Overview of the Knowledge Graph and summary of related presentations at the International Society for the Study of Knowledge Graphs (ISWC)

Machine Learning Artificial Intelligence Natural Language Processing Semantic Web Ontology Knowledge Information Processing Digital Transformation Probabilistic Generative Model Navigation of this blog

Overview of the Knowledge Graph and summary of related presentations at the International Society for the Study of Knowledge Graphs (ISWC)

A Knowledge Graph is a representation of information in a graph structure, which will play an important role in the field of Artificial Intelligence (AI). Knowledge graphs are used to represent the knowledge that multiple entities (e.g., people, places, things, concepts, etc.) have relationships among them (e.g., “A owns B,” “X is part of Y,” “C affects D,” etc.).

Specifically, knowledge graphs play an important role in search engine question answering systems, artificial intelligence dialogue systems, and natural language processing. These systems can use knowledge graphs to efficiently process complex information and provide accurate information to users.

Knowledge graphs are often constructed by automatically collecting information from the Web, with Google’s Knowledge Graph and Microsoft’s Concept Graph being well-known examples. They are also used for information management within companies and organizations, and for building knowledge bases in the medical field. In recent years, as described in “Knowledge Graphs and Big Data Processing,” there has been active activity to automatically generate such knowledge graphs using natural language processing and machine learning technologies.

The following technologies are required to realize knowledge graphs.

Semantic Web technologies: In order to build a knowledge graph, a way to represent structured data is needed. For this reason, Semantic Web technologies are commonly used to represent data in RDF, a graph structure consisting of entities (mainly nodes) and their attributes (mainly labels) connected by relationships (mainly edges). The theory and specific implementation of Semantic Web technologies are described in “Semantic Web Technologies.
Natural language processing technology: In order to construct a knowledge graph, it is necessary to extract information from natural language sentences using natural language processing technology. Specifically, it is necessary to identify proper nouns, predicates, semantic relationships, etc., and convert them into RDF format. The theory and implementation of natural language processing technology are described in “Natural Language Processing Technology.
Data Integration Technology: To build a knowledge graph, it is necessary to integrate data from various sources. This requires the use of data integration techniques to integrate data in different formats and represent them as a single knowledge graph. These will mainly involve integrating similarities of words and sentences using machine learning techniques. Semantic data integration approaches are described in “Ontology Matching Techniques” and database integration is described in “Schema Matching and Mapping,” while machine learning approaches are described in “Deep Learning Techniques” and “Graph Data Processing Algorithms and Applications to Machine Learning/Artificial Intelligence Tasks“.
knowledge engine: A knowledge engine is required to use the knowledge graph. The knowledge engine is an engine that acquires information from the knowledge graph and provides functions such as search, inference, and question answering. Knowledge engines often utilize the SPARQL query language to perform complex queries on the data in the knowledge graph. Approaches to knowledge information are described in “Knowledge Information Processing Techniques,” RDF Stores and SPARQL are described in “RDF Stores and SPARQL,” the theory and implementation of search techniques are described in “Search Techniques,” and inference techniques are described in “Inference Techniques. Question-and-answer techniques are also discussed in “Chatbots and Question-and-Answer Techniques“.

By combining the above techniques, a knowledge graph can be realized. However, the construction of a knowledge graph requires advanced technology and many resources because it requires the collection and accurate conversion of a large amount of data.

By applying machine learning techniques to the knowledge graph, a more accurate knowledge graph can be constructed and its use value can be increased. The following describes how to apply machine learning techniques to the knowledge graph.

