Chains of Reasoning over Entities, Relations, and Text using Recurrent Neural Networks

By Andrew McCallum, Arvind Neelakantan, Rajarshi Das, David Belanger. 2016

278 citations recorded by the source corpus.

Our goal is to combine the rich multi-step inference of symbolic logical reasoning with the generalization capabilities of neural networks. We are particularly interested in complex reasoning about entities and relations in text and large-scale knowledge bases (KBs). Neelakantan et al. (2015) use RNNs to compose the distributed semantics of multi-hop paths in KBs; however for multiple reasons, the approach lacks accuracy and practicality. This paper proposes three significant modeling advances: (1) we learn to jointly reason about relations, entities, and entity-types; (2) we use neural attention modeling to incorporate multiple paths; (3) we learn to share strength in a single RNN that represents logical composition across all relations. On a large-scale Freebase+ClueWeb prediction task, we achieve 25% error reduction, and a 53% error reduction on sparse relations due to shared strength. On chains of reasoning in WordNet we reduce error in mean quantile by 84% versus previous state-of-the-art.

Subfields: Artificial Intelligence, Natural Language Processing, Graph Learning

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

In Natural Language Processing

In Graph Learning

Citations in the corpus

A simple neural network module for relational reasoning (2017, background): “iew of some of the most relevant related work. Relational reasoning Relational reasoning is implicit in many symbolic approaches [11, 32] and has been explicitly pursued using neural networks as well [4]. There is recent work applying neural networks to graphs, which are a natural structure for form”
Constructing Datasets for Multi-hop Reading Comprehension Across Documents (2017, methodology): “ucing synthetic links via dense latent embeddings. Several other methods have been proposed, using composition functions such as vector addition (Bordes et al., 2014), RNNs (Neelakantan et al., 2015; Das et al., 2017), and memory networks (Jain, 2016). Another approach is the Neural Theorem Prover (”
End-to-end Differentiable Proving (2017, background): “Furthermore, NTPs are related to path encoding models [14, 16], but instead of encoding paths to predict a target predicate, reasoning steps in NTPs are explicit and only the comparison of symbols uses subsymbolic representations.”
NLProlog: Reasoning with Weak Unification for Question Answering in Natural Language (2019, background): “An area in which neural multi-hop reasoning models have been investigated is Knowledge Base Completion (KBC) (Das et al., 2016; Cohen, 2016; Neelakantan et al., 2015; Rocktäschel and Riedel, 2017; Das et al., 2017; Evans and Grefenstette, 2018).”
RotatE: Knowledge Graph Embedding by Relational Rotation in Complex Space (2018, background): “There are also a large body of relational approaches for modeling the relational patterns on knowledge graphs (Lao et al., 2011; Neelakantan et al., 2015; Das et al., 2016; Rockt ¨ aschel & Riedel, 2017; Yang et al., 2017).”
Towards a Continuous Knowledge Learning Engine for Chatbots (2018, methodology): “Unlike (Neelakantan et al., 2015; Das et al., 2016), while predicting the label for d ∈ Deval, we compute a relation-specific prediction threshold µr corresponding to r using Dval as: µr = 1 2 [µ+r + µ − r ] and infer d as +ve if P(r|h, t) ≥ µr and -ve otherwise.”
DeepPath: A Reinforcement Learning Method for Knowledge Graph Reasoning (2017, background): “Note that many of the recent KG reasoning methods (Neelakantan et al., 2015; Das et al., 2017) still rely on ﬁrst learning the PRA paths, which only operates in a discrete space.”
Entity Linking via Joint Encoding of Types, Descriptions, and Context (2017, background): “, 2016b), or their types (Das et al., 2017), but do not use multiple sources of”
A Survey on Knowledge Graphs: Representation, Acquisition, and Applications (2020, background): “Chain-of-Reasoning [70], a neural attention mechanism to enable multiple reasons, represents logical composition across all relations, entities and text.”
Learning to Exploit Long-term Relational Dependencies in Knowledge Graphs (2019, unknown): “(Section 3) (Guu et al., 2015; McCallum et al., 2017; Yang et al., 2017).”
Knowledge Graph Embedding: A Survey of Approaches and Applications (2017, unknown): “work, please refer to [98], [99], [100].”
Differentiable Reasoning on Large Knowledge Bases and Natural Language (2019, methodology): “Our work is also related to path encoding models (Das et al. 2017) and random walk approaches (Lao, Mitchell, and Cohen 2011; Gardner et al. 2014), both of which lack a rule induction mechanisms, and to approaches combining observable and latent features of the graph (Nickel, Jiang, and Tresp 2014;…”
Neural, symbolic and neural-symbolic reasoning on knowledge graphs (2021, methodology): “…the KG and the extended textural graph Neelakantan et al. (Neelakantan et al., 2015) use RNN to encode the most con ﬁ dent path Chain-of-Reasoning (Das et al., 2017) change the most con ﬁ dent path to multiple paths DeepPath (Xiong et al., 2017) use RL to evaluate the sampled paths AnyBURL…”
A Survey of Knowledge Graph Reasoning on Graph Types: Static, Dynamic, and Multi-Modal (2022, methodology): “LogSumExp [134] designs a logical composition method across all the elements with attention mechanisms for multiple reasoning.”
Query2box: Reasoning over Knowledge Graphs in Vector Space using Box Embeddings (2020, background): “Recently, a promising alternative approach has emerged, where logical queries as well as KG entities are embedded into a low-dimensional vector space such that entities that answer the query are embedded close to the query (Guu et al., 2015; Hamilton et al., 2018; Das et al., 2017).”
Iterative Visual Reasoning Beyond Convolutions (2018, background): “As an active research area, there are recent works also applying neural networks to the graph structured data [42, 17, 27, 23, 35, 7, 32], or attempting to regularize the output of networks with relationships [8] and knowledge bases [20].”
Knowledge Aware Conversation Generation with Explainable Reasoning over Augmented Graphs (2019, background): “Knowledge Graph Reasoning: Previous studies on KG reasoning can be categorized into three lines, path-based symbolic models (Das et al., 2017a; Lao et al., 2011), embedding-based neural models (Bordes et al.”
Go for a Walk and Arrive at the Answer: Reasoning Over Paths in Knowledge Bases using Reinforcement Learning (2017, methodology): “…inference paths in KBs. Multi-hop link prediction approaches (Lao et al., 2011; Neelakantan et al., 2015; Guu et al., 2015; Toutanova et al., 2016; Das et al., 2017) address the problems above, but the reasoning paths that they operate on are gathered by (i) Can learn general rules: . performing…”
RelNN: A Deep Neural Model for Relational Learning (2017, background): “Examples of these works include path-constrained random walks (e.g., (Lao and Cohen 2010; Lao, Mitchell, and Cohen 2011)), and several hybrid approaches (e.g., (Das et al. 2017; Rockt¨aschel and Riedel 2017; Yang, Yang, and Cohen 2017; Wang, Shi, and Yeung 2017)).”