edX @ From Graph to Knowledge Graph – Algorithms and Applications / Introduction

Link do Curso -> https://courses.edx.org/courses/course-v1:Microsoft+DAT278x+3T2019/course/
Microsoft: DAT278x: From Graph to Knowledge Graph – Algorithms and Applications

Syllabus
Module 1: Introduction and Overview
Module 2: Graph Properties and Applications
Module 3: Graph Representation Learning
Module 4: Knowledge Graph Fundamentals and Construction
Module 5: Knowledge Graph Inference and Applications

Microsoft Academic Graph (MAG) Project, which aims to build the largest publicly accessible academic domain specific knowledge graph, which could assist the human conducting scientific research by leveraging machine's cognitive power.

The graph is nodes connected by edges.
And the knowledge graph is entities connected by relations.

Module 1 - Introduction

What is a Graph - complex systems are everywhere

In biology, there also exists other kind of graph systems such as a protein-protein interaction networks with each node representing a protein and each edge representing the chemical process between different proteins.

Why are graphs important?

Basically, graph is a general language to interpret complex data.
It offers a natural representation to model data across various domains and fields.
In addition, graph also enable us to discover the shared property across different domains and unify problem space between different fields.

• scale free property (power law)
• small world phenomena (6-Degree Separation)
• community organization 
• information diffusion over network

Graph applications

A real world example can be seen in LinkedIn (Viral Marketing). A research study recently showed that between 60 and 90% of LinkedIn new users signed up due to invitation from another user.

What is a Knowledge Graph - definition

So we can see that a knowledge graph has three essential parts. Entities such as location, company, or person. Relationships, that states how entities are linked or related to each other. And attributes, or properties, associated with these entities.

Knowledge Graph datasets

The two largest private knowledge graph are Google's Knowledge Graph where later on  it becomes Knowledge Vault, and Microsoft Satori Knowledge Graph. These two knowledge graphs are powering Google and the Bing search result entity experience and their question-answering experience as well.
For domain-specific knowledge graph, we have Microsoft Academic Graph, LinkedIn Economic Graph, and the Common Sense Knowledge Graph.

Why are Knowledge Graphs important

Such structure is highly similar to how humans organize knowledge in our brain about the world, which has been indicated by various psychological experiments. 

However, based on our Microsoft academic project's, latest statistic, there are more than one million papers are added to our index every month. Or, on average, one new paper every 2.6 seconds. This is simply far beyond our human cognitive capability to consume and understand.

Knowledge graph plays a key role in current personal assistant applications. As a personal assistant takes voice questions, and converts it to text, the machine would try to parse and understand the text questions, and the query behind the knowledge base to generate the answers. 

Knowledge Graph application 1 - recommendation

It includes entity recommendations, and search personalization, as well as personal assistance scenarios.
When people are typing in some keywords into their search engine query box, sometimes, they know exactly what they need, while in some other cases, they are willing to explore and extend their  knowledge, such as learning-related topics or people.

Knowledge Graph application 2 - personalization

The first one is by utilizing user click log and the knowledge extracted from freebase to provide personalized entity recommendation. From the click logs, we can map the clicked URLs to specific entities in knowledge base. Based on the user click behavior, we can understand the personal interest of topics.

Since the Personal Assistant is usually taking questions which were asked by human spoken language, the only additional step required is to convert the speech signal to text. Then, it follows similar steps as the written text as input to query the knowledge base, and generate the answer.