A Survey of Knowledge Graph Reasoning on Graph Types: Static, Dynamic, and Multimodal (2212.05767v7)

Abstract: Knowledge graph reasoning (KGR), aiming to deduce new facts from existing facts based on mined logic rules underlying knowledge graphs (KGs), has become a fast-growing research direction. It has been proven to significantly benefit the usage of KGs in many AI applications, such as question answering, recommendation systems, and etc. According to the graph types, existing KGR models can be roughly divided into three categories, i.e., static models, temporal models, and multi-modal models. Early works in this domain mainly focus on static KGR, and recent works try to leverage the temporal and multi-modal information, which are more practical and closer to real-world. However, no survey papers and open-source repositories comprehensively summarize and discuss models in this important direction. To fill the gap, we conduct a first survey for knowledge graph reasoning tracing from static to temporal and then to multi-modal KGs. Concretely, the models are reviewed based on bi-level taxonomy, i.e., top-level (graph types) and base-level (techniques and scenarios). Besides, the performances, as well as datasets, are summarized and presented. Moreover, we point out the challenges and potential opportunities to enlighten the readers. The corresponding open-source repository is shared on GitHub https://github.com/LIANGKE23/Awesome-Knowledge-Graph-Reasoning.

• The paper presents a comprehensive survey and bi-level taxonomy categorizing KGR models across static, temporal, and multimodal graphs.
• It examines diverse methodologies including embedding, path-based, rule-based, RNN, and transformer-based approaches tailored to each graph type.
• The paper highlights challenges such as scalability, out-of-distribution reasoning, and multi-relational, explainable modeling as future research avenues.

Knowledge Graph Reasoning on Graph Types: A Comprehensive Survey

This paper presents a comprehensive survey of knowledge graph reasoning (KGR), focusing on the diverse types of knowledge graphs (KGs): static, temporal, and multi-modal. It aims to systematically categorize and review existing KGR models within these graph types, providing insights into their methodologies, performances, and future opportunities.

Overview of Knowledge Graph Reasoning

KGR involves deducing new knowledge from existing facts within KGs by applying learned logical rules. It has significant applications in fields such as question answering and recommendation systems, highlighting its importance in enhancing AI capabilities. The survey categorizes KGR models into three primary types based on the structure of the KGs:

1. Static KGs: Traditional models representing facts as static entities without temporal or multimodal data.
2. Temporal KGs: Incorporate time dynamics to represent changing relationships over time.
3. Multi-Modal KGs: Enrich KGs with diverse data types like images or textual information to support richer reasoning processes.

Taxonomy and Techniques

The paper introduces a bi-level taxonomy to classify KGR models. At the top level, the taxonomy is split into graph types, while the base level is divided into techniques and scenarios.

Static KGR Models:

• Embedding-based Models: Utilize translational, tensor decompositional, and neural network techniques to encode KGs into latent spaces. These models focus on accurate representation learning and ranking of facts. Recent advancements integrate GNNs for enhanced structure learning.
• Path-based Models: Explore logical paths within KGs, often using random walk or reinforcement learning techniques, emphasizing the discovery of reasoning paths.
• Rule-based Models: Employ logical mining to extract and utilize rules for reasoning. These models often combine symbolic reasoning with statistical methods to enhance performance.

Temporal KGR Models:

• RNN-based Models: Rely on RNN variants (LSTM, GRU) to capture temporal dependencies in evolving KGs.
• RNN-agnostic Models: Use time embeddings or operations to incorporate temporal information without relying on RNN structures.

Multi-Modal KGR Models:

• Transformer-based Models: Leverage pre-trained transformer architectures to integrate multi-modal data, attributing to enhanced reasoning capabilities through unified frameworks.
• Transformer-agnostic Models: Employ various fusion techniques to incorporate multi-modal information, extending traditional KG models to handle additional modalities.

Scenarios in KGR

The paper defines reasoning scenarios for static and temporal KGR:

• Transductive and Inductive: In static KGR, these scenarios refer to reasoning about elements seen during training versus newly encountered elements.
• Interpolation and Extrapolation: In temporal KGR, these relate to reasoning within the existing timeline versus predicting future states.

Datasets

The paper provides detailed information on widely used datasets for each KG type, emphasizing the variety and range of available resources for KGR tasks.

Challenges and Future Directions

The survey identifies several challenges and areas for future research:

• Out-of-Distribution Reasoning: Developing models capable of reasoning with new, unseen entities and relations.
• Large-Scale Reasoning: Efficiently handling reasoning tasks in large industrial-scale KGs.
• Multi-Relational Reasoning: Exploiting the complexities of real-world multi-relational data.
• Multi-Modal and Explainable Reasoning: Advancing the integration and explanation of diverse data sources in KGR.
• Interaction with LLMs: Exploring synergies between KGR and LLMs for improved scalability and generalization.

Conclusion

This survey provides an extensive overview of KGR methodologies across various graph types, organizing models into a structured taxonomy and identifying open challenges and potential research avenues. It serves as a valuable resource for researchers in designing and improving KGR systems, contributing to the broader AI landscape.