<p>The GraphRAG project is designed to address the challenge of efficient document retrieval, processing and query answering by combining vector search, BM25 and knowledge graph traversal. The system allows for extracting contextually relevant information from large sets of documents and organizing it into a graph structure, enabling effective querying and response generation. By leveraging large language models (LLMs) and embedding models, GraphRAG offers a powerful solution for retrieving information, analyzing document content and generating accurate answers in real time.</p>

GraphRAG is a document retrieval system that combines vector search, knowledge graph traversal and LLMs for accurate, context-aware query responses.

Graph-Enhanced Retrieval-Augmented Generation (GRAPH-RAG)

<h2>Project Overview</h2><p>The GraphRAG project represents a powerful system that uses vector search techniques and knowledge graph exploration to improve document searches and questioning processes. The system divides extensive documents into segments while creating embeddings, which get stored within a FAISS vector store for quick retrieval searches. A knowledge graph uses content similarity to establish document chunk relationships, which allows the chunks to connect into a comprehensive network. The system utilizes a QueryEngine to process queries by conducting graph traversals, which enable the retrieval of highly pertinent information. When the context falls short, the system uses a <a href="https://www.datacamp.com/blog/what-is-an-llm-a-guide-on-large-language-models?utm_source=google&amp;utm_medium=paid_search&amp;utm_campaignid=19589720824&amp;utm_adgroupid=152984013054&amp;utm_device=c&amp;utm_keyword=&amp;utm_matchtype=&amp;utm_network=g&amp;utm_adpostion=&amp;utm_creative=733936254665&amp;utm_targetid=aud-1645446892440:dsa-2222697810678&amp;utm_loc_interest_ms=&amp;utm_loc_physical_ms=1001441&amp;utm_content=DSA~blog~Artificial-Intelligence&amp;utm_campaign=230119_1-sea~dsa~tofu_2-b2c_3-apac_4-prc_5-na_6-na_7-le_8-pdsh-go_9-nb-e_10-na_11-na-feb25&amp;gad_source=1&amp;gclid=Cj0KCQiA_Yq-BhC9ARIsAA6fbAgvQKXl9_FfhupNC6zZMaA1oIinE1vUzTypZZWwzpXwtCS2eRZatdsaApfSEALw_wcB" target="_blank">large language model</a> (<strong>LLM</strong>) to create the missing answer. The Visualizer tool shows a graphical representation of graph traversal, allowing users to better understand the analysis process. Relevant context-based document understanding produces accurate results and supplies a strong method for textual information retrieval.</p><h2>Prerequisites</h2><ol><li><strong>Python 3.6+</strong>: Required for running the project.  </li><li><strong>Libraries</strong>: Install NetworkX, Matplotlib, FAISS, spaCy, OpenAIEmbeddings, NLTK, LangChain and PyPDFLoader.  </li><li><strong><a href="https://www.aionlinecourse.com/blog/openai-api-error-the-api-key-client-option-must-be-set-either-by-passing-api-key" target="_blank">OpenAI API Key</a></strong>: Needed for embeddings and LLM interactions.  </li><li><strong><a href="https://colab.research.google.com/" target="_blank">Google Colab</a> / Jupyter Notebook</strong>: Recommended for running the system.  </li><li><strong>pip</strong>: For managing and installing dependencies.  </li><li><strong>Basic Python Knowledge</strong>: Familiarity with Python libraries like <strong>pandas</strong>, <strong>numpy</strong> and <strong>scikit-learn</strong>.</li></ol><h2>Approach</h2><p>GraphRAG retrieves documents and generates responses through processes of document processing, knowledge graph building and query expansion. It initially chunked the documents into small pieces through the use of a text splitter and then embedded them using OpenAIEmbeddings; these embeddings were later stored in a FAISS vector store for efficient similarity-based searching. A knowledge graph is then constructed where the node represents a document chunk and the edge follows the content similarity and shared concepts. The QueryEngine traverses this graph using modified <strong><a href="https://en.wikipedia.org/wiki/Dijkstra%27s_algorithm" target="_blank">Dijkstra's algorithm</a></strong> to expand the context around the query to ensure the visiting of the most relevant information, which gives the entire answer; otherwise, a final response would be generated by an LLM. The visualizer graphically represents the graph traversal to make the procedure transparent so that clicking any node can show how the system comes up with that answer. This contribution makes the process a comprehensive, rich contextual, yet efficient querying method; it is highly suitable for applications where deep document understanding is required.