General Information
Welcome to ORKG Ask, an advanced search system designed to help you find and extract valuable information from a vast corpus of research articles. ORKG Ask revolutionizes the way researchers, academics, and enthusiasts navigate through scholarly literature by leveraging cutting-edge technologies to provide precise and relevant answers to your research queries.
Key Features:
- Natural Language Queries: Simply ask a question, and ORKG Ask will handle the rest.
- Semantic Search: Our system understands the context of your query and finds semantically similar research articles.
- Global Search: Conduct searches across the entire indexed corpus to ensure comprehensive and related results.
- Metadata Filtering: Refine your search results by applying filters based on metadata (and content coming soon).
- Specific Information Extraction: Request specific properties or information to be extracted from the related research papers for detailed insights.
How It Works
ORKG Ask employs a sophisticated pipeline to ensure you receive the most relevant information from the vast CORE dataset of research articles. Here’s an overview of the process:
- Query Submission: Start by entering your question into ORKG Ask.
- Semantic Search: The system uses an embedding model to encode your query and searches for semantically similar research articles using a vector store with an Approximate Nearest Neighbor (ANN) algorithm.
- Information Extraction: Once the relevant articles are identified, a large language model extracts the necessary information from the textual content of these articles.
- Filtered Results: Enhance your search with metadata filters to narrow down results and get more precise information.
Technical Details
ORKG Ask leverages the following advanced technologies to deliver its robust search and extraction capabilities:
- Vector Store: We use Qdrant for storing and retrieving vector representations of research articles efficiently.
- Embedding Model: The Nomic embedding model is utilized to encode search queries and research articles into high-dimensional vectors for semantic search.
- Large Language Model: Our system employs the Mistral Instruct 7B v0.2 LLM to extract and interpret relevant information from the selected research articles.
- Dataset: The research articles are sourced from the CORE dataset, a comprehensive collection of open-access research literature.
Using ORKG Ask
To make the most out of ORKG Ask, follow these simple steps:
- Enter Your Question: Input your research question in natural language.
- Apply Filters: Use metadata filters to refine your search if needed.
- Review Results: Browse through the list of semantically similar research articles.
How is ORKG Ask meant to be used:
Ask is not a Large Language Model (LLM) so don't use it the same way you use any other LLM out there.
- Queries that do NOT work:
- Tell me what is the weather like in Hannover
- Show me the research of John Smith
- Hey, how are you?
- Write a SQL query to drop a table
- Summarize papers about LLMs
- What is 4 minus 1?
- IF a car weighs a ton THEN how much fuel does it use
- who is Jon Grace Tumble (the second)?
- Please tell me which is better Coke or Pepsi
- Queries that do work:
- What strategies can improve disaster preparedness and response in vulnerable communities?
- How does exposure to art and culture influence cognitive development?
- What is the Open Research Knowledge Graph used for?
As you can notice that ORKG Ask is built to answer direct and conceret questions about research. Ask shouldn't be used with instructions like you would do a normal LLM (such as ChatGPT). You can even ask questions in other languages than English as long as they align with criteria mentioned above.
Source code
ORKG Ask is fully open source, visit our Gitlab page for more information. In case you experience any issues, or have any feature requests, please use our issue tracker. Or have a look at the individual repositories listed below:
In case of any queries or questions feel free to use the contact us using via the infromation provided in the Contact page. Also we have a feedback button in the website to collect any related comments and observations that you think are useful.