NLP Knowledge Graph

If you need to contribute to this open source project, please contact me.

• Deep learning for Knowledge-Graph
  • Resource external link
  • The past and present life of the knowledge graph
  • Research summary list
    • Special research
    • Analysis of the related theory of KG&QA
    • Analysis of NLP related papers
    • Research on the Structure of Chinese Financial Documents
    • Knowledge graph related conferences
    • Analogy of facts
    • Dialogue System Technology
    • Related to the commercialization of NLP/Dialogue/KG (for communication and learning only)
    • List of mainstream open source Q&A & conversation systems
    • List of mainstream semantic platforms
    • List of Chinese and English text preprocessing tools
    • Graph storage and query tools
    • Visualization tools
    • List of Chinese and English knowledge graph datasets
  • Contest
  • think
  • Star History
  • illustrate
  • Contact

Deep learning for Knowledge-Graph

Explore the cognitive intelligence series---Trends: 1. Data fusion knowledge; 2. All in LLM. Including knowledge acquisition, knowledge base construction, and a series of technical research and application of Q&A systems based on knowledge base. It involves cutting-edge technologies and papers in the field of NLP.

Resource external link

• NLP-Progress

  Repository to track the progress in Natural Language Processing (NLP), including the datasets and the current state-of-the-art for the most common NLP tasks.

• paperswithcode

  Papers and codes are provided.

• Technology Online

  Personal daily technology and chat websites.

---

The past and present life of the knowledge graph

Serial number	article
1	Why is the "Knowledge Graph" going back to 1956?

Research summary list

Special research

1. A dialogue system based on knowledge graph

Analysis of the related theory of KG&QA

1. A summary of knowledge graphs
2. The Challenge of Knowledge Graph
3. Deep Learning and Knowledge Graph
4. CN-DBpedia: A Never-Ending Chinese Knowledge Extraction System
5. KBQA
6. Zhao Tingyang: How is the self-awareness of artificial intelligence possible? 1-3
7. NLP University Opening Daji--Teacher Li Wei, Simon Fraser University, Ph.D.
8. ACL 2019 A comprehensive summary of the knowledge graph

Analysis of NLP related papers

1. The Illustrated Transformer
2. An Attentive Survey of Attention Models
3. BERT:Bidirectional Encoder Representations from Transformers
4. ERNIE:Enhanced Representation through Knowledge Integration
5. ERNIE: Enhanced Language Representation with Informative Entities
6. Google T5: Text-To-Text Transfer Transformer

Research on the Structure of Chinese Financial Documents

1. Doc2EDAG

Knowledge graph related conferences

serial number	name	grade	type	field
1	ACL	Class A	Academic Conference	AI
2	CVPR	Class A	Academic Conference	AI
3	ICML	Class A	Academic Conference	AI
4	IJCAI	Class A	Academic Conference	AI
5	EMNLP	Class B	Academic Conference	AI
6	CIKM	Class B	Academic Conference	Database/Data Mining/Content Retrieval
7	AAAI	Class A	Academic Conference	AI
8	SIGKDD	Class A	Academic Conference	Database/Data Mining/Content Retrieval
9	TKDE	Class A	Academic Journals	(Database/Data Mining/Content Retrieval)
10	SIGIR	Class A	Academic Conference	Database/Data Mining/Content Retrieval

Analogy of facts

1. Summary of the matter map
2. Bai Shuo: Six Questions and Six Answers at the Map of Things

Dialogue System Technology

1. CSDN|Open Class Notes: A detailed explanation of knowledge management in heuristic dialogues_aispeech
2. A systematic review of dialogue
3. Spoken comprehension technology (SLU) in dialogue systems

Related to the commercialization of NLP/Dialogue/KG (for communication and learning only)

1. Natural language processing technology in voice interaction-xiaomi
2. Lean Knowledge Graph Methodology-Wenyin Internet
3. Research and progress of Xiaoi Robot in Chinese semantic open platform--Xiaoi Robot
4. iFlytek-Looking at the value and challenges of knowledge graph from an application perspective--iFlytek

---

List of mainstream open source Q&A & conversation systems

serial number	name	address
1	questionAnsweringsystem	QuestionAnsweringSystem is a Java-implemented human-computer question and answer system that can automatically analyze questions and give candidate answers.
2	QABasedOnMedicaKnowledgeGraph	From scratch, we build a certain disease-centered medical field knowledge graph, and use this knowledge graph to complete automatic question-and-answer and analysis services. python
3	DeepPavlov	An open source library for deep learning end-to-end dialog systems and chatbots. python

