An Adaptive Logical Rule Embedding Model for Inductive Reasoning over Temporal Knowledge Graphs

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 7304 - 7316 December 7-11, 2022 ©2022 Association for Computational Linguistics An Adaptive Logical Rule Embedding Model for Inductive Reasoning over Temporal Knowledge Graphs Xin Mei∗, Libin Yang∗, Zuowei Jiang, Xiaoyan Cai† Northwestern Polytechnical University, Xi’an, China meixin@mail.nwpu.edu.cn, libiny@nwpu.edu.cn jiangzw@mail.nwpu.edu.cn, xiaoyanc@nwpu.edu.cn Abstract Temporal knowledge graphs (TKGs) extrapola- tion reasoning predicts future events based on historical information, which has great research significance and broad application value. Exist- ing methods can be divided into embedding- based methods and logical rule-based meth- ods. Embedding-based methods rely on learned entity and relation embeddings to make pre- dictions and thus lack interpretability. Logi- cal rule-based methods bring scalability prob- lems due to being limited by the learned log- ical rules. We combine the two methods to capture deep causal logic by learning rule em- beddings, and propose an interpretable model for temporal knowledge graph reasoning called adaptive logical rule embedding model for inductive reasoning (ALRE-IR). ALRE-IR can adaptively extract and assess reasons contained in historical events, and make predictions based on causal logic. Furthermore, we propose a one-class augmented matching loss for op- timization. When evaluated on ICEWS14, ICEWS0515 and ICEWS18 datasets, the per- formance of ALRE-IR outperforms other state- of-the-art baselines. The results also demon- strate that ALRE-IR still shows outstanding performance when transferred to related dataset with common relation vocabulary, indicating our proposed model has good zero-shot reason- ing ability.1 1 Introduction Knowledge graphs (KGs) are a form of structured human knowledge. They represent events as triples (subject, relation, object), where subject and object are entities. Entities usually are objects and abstract concepts in the real world, and relations represent relationships between entities. KGs have caused great research both in academia and industry (Dong ∗Equal contribution †Corresponding author 1Our code is released at https://github.com/mxadorable/ ALRET-IR. et al., 2014; Nickel et al., 2015; Wang et al., 2017; Hogan et al., 2021), and have been widely used in many real-world applications including relation ex- traction (Min et al., 2013; Zeng et al., 2015), entity linking (Hua et al., 2015; Mendes et al., 2011), and question answering (Luo et al., 2018; Yih et al., 2015). However, most of the knowledge graphs are incomplete (Shi and Weninger, 2018; Toutanova and Chen, 2015), which affects their effectiveness and limits the performance of KG-based applica- tions. Reasoning over KGs aims to infer new con- clusions based on existing data and predict the missing events, which can effectively alleviate this problem. Traditional knowledge graphs contain only static events, and there is a large amount of available event data with temporal correlations, where entities interact differently over time. There- fore, many temporal knowledge graphs (TKGs) composed of entity interaction data with tempo- ral attributes have emerged (Boschee et al., 2015; Gottschalk and Demidova, 2018, 2019).TKGs ex- tend static triples with timestamp to represent dy- namic events in the form of quadruples (subject, relation, object, timestamp), where timestamp rep- resents valid time of static triple. Compared with traditional static …

Content truncated to 500 words