Learning Sequence Encoders for Temporal Knowledge Graph Completion

Abstract

Research on link prediction in knowledge graphs has mainly focused on static multirelational data. In this work we consider temporal knowledge graphs where relations between entities may only hold for a time interval or a specific point in time. In line with previous work on static knowledge graphs, we propose to address this problem by learning latent entity and relation type representations. To incorporate temporal information, we utilize recurrent neural networks to learn timeaware representations of relation types which can be used in conjunction with existing latent factorization methods. The proposed approach is shown to be robust to common challenges in real-world KGs: the sparsity and heterogeneity of temporal expressions. Experiments show the benefits of our approach on four temporal KGs.

A survey on knowledge graph-based recommender systems

“…范数. 进而一些研究者 [105,106] 将动态知识图谱的图嵌入方法运用到知识补全、知识推理 [36] 、机器阅 读 [107] ; (3) 学习新的数据知识时可以保 留原有大部分的知识. 我们可以融合近几年来被提出的增量图嵌入方法 [108,109] 对我们原有推荐场景 进行扩展, 以适应实时性要求较高的推荐问题, 降低对知识图谱与用户交互数据训练的时间成本.…”

Timespan-Aware Dynamic Knowledge Graph Embedding by Incorporating Temporal Evolution

“…Know-Evolve [47] models the occurrence of a fact as a temporal point process and deals with concurrent events based on a problematic formulation. Reference [48] regards timestamps as a sequence of digits (from 0 to 9), then uses LSTMs to encode the relation vectors and the time digits. [49] models the interactions between relations and time, and studies various ways to combine the time embedding vector with relation embedding vector, such as concatenate, sum or dot product operations.…”

3DRTE: 3D Rotation Embedding in Temporal Knowledge Graph

“…HyTE [11] extends TransH to define a hyperplane for each time slice. García-Durán et al [12] propose a general relation-time fusion framework TA-Model with LSTM, and verify its effectiveness on some models including DistMult [4]. Rishab Goel et al [13] also focus on the design of a universal temporal fusion framework, but argue that time information should be fused with entities.…”

“…2). For example, "2022-06-09" will be converted into a token sequence "2y-0y-2y-2y-0m-6m-0d-9d" in which l = 8, following the mapping protocol used in [12] so as to retain the purest symbolic characteristics of time.…”

“…However, methods developed in this way are restricted by original models when dealing with the basic relation patterns. For example, TA-TransE [12] is haunted by the same problem as TransE, because its score function is defined as s + r t − o, where r t is obtained by inputting the relation with time tokens to LSTM. Similar problems also exist in TA-DistMult, an extension to DistMult.…”