The Past and Present of Knowledge Graphs

Follow the WeChat account “ML_NLP“

Set it as “Starred“, heavy content delivered first!

From | Zhihu Author | PMwei

Link | https://zhuanlan.zhihu.com/p/37057052

Editor | Deep Learning This Small Matter WeChat Account

This article is for academic sharing only. If there is any infringement, please contact us to delete it.

1 What is a Knowledge Graph??

In simple terms, a knowledge graph is a relational network that connects all different types of information (Heterogeneous Information).

The knowledge graph network has the following 3 characteristics:

1.1 Composed of Nodes (Point) and Edges (Edge)

1.2 Each Node Represents an “Entity” in the Real World, and Each Edge Represents the “Relationship” Between Entities

1.3 The Knowledge Graph is the Most Effective Representation of Relationships

Therefore, the knowledge graph is essentially a semantic network, a type of data structure based on graphs;

2 What Can Knowledge Graphs Do?

Knowledge graph applications can be broadly divided into two categories based on their depth:

One is the general knowledge graph, which is a public version without particularly deep industry knowledge or specialized content, usually solving popular science and common knowledge problems.

The other is the industry knowledge graph, which is a professional version customized based on in-depth research of a specific industry or niche field, primarily solving professional problems in the current industry or niche.

Below, I will introduce the breadth of knowledge graph applications based on these two categories:

2.1 General Knowledge Graph

We mostly encounter general knowledge graphs in our daily lives, primarily applied in internet-facing search, recommendation, Q&A, and other business scenarios;

Here are 3 examples of general knowledge graphs:

2.1.1 Baidu Knowledge Graph (tupu.baidu.com/)

2.1.2 Sogou Search (sogou.com/)

2.1.3 360 Search (so.com)

2.2 Industry Knowledge Graph

The industry knowledge graph refers to knowledge graphs aimed at specific fields, where the target audience needs to consider personnel at all levels within the industry, as different personnel correspond to different operations and business scenarios, requiring a certain depth and completeness. The industry knowledge graph has very high accuracy requirements, usually used to assist various complex analysis applications or decision support, with strict and rich data models. Entities in the industry knowledge graph usually have many attributes and industry significance.

2.2.1 Path Querying of Connections

Find the connection path between two users based on the associated entities (e.g., workplace, colleagues, classmates, friends, family, etc.).

2.2.2 Corporate Social Graph Querying

Based on investments, positions, patents, bidding, and litigation relationships, a network relationship diagram is formed centered on the target company, visually displaying corporate connections.

2.2.3 Querying of Ultimate Beneficial Owners

Find the shareholder with the largest holding percentage based on equity investment relationships, ultimately tracing back to individuals or state resource management departments.

2.2.4 Assisting Credit Review

Based on unified queries of knowledge graph data, comprehensively grasp customer information; avoid issues such as credit reuse and incomplete information caused by isolated systems and data inconsistencies.

2.2.5 Anti-Fraud Group Loan Schemes

The same person uses multiple identities to apply for loans. As detailed below: although lenders A, B, and C have no direct relationship, the knowledge graph shows that all three share certain information, indicating a risk of group loan fraud.

There are more industry cases:

1. Corporate Development Timeline Graph (Financing)

Based on the chronological order of investment and financing events in the corporate knowledge graph, record the financing development history of the company.

2. Competitive Product Analysis

E-commerce platforms often use this; the more similar the knowledge paths of two companies, the tenser the competitive relationship.

3. Credit Reporting System

Associate multiple platform credit records based on existing user information (e.g., education information, identity information, contact information, guarantor or guaranteed person information).

Building a knowledge graph system requires including 5 major parts: knowledge modeling, knowledge acquisition, knowledge fusion, knowledge storage, and knowledge application:

1. Knowledge Modeling:Construct a multi-level knowledge system that defines, organizes, and manages abstract knowledge, attributes, relationships, etc., transforming them into a real database.

2. Knowledge Acquisition:Transform data from different sources and structures into graph data, including structured data, semi-structured data (parsing), knowledge indexing, knowledge reasoning, etc., ensuring data effectiveness and completeness.

3. Knowledge Fusion:Fuse knowledge information from multiple sources and duplicates, including fusion computation, fusion computation engines, manual fusion, etc.

4. Knowledge Storage:Provide reasonable knowledge storage solutions based on business scenarios, with flexible, diverse, and scalable storage solutions.

5. Knowledge Application:Provide graph retrieval, knowledge computation, graph visualization, and other analytical and application capabilities for the constructed knowledge graph. Also provide SDKs for various knowledge computation needs, including basic graph applications, graph structure analysis, graph semantic applications, natural language processing, graph data acquisition, graph statistics, dataset data acquisition, and dataset statistics.

Download 1: Four Essentials
Reply "Four Essentials" in the backend of the Machine Learning Algorithms and Natural Language Processing WeChat account to obtain the four essentials for learning TensorFlow, Pytorch, machine learning, and deep learning!

Download 2: Repository Address Sharing
Reply "Code" in the backend of the Machine Learning Algorithms and Natural Language Processing WeChat account to obtain 195 NAACL papers + 295 ACL 2019 papers with open-source code. The open-source address is as follows: https://github.com/yizhen20133868/NLP-Conferences-Code

Big News! The Machine Learning Algorithms and Natural Language Processing QQ group has officially been established! There are abundant resources in the group, and everyone is welcome to join and learn!

Additional free resources! Deep learning and neural networks, official Chinese tutorials for Pytorch, data analysis using Python, machine learning notes, official Chinese documentation for pandas, effective java (Chinese version), and 20 other welfare resources.

How to obtain: After entering the group, click on the group announcement to get the download link.
Note: Please modify the remarks when adding to [School/Company + Name + Direction]. For example —— Harbin Institute of Technology + Zhang San + Dialogue System.
The account owner and micro-business please consciously bypass. Thank you!

Recommended Reading:
Implementation of NCE-Loss in Tensorflow and word2vec
Overview of multimodal deep learning: summary of network structure design and modal fusion methods
Awesome adversarial machine learning resource list