GEOAssist V2.0是一款开源的地质AI应用，能够从PDF文档中提取地学实体，并构建地学知识图谱（GeoKG）。该工具利用大型语言模型（LLM）自动提取地学数据和关联信息，支持本地化运行，确保数据隐私。构建的知识图谱可以形式化信息为相互连接的结构，并以表格形式展示实体及其关联，便于发现新的思考方向。GeoKG可导出为RDF格式，用于图算法和图神经网络（GNN）分析，以发现相似地质构造、预测矿产关联、挖掘资源富集区域模式等。GEOAssist V2.0支持地质位置、地层年代、岩性、矿物、构造和矿体特征等多种实体提取，并可根据用户需求进行定制，服务于地质研究、城市规划、地质灾害防治及自然资源开发等多个领域，助力联合国可持续发展目标。

💡 GEOAssist V2.0是一款开源的地质AI应用，能够高效地从PDF文档中提取地学实体，并构建结构化的地学知识图谱（GeoKG）。通过利用大型语言模型（LLM）技术，该工具实现了地学数据和关联信息的自动化提取，为地质研究和数据分析提供了强大的支持。

🔒 用户可以本地化运行GEOAssist V2.0，确保敏感的地质数据在数据传输过程中得到保护，从而满足严格的数据隐私和安全要求。这种本地部署的模式使得用户能够安心处理大量地质文献和数据。

📊 GEOAssist V2.0构建的GeoKG能够将地质信息形式化为相互关联的结构，并提供表格视图，方便用户直观地识别实体及其关联，从而发现潜在的、非显而易见的联系，激发新的研究思路。GeoKG还可以导出为RDF格式，便于与其他应用程序集成和进行高级分析。

🚀 GeoKG支持多种图算法和图神经网络（GNN）的应用，例如识别相似地质构造、预测缺失的矿产-岩石类型关联、发现预示资源富集区域的地质模式，以及挖掘前所未见的异常地质联系，极大地增强了地质勘探和发现的效率。

🌐 该工具支持提取多种关键地学实体，包括地理位置、地层年代、岩性、矿物、构造和矿体特征等，并且允许用户根据具体的应用场景进行定制。GEOAssist V2.0的应用范围广泛，涵盖了地质研究、城市规划、地质灾害应对以及能源转型等多个重要领域，并致力于支持联合国可持续发展目标。

GEOAssist V2.0: Opensource Geological AI App. Extract geoscience entities from your PDFs and create Geoscience Knowledge Graphs (GeoKG). Surface insights, find patterns, validate structure and support discovery. I’ve added an extra feature this weekend allowing automatic extraction of geoscience data and associations from your PDFs using Large Language Models (LLM).

You can run GEOAssist locally on a single PDF or thousands downloaded by GEOAssist (or files you already have), ensuring data never leaves your firewall for privacy.

Knowledge Graphs formalise information as interconnected structures. You can also view the entities extracted and their associations in tabular form, spotting unusual associations that may lead to new lines of thinking. The GeoKG can be exported via an RDF option (as it can be very large) for use in other applications.

For example, specific graph algorithms can be applied to a GeoKG, which can also be used in Graph Neural Networks (GNN). These can help find similar formations based on shared connections; to run link prediction to identify missing mineral-rock type links, or new plausible mineral associations; for pattern mining to find geological configurations commonly preceding resource-rich areas, and unusual patterns not previously documented; or perhaps discover novel geological connections, e.g. links between tectonics and a mineral previously unassociated.

The GeoKG option uses:
1. Geographical Location
2. Chronostratigraphy (Geol Age)
3. Lithology (Rock Type)
4. Minerals
5. Tectonics
6. Ore Body Feature

However, you can add/change these and extract anything based on your use case. This might be focused on research such as deep time, palaeontology and stratigraphy; urban planning geotechnical engineering; mitigating geohazards and disaster preparedness; to natural resource industry sectors such as water – hydrogeology; and the move towards the energy transition such as economic mining for critical minerals, geothermal, natural hydrogen, oil & gas exploration, carbon capture and storage and underground storage such as radioactive waste etc. Supporting the UN Sustainable Development Goals (SDG).

Out-of-the-box foundation LLMs have been trained on vast amounts of geological content, so have some ‘understanding’ of terminology without the need to perform fine tuning. Hopefully releasing all this code can help towards building equitable and sustainable geodata science and AI capabilities, and help spark new ideas!

I have updated the V1.0 GEOAssist code to V2.0 in Github : https://github.com/PCleverleyGeol/GeoAssist—An-open-source-autonomous-research-agent-for-geoscience-data-and-literature.-