Paper address: http://arxiv.org/abs/2002.00388
This review is the latest review paper on the knowledge map field by the data science authority Philip S. Yu team. The paper describes the overall situation of the knowledge map from the development history of the knowledge map, knowledge representation learning, knowledge acquisition, knowledge application, and future research directions.
First, let's take a look at the abstract of this review paper:
Abstract: Human knowledge provides a formal understanding of the world. Knowledge graphs representing structural relationships among entities have become an increasingly popular research direction for cognitive systems and human intelligence. In this review, we provide a comprehensive overview of knowledge graphs, covering general research topics on 1) knowledge graph representation learning, 2) knowledge acquisition and completion, 3) temporal knowledge graphs and, 4) knowledge applications, and summarize recent breakthroughs and visionary research directions to facilitate future research. We propose a taxonomy and a new taxonomy for a full range of perspectives on these topics. Knowledge graph embedding is organized from four aspects: representation space, scoring function, encoding model and auxiliary information. For knowledge acquisition, especially the completion of knowledge graphs, embedding methods, path inference and logic rule reasoning are reviewed. We further explore several emerging topics, including meta-relational learning, commonsense reasoning, and temporal knowledge graphs. To facilitate future research on knowledge graphs, we also provide curated datasets and open-source libraries on different tasks. Finally, we give a comprehensive outlook on several promising research directions.
Integrating human knowledge is one of the research directions of artificial intelligence (AI). Inspired by solving problems encountered by humans, knowledge representation and reasoning is the ability to represent knowledge for intelligent systems to solve complex tasks. In recent years, knowledge graphs, as a form of structured human knowledge, have attracted great attention from academia and industry. A knowledge graph is a structured representation of facts, consisting of entities, relations, and semantic descriptions. Entities can be objects and abstract concepts in the real world, relations represent the relationship between entities, and the semantic description of entities and their relations contain types and properties with well-defined meanings. Property graphs or attribute graphs are widely used where nodes and relationships have properties or properties.
Features of this paper include:
- A full-perspective classification and a new classification system for knowledge graphs are given. A full-perspective classification of knowledge graph research and a new fine-grained classification system are proposed. Specifically, knowledge graphs are reviewed from three aspects at a high level: knowledge representation learning, knowledge acquisition, and knowledge application. For knowledge representation learning methods, we further propose a fine-grained classification from four perspectives, including: representation space, scoring function, encoding model, and auxiliary information. The classification of knowledge acquisition includes: Knowledge graph completion based on knowledge graph embedding sorting includes relational path reasoning, logical rule reasoning and meta-relational learning. Entity relationship extraction tasks are divided into: entity recognition, entity classification, disambiguation and entity alignment, and relationship extraction is discussed according to the neural network paradigm.
- The latest cutting-edge technologies and emerging topics are summarized, including Transformer-based knowledge encoding, graph neural network (GNN)-based knowledge dissemination, path reasoning-based reinforcement learning, and meta-relational learning.
- A summary and an outlook on future directions provide a summary of each category and highlight promising future research directions.
The overall classification and sorting of the knowledge map in this paper is shown in the figure:
Considering that this article covers a wide range of areas and contains a large amount of knowledge, the author introduces the core parts of knowledge representation learning, knowledge acquisition, and knowledge application separately, and will introduce these parts separately later.
Welcome to the WeChat public account of the same name " Artificial Intelligence Meets Knowledge Graph ", and welcome to Zhihu's column "Artificial Intelligence Meets Knowledge Graph". Also, let us learn and discuss artificial intelligence and knowledge graph technology together.