List of mainstream semantic platforms

serial number	name
1	Tencent Wenzhi Chinese Semantic Platform
2	iFLYTEK Open Semantic Platform
3	Bosen Chinese Semantics Open Platform
4	Harbin Institute of Technology Language Cloud

List of Chinese and English text preprocessing tools

serial number	name	Main functions
1	THULAC	Chinese lexical analysis tool, supports C++/JAVA/Python by Tsinghua
2	LTP	Language technology platform pylyp LTP Python packaging by Harbin Institute of Technology
3	HanLP	Support Java
4	Stutter participle	Chinese word participle (only word participle participle), Java, python, C++
5	jiagu	Provides common natural language processing functions such as Chinese word segmentation, part-of-speech annotation, naming entity recognition, keyword extraction, text summary, and new word discovery. Python
6	fudanNLP	Chinese word participle (word participle, part-of-speech annotation, naming entity recognition), supports Java
7	deepdive deepdive	Stanford University’s open source knowledge extraction tool (triple extraction), supports python
8	FudanDNN-NLP3.0	Used for Chinese word segmentation, naming recognition, part-of-speech annotation, sentence classification, semantic analysis, knowledge base access, dialogue question and answer, and supports Java-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
9	Stanford CoreNLP	Multilingual part-of-speech marker, naming entity recognizer, parser (sentence and grammatical structure), reference digester (that is, the question of determining which noun phrase the pronoun points to in the chapter), sentiment analyzer, guided mode learner, open information extractor, Java/Python----- Stanford CoreNLP is Stanford University's natural language processing toolkit, the toolkit requires Java support

Graph storage and query tools

serial number	name	Main functions
1	Neo4j	Open source graph database developed by Java.
2	OrientDB	Open source noSQL database that can handle documents, graphics, and traditional database components. Written by Java, fast storage.
3	Virtuoso	Supports RDF and SPARQL queries.
4	Titan	It can be integrated with Gremlin/Hbase to enable distributed storage and computing graph data processing.
5	Apache Jena-IDB	Operate RDF under JAVA. Among them, TDB uses the triple store to provide persistent store for RDF data. TDB is faster and scalable than RDB and SDB.
6	Cypher	Declarative graph query language, express efficient query and update graph database.
7	Gremlin	A functional data flow language that allows users to express traversal or query of complex property graphs in a concise way.
8	SPARQL	A query language and data acquisition protocol developed for RDF.
9	rdflib	A parser and serializer written based on Python, RDF/XML, N3, NTriples, N-Quads, Turtle, TriX, RDFa and Microdata, supports SPARQL 1.1 query and update statements.

Visualization tools

serial number	name	Main functions
1	ECharts	Baidu open source tool, complete API encapsulation, simple and easy to use, and easy to use, but does not support event processing.
2	Cytoscape.js	For graphics and network, event interactivity is good and easy to use.
3	D3.js	The threshold for use is high, but it supports event processors. D3 has extremely small overhead, supports dynamic behavior of large data sets and interactive animations, and supports rich graphics.

List of Chinese and English knowledge graph datasets

serial number	name	Application tasks
1	OpenKG	Open Knowledge Graph
2	CN-Probase	Large-scale Chinese concept map
3	SentiBridge	Chinese entity emotional knowledge base, depicting how people describe an entity, including news, tourism, and catering, a total of 300,000 pairs
4	Music Knowledge Graph	Chinese music knowledge graph, singers, songs and other information
5	Character RDF knowledge	Collected character knowledge from the encyclopedia website, a total of 650,000 RDF triples
6	Knowledge Graph of Chinese Tourism Attractions	The Chinese Knowledge Map of Chinese Tourism Attractions is part of the CASIA-KB Knowledge Map. Extracted from Baidu Encyclopedia and Interactive Encyclopedia. The knowledge map of tourist attractions can be used in geography, life, entertainment and other applications.
7	2 million product portrait data	This data is a summary of the product portrait data accumulated by Bai Dan in the seven years of operation. Bai Dan has built a rich e-commerce classification system and media classification system.
8	Chinese Symptom Bank	This is a dataset that contains symptom entities and symptom-related triplets. The data from the Chinese symptom database comes from 8 mainstream health consultation websites, 3 Chinese encyclopedia websites and electronic medical records.
9	cnSchema Airport Knowledge Graph	The airport knowledge graph based on cnSchema can query the properties of airports around the world, including name, time zone, airport code, geographical location (latitude and longitude), etc.
10	Seven-character verses-General Knowledge Graph	This data contains a total of 80 million encyclopedia triplets, which are part of the subset of the seven-character poems, and will continue to be more open in the future. Qiluo-7Lore is an encyclopedia knowledge graph carefully created by Dogtail Grass Technology. It is a collection of massive knowledge in the human world. It contains things, facts, concepts, rules, etc.

