Collaborative Policy Learning for Open Knowledge Graph Reasoning

Abstract

In recent years, there has been a surge of interests in interpretable graph reasoning methods. However, these models often suffer from limited performance when working on sparse and incomplete graphs, due to the lack of evidential paths that can reach target entities.Here we study open knowledge graph reasoning-a task that aims to reason for missing facts over a graph augmented by a background text corpus.A key challenge of the task is to filter out "irrelevant" facts extracted from corpus, in order to maintain an effective search space during path inference. We propose a novel reinforcement learning framework to train two collaborative agents jointly, i.e., a multi-hop graph reasoner and a fact extractor. The fact extraction agent generates fact triples from corpora to enrich the graph on the fly; while the reasoning agent provides feedback to the fact extractor and guides it towards promoting facts that are helpful for the interpretable reasoning. Experiments on two public datasets demonstrate the effectiveness of the proposed approach. Source code and datasets used in this paper can be downloaded at https://github. com/shanzhenren/CPL.

Knowledge Graph Completion: A Review

“…The current reviews [1], [8], [10], [25], [26], [27], [28], [29], [30] related to knowledge graphs mainly involve research on knowledge representation, knowledge graph construction, knowledge reasoning, etc., lacking of systematic review articles of knowledge graph completion. Literature [31] analyzes the graph completion algorithm based on knowledge representation, but it lacks comprehensiveness. Starting from the definition of the knowledge graph completion problem, this paper classifies the existing methods, and analyzes the advantages and disadvantages of each class, points out applicable fields of each type of method, introduces the evaluation indicators of the knowledge graph completion algorithms, and discusses the main challenges and problems in the field, as well as its potential research directions.…”

Dynamic Anticipation and Completion for Multi-Hop Reasoning over Sparse Knowledge Graph

“…For embedding-based models, we compared with TransE (Bordes et al, 2013), Dist-Mult , ConvE (Dettmers et al, 2018) and TuckER (Balazevic et al, 2019). For multi-hop reasoning, we evaluate the following five models 1 , Neural Logical Programming (NeuralLP) , Neural Theorem Prover (NTP) (Rocktäschel and Riedel, 2017), MINERVA (Das et al, 2018), MultiHopKG (Lin et al, 2018) and CPL 2 (Fu et al, 2019) . Besides, our model has three variations, DacKGR (sample), DacKGR (top) and DacKGR (avg), which use sample, top-one and average strategy (introduced in Section 3.3) respectively.…”

“…As the performance of most existing multi-hop reasoning methods drops significantly on sparse KGs, some preliminary efforts, such as CPL (Fu et al, 2019), explore to introduce additional text information to ease the sparsity of KGs. Although these explorations have achieved promising results, they are still limited to those specific KGs whose entities have additional text information.…”

On Aligning OpenIE Extractions with Knowledge Bases: A Case Study

“…This triple can be represented with the KB formula ∃x : (J. F. K.; dbo:parent; x) ∧ (x; dbo:parent; P. J. K.), because there is no KB relation expressing "grandchild" relationship between two entities. Similarly, Das et al (2016) use multi-hop reasoning between two entities in a KB to infer new relations, while Fu et al (2019) do multi-hop reasoning over OIE data.…”