers, thus, we overcome this issue by applying a question-aware gating [25] and a general gating mechanism [31] to optimize the procedure.…”

Section: Heterogeneous Reasoning Graph Constructionmentioning

confidence: 99%

See 2 more Smart Citations

Heterogeneous Graph Attention Network for Multi-hop Machine Reading Comprehension

Gao¹,

Ni²,

Gao

et al. 2021

Preprint

View full text Add to dashboard Cite

Multi-hop machine reading comprehension is a challenging task in natural language processing, which requires more reasoning ability and explainability. Spectral models based on graph convolutional networks grant the inferring abilities and lead to competitive results, however, part of them still face the challenge of analyzing the reasoning in a human-understandable way. Inspired by the concept of the Grandmother Cells in cognitive neuroscience, a spatial graph attention framework named ClueReader, imitating the procedure was proposed. This model is designed to assemble the semantic features in multiangle representations and automatically concentrate or alleviate the information for reasoning. The name "ClueReader" is a metaphor for the pattern of the model: regard the subjects of queries as the start points of clues, take the reasoning entities as bridge points, and consider the latent candidate entities as the grandmother cells, and the clues end up in candidate entities. The proposed model allows us to visualize the reasoning graph and analyze the importance of edges connecting two entities and the selectivity in the mention and candidate nodes, which can be easier to be comprehended empirically. The official evaluations in open-domain multi-hop reading dataset WIKIHOP and Drugdrug Interactions dataset MEDHOP prove the validity of our approach and show the probability of the application of the model in the molecular biology domain.

show abstract

Section: Results and Analysessupporting

confidence: 78%

Section: B Graph Neural Network For Multi-hop Mrcmentioning

confidence: 99%

Section: Heterogeneous Reasoning Graph Constructionmentioning

confidence: 99%

See 1 more Smart Citation

Heterogeneous Graph Attention Network for Multi-hop Machine Reading Comprehension

Gao¹,

Ni²,

Gao

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…We also compare our model with two state-ofthe-art GNN models (i.e., SAE and HGN), shown in Table 2. Both of them need to set the number of GNN layers manually while BFR-Graph can pass through all the connected nodes automatically with Model Accuracy HDE (Tu et al, 2019) 68.1 DynSAN (Zhuang and Wang, 2019) 70.1 Path-based GCN (Tang et al, 2020) 70.8 ChainEx 72.2 Longformer * 73.8 Longformer+BFR 74.4 an extremely low risk of over-smoothing (Kipf and Welling, 2017). SAE and HGN set a fixed types of edges, which is still not fine-grained enough, while BFR-Graph define different weights (can up to ∞ different weights depends on the dataset) to distinguish nodes in a finer granularity.…”

Section: Resultsmentioning

confidence: 99%

Breadth First Reasoning Graph for Multi-hop Question Answering

Huang¹,

Yang²

2021

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langua

View full text Add to dashboard Cite

Recently Graph Neural Network (GNN) has been used as a promising tool in multi-hop question answering task. However, the unnecessary updations and simple edge constructions prevent an accurate answer span extraction in a more direct and interpretable way. In this paper, we propose a novel model of Breadth First Reasoning Graph (BFR-Graph), which presents a new message passing way that better conforms to the reasoning process. In BFR-Graph, the reasoning message is required to start from the question node and pass to the next sentences node hop by hop until all the edges have been passed, which can effectively prevent each node from over-smoothing or being updated multiple times unnecessarily. To introduce more semantics, we also define the reasoning graph as a weighted graph with considering the number of co-occurrence entities and the distance between sentences. Then we present a more direct and interpretable way to aggregate scores from different levels of granularity based on the GNN. On Hot-potQA leaderboard, the proposed BFR-Graph achieves state-of-the-art on answer span prediction.

show abstract

“…Gated Graph Neural Network (GGNNs) [43] was designed to encode the node feature with gated recurrent units. Tang et al [44] employed the Gated Relational Graph Convolution Network (Gated-RGCN) to aggregate messages on the path-based reasoning graph, where the attention and gate mechanisms were employed to adjust the usefulness of information propagating across the graph.…”

Section: B Graph-based Reasoningmentioning

confidence: 99%

Coarse-to-Fine Spatial-Temporal Relationship Inference for Temporal Sentence Grounding

Yang

et al. 2021

IEEE Access

View full text Add to dashboard Cite

Temporal sentence grounding aims to ground a query sentence into a specific segment of the video. Previous methods follow the common equally-spaced frame selection mechanism for appearance and motion modeling, which fails to consider redundant and distracting visual information. There is also no guarantee that all meaningful frames can be obtained. Moreover, this task needs to detect the location clues precisely from both spatial and temporal dimensions, but the relationship between spatialtemporal semantic information and query sentence is still unexplored in existing methods. Inspired by human thinking patterns, we propose a Coarse-to-Fine Spatial-Temporal Relationship Inference (CFSTRI) network to progressively localize fine-grained activity segments. Firstly, we present a coarse-grained crucial frame selection module, where the query-guided local difference context modeling from adjacent frames helps discriminate all the coarse boundary locations relevant to the sentence semantics, and the soft assignment vector of locally aggregated descriptors are employed to enhance the representation of selected frames. Then, we develop a fine-grained spatial-temporal relationship matching module to refine the coarse boundaries, which disentangles the spatial and temporal semantic information from query sentence to guide the excavation of visual grounding clues of corresponding dimensions. Furthermore, we devise a gated graph convolution network to incorporate the spatial-temporal semantic information by leveraging a gate operation to highlight frames referred to by the query sentence from spatial and temporal dimensions, and propagate fused information on the graph. Extensive experiments on two benchmark datasets demonstrate that our CFSTRI significantly outperforms most state-of-the-art methods.

show abstract

Multi-hop Reading Comprehension across Documents with Path-based Graph Convolutional Network

Cited by 13 publications

References 2 publications

Heterogeneous Graph Attention Network for Multi-hop Machine Reading Comprehension

Heterogeneous Graph Attention Network for Multi-hop Machine Reading Comprehension

Breadth First Reasoning Graph for Multi-hop Question Answering

Coarse-to-Fine Spatial-Temporal Relationship Inference for Temporal Sentence Grounding

Contact Info

Product

Resources

About