{"title":"Reinforcement learning with time intervals for temporal knowledge graph reasoning","authors":"Ruinan Liu , Guisheng Yin , Zechao Liu , Ye Tian","doi":"10.1016/j.is.2023.102292","DOIUrl":null,"url":null,"abstract":"<div><p>Temporal reasoning methods have been successful in temporal knowledge graphs (TKGs) and are widely employed in many downstream application areas. Most existing TKG reasoning models transform time intervals into continuous time snapshots, with each snapshot representing a subgraph of the TKG. Although such processing can produce satisfactory outcomes, it ignores the integrity of a time interval and increases the amount of data. Meanwhile, many previous works focuses on the logic of sequentially occurring facts, disregarding the complex temporal logics of various time intervals. Consequently, we propose a <strong>R</strong>einforcement Learning-based Model for <strong>T</strong>emporal Knowledge Graph Reasoning with <strong>T</strong>ime <strong>I</strong>ntervals (RTTI), which focuses on time-aware multi-hop reasoning arising from complex time intervals. In RTTI, we construct the time learning part to obtain the time embedding of the current path. It simulates the temporal logic with relation historical encoding and compute the time interval between two facts through the temporal logic feature matrix. Furthermore, we propose a new method for representing time intervals that breaks the original time interval embedding method, and express the time interval using median and embedding changes of two timestamps. We evaluate RTTI on four public TKGs for the link prediction task, and experimental results indicate that our approach can still perform well on more complicated tasks. Meanwhile, our method can search for more interpretable paths in the broader space and improve the reasoning ability in sparse spaces.</p></div>","PeriodicalId":50363,"journal":{"name":"Information Systems","volume":null,"pages":null},"PeriodicalIF":3.0000,"publicationDate":"2023-10-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Information Systems","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S030643792300128X","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
引用次数: 0
Abstract
Temporal reasoning methods have been successful in temporal knowledge graphs (TKGs) and are widely employed in many downstream application areas. Most existing TKG reasoning models transform time intervals into continuous time snapshots, with each snapshot representing a subgraph of the TKG. Although such processing can produce satisfactory outcomes, it ignores the integrity of a time interval and increases the amount of data. Meanwhile, many previous works focuses on the logic of sequentially occurring facts, disregarding the complex temporal logics of various time intervals. Consequently, we propose a Reinforcement Learning-based Model for Temporal Knowledge Graph Reasoning with Time Intervals (RTTI), which focuses on time-aware multi-hop reasoning arising from complex time intervals. In RTTI, we construct the time learning part to obtain the time embedding of the current path. It simulates the temporal logic with relation historical encoding and compute the time interval between two facts through the temporal logic feature matrix. Furthermore, we propose a new method for representing time intervals that breaks the original time interval embedding method, and express the time interval using median and embedding changes of two timestamps. We evaluate RTTI on four public TKGs for the link prediction task, and experimental results indicate that our approach can still perform well on more complicated tasks. Meanwhile, our method can search for more interpretable paths in the broader space and improve the reasoning ability in sparse spaces.
期刊介绍:
Information systems are the software and hardware systems that support data-intensive applications. The journal Information Systems publishes articles concerning the design and implementation of languages, data models, process models, algorithms, software and hardware for information systems.
Subject areas include data management issues as presented in the principal international database conferences (e.g., ACM SIGMOD/PODS, VLDB, ICDE and ICDT/EDBT) as well as data-related issues from the fields of data mining/machine learning, information retrieval coordinated with structured data, internet and cloud data management, business process management, web semantics, visual and audio information systems, scientific computing, and data science. Implementation papers having to do with massively parallel data management, fault tolerance in practice, and special purpose hardware for data-intensive systems are also welcome. Manuscripts from application domains, such as urban informatics, social and natural science, and Internet of Things, are also welcome. All papers should highlight innovative solutions to data management problems such as new data models, performance enhancements, and show how those innovations contribute to the goals of the application.