Entity Linking: Entity linking is a technology that links entity information obtained from different data sources. By using natural language processing and machine learning technologies, entity characteristics can be learned and entities with different notations, entities in different categories, etc. can be linked as the same entity. By implementing entity linking, entities in the knowledge graph can be automatically linked to each other to construct a richer knowledge graph. Entity linking techniques include labeling words using the NER (Named Entity Recognition) module in various open-source natural language processing tools described in “OpenNLP” and “Introduction to Various Tools for Natural Language Processing,” etc., and labeling sentences as described in “Topic Model“, and link these topics and labels using machine learning techniques described in “Relational Data Learning” and “Graph Data Processing Algorithms and Applications to Machine Learning/Artificial Intelligence Tasks“.
Relationship Extraction: In a knowledge graph, relationships between entities are important information. Machine learning techniques can be used to extract relationships from natural language text. For example, if an entity appears in the context of “A owns B,” it can be extracted that there is an ownership relationship between the entities. This allows us to add new relationships to the knowledge graph. Relationship extraction can be achieved by using the machine learning techniques described in “Relational Data Learning” and “Graph Data Processing Algorithms and Their Application to Machine Learning/Artificial Intelligence Tasks” above, or by using machine learning techniques that combine logical data and probabilistic models as described in “Combining Logic, Rules, and Probability/Machine Learning“.
Entity Classification: In the knowledge graph, there are various categories of entities. For example, there are people, organizations, places, etc. Machine learning techniques can be used to automatically classify entity categories from natural language text. This enables search and analysis based on entity categories in the knowledge graph. Entity classification can be done using the NER (Named Entity Recognition) module in the various open-source natural language processing tools described in “OpenNLP” and “Introduction to Various Tools for Natural Language Processing” above, or by assigning topics to sentences as described in “Topic Model. This can be achieved by assigning topics to sentences described in the “topic model”.
Inference engine: By using the knowledge graph, new relationships can be found by reasoning between entities. Combining logical data and knowledge graphs enables inference using expert systems as described in “Rule Bases, Knowledge Bases, Expert Systems, and Relational Data. Also, flexible inference can be achieved using various meta-heuristics algorithms as described in “Mathematical Metaheuristics: A Note and Probabilistic Generative Models as described in “Probabilistic Generative Models“. The application of reinforcement learning approaches such as those described in “Theory and Algorithms of Various Reinforcement Learning Techniques” and “Theory and Algorithms of Bandit Problems” is also possible. Furthermore, we can also consider causal inference as described in “Statistical Causal Inference and Search“.

There are many applications of the knowledge graph in various fields, as shown below.

Search Engines: Knowledge graphs are used to improve the accuracy of search engines. Knowledge graphs can provide more accurate answers to search queries by graphically representing entities and the relationships between them. For example, if a user enters the query “players who played in tennis matches,” the knowledge graph can identify the entities of players who played in tennis matches and present information about the matches they played in and other related entities.
Natural Language Processing: The knowledge graph can also be used in natural language processing. In natural language processing, machines need to interpret natural language text, and the knowledge graph can be used to extract entities and related relationships.
Robotics/IOT: Knowledge graphs are used for autonomous robot behavior. Knowledge graphs enable robots to self-locate and understand their surroundings, while knowledge about the types and placement of surrounding objects is necessary for robots to be aware of their surroundings.
Data Analysis: Knowledge graphs are also used in data analysis. Knowledge graphs can be used to visualize relationships among data and analyze data interactions and trends. For example, by analyzing the relationship between the number of products sold and weather conditions on a knowledge graph, it is possible to improve product sales strategies and the accuracy of weather forecasts.

As described above, knowledge graphs are expected to be applied in various fields, and in recent years, there have been many active presentations at AI-related international conferences. The following is a summary of presentations related to knowledge graphs at ISWC (International Semantic Web Conference), an international conference on Semantic Web technologies, as an example.

The figure below shows the trends by year. The trend is a large increase in presentations from 2018 to 2020. (The announcement of google knowledge graph was made in 2012)

<ISWC2021>

<ISWC2020>

<ISWC2019>

<ISWC2018>

<ISWC2017>

<ISWC2016>

<ISWC2015>

<ISWC2014>

Entity Linking with Multiple Knowledge Bases: An Ontology Modularization Approach

<ISWC2013>

An Evidence-Based Verification Approach to Extract Entities and Relations for Knowledge Base Population