Question Answering on Knowledge Bases and Text using Universal Schema and Memory Networks
By Rajarshi Das, Manzil Zaheer, Siva Reddy, Andrew McCallum. 2017
144 citations recorded by the source corpus.
Existing question answering methods infer answers either from a knowledge base or from raw text. While knowledge base (KB) methods are good at answering compositional questions, their performance is often affected by the incompleteness of the KB. Au contraire, web text contains millions of facts that are absent in the KB, however in an unstructured form. Universal schema can support reasoning on the union of both structured KBs and unstructured text by aligning them in a common embedded space. In this paper we extend universal schema to natural language question answering, employing Memory networks to attend to the large body of facts in the combination of text and KB. Our models can be trained in an end-to-end fashion on question-answer pairs. Evaluation results on Spades fill-in-the-blank question answering dataset show that exploiting universal schema for question answering is better than using either a KB or text alone. This model also outperforms the current state-of-the-art by 8.5 F1 points.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Retrieval-Augmented Generation, Natural Language Processing
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Luke S. Zettlemoyer
- Percy Liang
- Mohit Bansal
- Hannaneh Hajishirzi
- Jianfeng Gao
- Graham Neubig
- Quoc V. Le
- Ruslan Salakhutdinov
In Computer Vision and Pattern Recognition
- Hannaneh Hajishirzi
- Luc Van Gool
- Mohit Bansal
- Luke S. Zettlemoyer
- Dacheng Tao
- Trevor Darrell
- Xiaogang Wang
- Jianfeng Gao
In Retrieval-Augmented Generation
- Hannaneh Hajishirzi
- Danqi Chen
- William W. Cohen
- Bhuwan Dhingra
- Luke S. Zettlemoyer
- Haitian Sun
- Sewon Min
- Ruslan Salakhutdinov
In Natural Language Processing
- Luke S. Zettlemoyer
- Percy Liang
- Hannaneh Hajishirzi
- Mohit Bansal
- Jonathan Berant
- Graham Neubig
- William W. Cohen
- Kyunghyun Cho
Citations in the corpus
- Knowledge Guided Text Retrieval and Reading for Open Domain Question Answering (2019, background): “Our text-match retrieval method extends TF-IDF-based retrieval from Chen et al. (2017) to select top k articles from Wikipedia, split each article into passages, and then run BM25 (Robertson et al., 2009) on top of these passages to score them.”
- E-BERT: Efficient-Yet-Effective Entity Embeddings for BERT (2019, background): “QA has typically been tackled as a supervised problem (e.g., Das et al. (2017); Sun et al. (2018)).”
- Knowledge Graph Based Synthetic Corpus Generation for Knowledge-Enhanced Language Model Pre-training (2021, background): “Das et al. (2017) use universal schema (Riedel et al., 2013) that embeds text and KGs in a shared space for their integration.”
- LEGO: Latent Execution-Guided Reasoning for Multi-Hop Question Answering on Knowledge Graphs (2021, methodology): “…of branches B ∈ [ ∅ , { b 1 } , . . . , { b 1 , b 2 } , . . . , { b 1 , . . . , b n } ] as input, we adopt an order-invariant DeepSets architecture (Zaheer et al., 2017), where we first use a 2-layer MLP to obtain the initial representation for each branch in the set and then use max-pooling,…”
- Learning Cross-Context Entity Representations from Text (2020, methodology): “Subsequently, the contents are very limited in domain and Das et al. (2017) show that it is essential to use the contents of Freebase to do well on this task.”
- TAT-QA: A Question Answering Benchmark on a Hybrid of Tabular and Textual Content in Finance (2021, background): “Though receiving growing interests (Das et al., 2017; Sun et al., 2019; Chen et al., 2020b, 2021), works on hybrid data comprising of unstructured text and structured or semi-structured KB/tables are rare.”
- BERT is Not a Knowledge Base (Yet): Factual Knowledge vs. Name-Based Reasoning in Unsupervised QA (2019, background): “Factual QA is typically tackled as a supervised problem (e.g., Das et al. (2017); Sun et al. (2018)).”
- Large Scale Knowledge Graph Based Synthetic Corpus Generation for Knowledge-Enhanced Language Model Pre-training (2020, background): “(Das et al., 2017) use universal schema (Riedel et al., 2013) that embeds text and KGs in a shared space for their integration.”
- ExFaKT: A Framework for Explaining Facts over Knowledge Graphs and Text (2019, background): “Second, we do not merely estimate the truthfulness of the fact, but also output human-readable semantic traces that support or refute it, while the output of [10] cannot be explained to humans.”
- Knowledge Graph Embedding Based Question Answering (2019, methodology): “[11] achieved this projection by using matrix factorization to incorporate the corpus into the KG, and LSTM to embed a question.”
- Knowledge Aware Conversation Generation with Explainable Reasoning over Augmented Graphs (2019, background): “…texts: In the task of QA, combination of a KG and a text corpus has been studied with a strategy of late fusion (Gardner and Krishnamurthy, 2017; Ryu et al., 2014) or early fusion (Das et al., 2017b; Sun et al., 2018), which can help address the issue of low coverage to answers in KG based models.”
- Statistical learning for semantic parsing: A survey (2019, background): “Reference [83] achieves the state-ofthe-art performance with an F1 score 39.”
- CPGAN: Full-Spectrum Content-Parsing Generative Adversarial Networks for Text-to-Image Synthesis (2019, methodology): “It was then extensively applied in tasks of natural language processing (NLP) [4, 5, 16, 29] and computer vision (CV)[14, 18, 20].”
- Using Natural Language Relations between Answer Choices for Machine Comprehension (2019, background): “, 2018) or an external commonsense knowledge base (Das et al., 2017; Mihaylov and Frank, 2018; Bauer et al., 2018; Weissenborn et al., 2017).”
- CPGAN: Content-Parsing Generative Adversarial Networks for Text-to-Image Synthesis (2020, methodology): “It was then extensively applied in tasks of natural language processing (NLP) [6, 7, 24, 41] and computer vision (CV)[22, 26, 30].”
- PullNet: Open Domain Question Answering with Iterative Retrieval on Knowledge Bases and Text (2019, methodology): “In (Das et al., 2017b) one approach for a hybrid text-based and KB-based QA system is described, where key-value memory networks are used to store text as well as KB facts encoded with a universal schema representation (Riedel et al., 2013).”
- UniK-QA: Unified Representations of Structured and Unstructured Knowledge for Open-Domain Question Answering (2020, background): “Somewhat closer to our approach are UniversalSchemas (Riedel et al., 2013; Das et al., 2017), which embed KB relations and textual relations in a common space.”