Less is more: Data-efficient complex question answering over knowledge bases

Hua, Yuncheng; Li, Yuanfang; Qi, Guilin; Wu, Wei; Zhang, Jingyao; Qi, Daiqing

doi:10.1016/j.websem.2020.100612

Cited by 23 publications

(10 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The Q pre was used to train the LSTM-based programmer, which was further optimized through the Policy Gradient (PG) algorithm (Williams, 1992;Sutton et al, 2000) with another 1% unannotated questions from the training set. We denoted this model by PG, which is also a model variant proposed in (Hua et al, 2020). We trained the meta learner on another 2K training samples (Q meta in Alg.1), representing only approx.…”

Section: Methodsmentioning

confidence: 99%

“…NSM annotates the questions and then anchors the model to the high-reward programs by assigning them with a deterministic probability. Neural-Symbolic Complex Question Answering (NS-CQA) model (Hua et al, 2020) augments the NPI approach with a memory buffer to alleviate the sparse reward and data inefficiency problems appear in the CQA task. Complex Imperative Compared with the NPI models, our model can flexibly adapt to the question under processing.…”

Section: Related Workmentioning

confidence: 99%

“…However, BFS inevitably produces a single annotation and is ignorant to many other plausible annotations yielding the correct answer. To alleviate this issue, a second approach was proposed based on reinforcement learning (RL) to use the search policy prescribed by the programmer (Hua et al, 2020;Neelakantan et al, 2016;. Compared to BFS which is a blind search algorithm, the RL-trained programmer can be regarded as an informed search algorithm for target programs.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Few-Shot Complex Knowledge Base Question Answering via Meta Reinforcement Learning

Hua

Li²,

Haffari³

et al. 2020

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)

Self Cite

View full text Add to dashboard Cite

Complex question-answering (CQA) involves answering complex natural-language questions on a knowledge base (KB). However, the conventional neural program induction (NPI) approach exhibits uneven performance when the questions have different types, harboring inherently different characteristics, e.g., difficulty level. This paper proposes a metareinforcement learning approach to program induction in CQA to tackle the potential distributional bias in questions. Our method quickly and effectively adapts the meta-learned programmer to new questions based on the most similar questions retrieved from the training data. The meta-learned policy is then used to learn a good programming policy, utilizing the trial trajectories and their rewards for similar questions in the support set. Our method achieves state-of-the-art performance on the CQA dataset (Saha et al., 2018) while using only five trial trajectories for the top-5 retrieved questions in each support set, and metatraining on tasks constructed from only 1% of the training set. We have released our code at https://github.com/DevinJake/MRL-CQA.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Few-Shot Complex Knowledge Base Question Answering via Meta Reinforcement Learning

Hua

Li²,

Haffari³

et al. 2020

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)

Self Cite

View full text Add to dashboard Cite

show abstract

“…Complex questions contain multiple entities and relationships; therefore, C-KBQA involves multiple knowledge triples [ 12 ], including operations, such as multi-hop, aggregation, logical operations, and reasoning [ 13 ]. At present, the mainstream methods of C-KBQA include semantic parsing, information retrieval, and template matching [ 14 ].…”

Section: Related Workmentioning

confidence: 99%

Complex Knowledge Base Question Answering for Intelligent Bridge Management Based on Multi-Task Learning and Cross-Task Constraints

Yang

et al. 2022

Entropy

View full text Add to dashboard Cite

In the process of bridge management, large amounts of domain information are accumulated, such as basic attributes, structural defects, technical conditions, etc. However, the valuable information is not fully utilized, resulting in insufficient knowledge service in the field of bridge management. To tackle these problems, this paper proposes a complex knowledge base question answering (C-KBQA) framework for intelligent bridge management based on multi-task learning (MTL) and cross-task constraints (CTC). First, with C-KBQA as the main task, part-of-speech (POS) tagging, topic entity extraction (TEE), and question classification (QC) as auxiliary tasks, an MTL framework is built by sharing encoders and parameters, thereby effectively avoiding the error propagation problem of the pipeline model. Second, cross-task semantic constraints are provided for different subtasks via POS embeddings, entity embeddings, and question-type embeddings. Finally, using template matching, relevant query statements are generated and interaction with the knowledge base is established. The experimental results show that the proposed model outperforms compared mainstream models in terms of TEE and QC on bridge management datasets, and its performance in C-KBQA is outstanding.

show abstract

“…Traditional approaches to KGQA rely on semantic parsing (SP) to translate natural language into a logical form. Weakly supervised SP is a wellstudied topic with increasing interest in applying Reinforcement Learning (RL) (Hua et al, 2020;Agarwal et al, 2019). ER is rarely considered in the scope of surveyed solutions, and if it is, it's treated as an independent component and not included in the weak supervision scope (Ansari et al, 2019).…”

Section: Related Workmentioning

confidence: 99%

End-to-End Entity Resolution and Question Answering Using Differentiable Knowledge Graphs

Oliya¹,

Saffari²,

Sen³

et al. 2021

Preprint

View full text Add to dashboard Cite

Recently, end-to-end (E2E) trained models for question answering over knowledge graphs (KGQA) have delivered promising results using only a weakly supervised dataset. However, these models are trained and evaluated in a setting where hand-annotated question entities are supplied to the model, leaving the important and non-trivial task of entity resolution (ER) outside the scope of E2E learning. In this work, we extend the boundaries of E2E learning for KGQA to include the training of an ER component. Our model only needs the question text and the answer entities to train, and delivers a stand-alone QA model that does not require an additional ER component to be supplied during runtime. Our approach is fully differentiable, thanks to its reliance on a recent method for building differentiable KGs . We evaluate our E2E trained model on two public datasets and show that it comes close to baseline models that use handannotated entities.

show abstract

Less is more: Data-efficient complex question answering over knowledge bases

Cited by 23 publications

References 24 publications

Few-Shot Complex Knowledge Base Question Answering via Meta Reinforcement Learning

Few-Shot Complex Knowledge Base Question Answering via Meta Reinforcement Learning

Complex Knowledge Base Question Answering for Intelligent Bridge Management Based on Multi-Task Learning and Cross-Task Constraints

End-to-End Entity Resolution and Question Answering Using Differentiable Knowledge Graphs

Contact Info

Product

Resources

About