Contest

1. ATEC Ant Developer Competition Learning Competition--NLP Problem Similarity Calculation-2019-3
2. Baidu 2019 Language and Intelligent Technology Competition Second Track--Knowledge-Driven Dialogue--2019-4
3. SMP2019 Chinese Human-Computer Dialogue Technology Evaluation (ECDT), Subtask: Domain Classification, Intent Recognition and Semantic Slot Filling -Task1: Natural Language Understanding Evaluation-2019-5

---

think

• The construction of knowledge graph is a systematic engineering project. Currently, combining deep learning technology and natural language processing knowledge has improved the efficiency of construction to a certain extent and reduced costs. The status of the knowledge graph is in the sense that cognitive intelligence is a basic resource facility, based on which it can be used in question-and-answer, dialogue, recommendation systems, and enhance interpretability.
• At present, there is a lot of demand for knowledge graph construction in vertical fields. In terms of application, the practice of intelligent question-and-answer and active dialogue systems is a relatively hot topic.
• I personally think that a crawler system can provide a dynamically updated knowledge source for the knowledge base.
• A dialogue system that combines knowledge/common sense, with more accuracy and diversity, is a direction worth studying, allowing machines to guide users to communicate topics. For example, the above game 2 is the application in this direction.
• The threshold for natural language processing/comprehension has been raised to BERT, which is basically a must-have in the vast number tasks, based on Transformer as an encoder! How to integrate knowledge (can be multimodal) into model training has become a hot topic in research. For example, Baidu recently proposed that ERNIE representation model introduces this idea, which has also achieved good results in downstream tasks.
• Knowledge Graph Embedding (KGE) is a current research hotspot.
• The integration of multimodal data knowledge should be an important exploration stage for domain machine intelligence to move towards general knowledge intelligence. First, due to the characteristics of multimodal data itself, structure is very important. After structure, it can greatly enrich the knowledge composition of the knowledge graph and enrich the expression. Second, it feels more interesting if the CV and NLP components can be connected and work together, such as the reply content of the Q&A scenario is richer and more natural communication with people.
• <strong> For the structure of text data, in addition to using machine learning methods, regular expressions can also be used for data extraction, intermediate processing links in modeling, etc. For example: regular expression combined with deep learning </strong>
• 20230226: The development of NLP is another paradigm change from the second half of 2022 to 2023. With the advent of ChatGPT, everyone inside and outside the field has been known, which has caused a hot discussion in all walks of life, and major domestic and foreign companies have followed suit. For example, LLM (Large Language Model), AIGC (AI Generated Content), and some subdivision tasks of NLP have become unified due to the LLM, and LLM has become a topic of infrastructure in various industries. The emergence of ChatGPT is to give people a surprise, to give people new exploration possibilities, and to AI, to be popularized.
  • From the perspective of experiencing chatgpt, I basically covered all the previous thoughts, including information extraction, dialogue, opinion generation, thinking chain--there is reasoning, new knowledge can be added to update, and code. The most surprising thing is that I can talk seriously and list the steps for logical problems.
  • Learn more about the principles from the paper. For the development of GPT-1 to GPT3, understand the Transformer infrastructure.
• 20230521: Domestic LLM follow-up speed is very fast, and self-developed large language models have been released, and there are also open source ones, such as Tsinghua’s ChatGLM-6/130b. The ecosystem around LLM is also developing rapidly. Various applications, such as Chatpdf, plug-ins, etc. Recently, OpenAI released the ChatGpt IOS version, which has taken the implementation of LLM a big step. It can imagine that many applications in the future will be based on the capabilities of LLM. Regarding how to build some applications based on LLM, the open source framework LangChain iterates very quickly. A few lines of code can allow you to quickly experience such as Q&A based on personal documents, vector calculation, storage, etc., which is convenient and fast. It is recommended that everyone learn and use it. For text, most text is actually unstructured, such as pdf, word, html, etc., need to be structured based on other tools, and langchain also provides an open source method. The ability of LLM privatized deployment + localized knowledge access + plug-in expansion of LLM is the current development direction.

Star History

illustrate

• The above content is all related to the individual's learning process (including Internet content, personal notes, etc.). I hope it will be helpful to friends in the field of studying cognitive intelligence.

Contact

• CSDN
• Personal website
• wiki-technical summary
• Email: [email protected]
• Update time: 2023.5.21
• Li Hanghang