Knowledge Graphs - Computerphile
Education
Introduction
Knowledge graphs are an advanced AI technology widely utilized across various online platforms and applications that we routinely engage with. Whenever you input a query into search engines like Google or Bing, they often employ knowledge graphs in conjunction with other techniques to deliver information and answer your questions effectively.
For instance, if you seek information about "Bush House," the building where this discussion is taking place, search engines will respond with a plethora of documents, potentially some advertisements, and, prominently, specific information about Bush House. This feature might include a picture of the entrance, its address, and perhaps details about the departments at King's College located there. Even without human-level understanding, a computer system can identify that "Bush House" refers to a physical location connected to specific attributes such as its address and operational hours.
The remarkable aspect of knowledge graphs is how they represent the multitude of facts and relationships about entities within the world. When you search for a different structure like the "National Theater," for example, you will retrieve comparable information about its address, though it might lack details related to university departments since it doesn’t serve the same function.
So, how does Google accumulate all this knowledge about our surroundings? The term "knowledge graph" was indeed popularized by Google; however, the underlying principles and technologies have been in development for decades. AI researchers have long been constructing knowledge bases, which are akin to knowledge graphs, designed to encode and organize information systematically.
At their core, knowledge graphs are vast databases, comprising billions of facts about various entities and the interrelations among them. For example, "Bush House is located in London" is a piece of information that gets extracted from diverse documents. Google, or similar organizations, uses sophisticated machinery to process textual data and recognize factual relationships, storing this structured information in knowledge graphs.
In knowledge graph development, we differentiate between concrete data (like specific landmarks) and abstract data (such as categories like theaters and buildings). This structure encompasses common properties shared across entities, thereby aiding in efficient web searches and multiple AI applications like question-answering systems and product recommendations.
When you input a query into a search engine, you might notice suggestions such as "people also asked." Some of these prompts might be relevant, while others can be tangential. This disparity stems from the difficulty of extracting precise information from unstructured data, which includes text, images, and videos. In contrast, structured data residing in a graph is far easier for computers to process and utilize.
A knowledge graph can be mathematically visualized as a set of nodes (entities) connected by edges (relationships). For example, various relationships may exist between Bush House and King's College London, such as "Campus of" or "contains image of." The edges can possess varied meanings and functionalities depending on the defined relationships, allowing for efficient queries regarding shared characteristics among nodes.
Understanding relationships is vital for knowledge graphs. For instance, consider the relationship "has sibling"; this could be symmetrical. Meanwhile, transitive relationships enable the inference of additional knowledge. For example, if we know that Bush House is located in London and London is in England, we can infer that Bush House is in England without needing a specific entry stating that.
Knowledge graphs fundamentally rely on ontologies, structured frameworks that dictate the types of entities and their attributes within a certain domain, such as cities, buildings, and the relationships among them. Open-source resources like Google’s Knowledge Graph API and WikiData enable developers and researchers to interact with and leverage these rich databases for various applications.
In conclusion, knowledge graphs streamline the connection and retrieval of complex relationships within vast datasets, providing specific and accurate answers far more efficaciously than traditional methods.
Keywords
- Knowledge Graphs
- AI technology
- Search Engines
- Structured Data
- Relationships
- Ontology
- Google Knowledge Graph API
- WikiData
FAQ
What are knowledge graphs?
Knowledge graphs are large databases that store facts about entities and their relationships, allowing for efficient information retrieval and inference.
How do search engines use knowledge graphs?
Search engines leverage knowledge graphs to provide specific information about entities, including their attributes and related facts, enhancing user query responses.
What is the difference between knowledge, information, and data?
Data refers to raw facts, information is processed data that has meaning, and knowledge encompasses the understanding and insights gained from analyzing information.
Where can developers access knowledge graph technology?
Developers can access knowledge graph technology through resources like the Google Knowledge Graph API and WikiData, which offer structured information for public use.
Can knowledge graphs infer additional information?
Yes, knowledge graphs have the capability to infer new knowledge through transitive relationships. If A is related to B, and B is related to C, then A can also be inferred to be related to C.
