AI knowledge graph

Knowledge Table is an open source toolkit designed to simplify the process of extracting and exploring structured data from unstructured documents. It enables users to create structured knowledge representations such as tables and charts through a natural language query interface. The toolkit features customizable extraction rules, fine-tuned formatting options, and data provenance displayed through the UI to accommodate a variety of use cases. Its goal is to provide business users with a familiar spreadsheet interface, while providing developers with a flexible and highly configurable backend, ensuring seamless integration with existing RAG workflows.

GraphReasoning is a project that uses generative artificial intelligence technology to transform 1,000 scientific papers into knowledge graphs. Through structured analysis, calculating node degrees, identifying communities and connectivity, and evaluating clustering coefficients and betweenness centralities of key nodes reveal fascinating knowledge architectures. The graph is scale-free, highly interconnected, and can be used for graph reasoning, using transitive and isomorphic properties to reveal unprecedented interdisciplinary relationships for answering questions, identifying knowledge gaps, proposing unprecedented materials designs, and predicting material behavior.

Fact Finder is an open source intelligent question answering system that uses language models and knowledge graphs to generate natural language answers and provide evidence. The system generates Cypher queries by calling a language model, queries the knowledge graph for answers, and uses another language model call to generate the final natural language answer. Key benefits of Fact Finder include its ability to provide transparency, allowing users to view queries and evidence, and providing intuitive evidence through visual subgraphs.

Triplex is an innovative open source model that can convert large amounts of unstructured data into structured data. Its performance in building knowledge graphs exceeds that of gpt-4o, and the cost is only one-tenth of the cost. It greatly reduces the cost of generating knowledge graphs by efficiently converting unstructured text into semantic triples, the basis for knowledge graph construction.

knowledge_graph_maker is a Python library capable of converting arbitrary text into a knowledge graph based on a given ontology. A knowledge graph is a semantic network that represents the network between real-world entities and the relationships between them. This library uses graph algorithms and centrality calculations to help users deeply analyze text content, realize connectivity analysis between concepts, and improve the depth of communication with text through graph retrieval enhanced generation (GRAG) technology.

llm-graph-builder is an application that uses large language models (such as OpenAI, Gemini, etc.) to extract nodes, relationships and their attributes from unstructured data (PDF, DOCS, TXT, YouTube videos, web pages, etc.), and uses the Langchain framework to create a structured knowledge graph. It supports uploading files from local machines, GCS or S3 buckets, or network resources, selecting LLM models and generating knowledge graphs.

Knowledge Graph RAG is an open source Python library that enhances the performance of large language models (LLMs) by creating knowledge graphs and document networks. This library allows users to search and correlate information through graph structures, providing richer context for language models. It is mainly used in the field of natural language processing, especially in document retrieval and information extraction tasks.

prettygraph is a Python-based web application developed by @yoheinakajima that showcases a new UI pattern for dynamically converting text input into a knowledge graph. This project is a rapid prototype that aims to provide a simple UI idea to generate a knowledge graph by updating text highlighting in the UI in real time.

MyGO is a tool for multi-modal knowledge graph completion that improves the accuracy of completion by processing discrete modal information as fine-grained tags. MyGO uses the transformers library to embed text tags and then train and evaluate on multi-modal datasets. It supports custom data sets and provides training scripts to reproduce experimental results.

ULTRA is a basic model for knowledge graph reasoning. A single pre-trained ULTRA model can perform link prediction tasks on any multi-relation graph and support any entity/relation vocabulary. Outperforms many SOTA models trained specifically for each map. Following the pre-training-fine-tuning paradigm of the base model, pre-trained ULTRA checkpoints can be used immediately on any graph for zero-shot inference, or further fine-tuning can be performed. ULTRA provides a unified, learnable, and transferable representation for any knowledge graph. ULTRA uses graph neural networks and a modified version of NBFNet. It does not learn specific entity and relationship embeddings for the downstream graph, but obtains relative relationship representations based on the interactions between relationships.