</p><h2>Workflow and Methodology</h2><h3>Workflow of GraphRAG System</h3><ol><li><strong>Document Loading</strong>: Documents are loaded using <strong><a href="https://python.langchain.com/docs/integrations/document_loaders/pypdfloader/" target="_blank">PyPDFLoader</a></strong>.  </li><li><strong>Document Processing</strong>: Documents are split into chunks embedded using <strong><a href="https://platform.openai.com/docs/guides/embeddings" target="_blank">OpenAIEmbeddings</a></strong><a href="https://platform.openai.com/docs/guides/embeddings" target="_blank"> </a>and stored in a <strong>FAISS vector store</strong>.  </li><li><strong>Knowledge Graph Construction</strong>: A knowledge graph is built, connecting document chunks based on similarity and shared concepts.  </li><li><strong>Query Handling</strong>: The <strong>QueryEngine</strong> retrieves relevant documents and traverses the knowledge graph to expand the context.  </li><li><strong>Context Expansion</strong>: The engine explores the graph, updates the context and uses an <strong>LLM</strong> to generate an answer if needed.  </li><li><strong>Visualization</strong>: The <strong>Visualizer</strong> shows the graph traversal, highlighting explored nodes and edges.  </li><li><strong>Final Answer</strong>: The system provides the final answer to the query, generated either from the context or via the <strong>LLM</strong>.</li></ol><h3>Methodology</h3><ol><li><strong>Chunking and Embedding:</strong> The documents are chunked into smaller pieces and embedded into vectors, facilitating comparison and retrieval.  </li><li><strong>Graph Construction:</strong> The initial document chunk embeddings and their cosine similarity are used to draw meaningful links in the knowledge graph.  </li><li><strong>Query Expansion:</strong> The system expands the query context while traversing the graph and, thus, retrieves the most relevant information.  </li><li><strong>LLM-based Answering: W</strong>hen the context is still insufficient, the system generates the answer with an LLM according to the expanded context.  </li><li><strong>Graph Visualization:</strong> Matplotlib visualizing the traversal path ensures transparency for how the system arrived at the final answer.</li></ol><h2>Data Collection and Preparation</h2><h3>Data Collection</h3><p>For this project, data is collected in the form of text documents, such as PDFs or other text-based formats. The documents can cover a variety of topics and in this case, a sample document named <strong>"Understanding_Climate_Change.pdf"</strong> is used. The goal is to extract meaningful information from these documents by splitting them into smaller, manageable chunks for further analysis. These documents are processed by the <strong>PyPDFLoader</strong> or similar document loaders to extract the content, which is then stored as a list of text documents for embedding and further processing.</p><h3>Data Preparation Workflow</h3><ul><li><strong>Collect Documents</strong>: Gather text documents (e.g., PDFs).  </li><li><strong>Load Text</strong>: Extract content using <strong>PyPDFLoader</strong>.  </li><li><strong>Chunk Text</strong>: Split documents into smaller chunks with <strong>RecursiveCharacterTextSplitter</strong>.  </li><li><strong>Generate Embeddings</strong>: Convert chunks to vector embeddings using <strong>OpenAIEmbeddings</strong>.  </li><li><strong>Store Embeddings</strong>: Save embeddings in <strong><a href="https://ai.meta.com/tools/faiss/" target="_blank">FAISS</a></strong><a href="https://ai.meta.com/tools/faiss/" target="_blank"> </a>for fast retrieval.  </li><li><strong>Extract Concepts</strong>: Use spaCy and LLMs for named entity and concept extraction.  </li><li><strong>Build Knowledge Graph</strong>: Link document chunks based on similarity and shared concepts.  </li><li><strong>Prepare Data</strong>: Organize the data for efficient querying and analysis.</li></ul>


<h2>Code Explanation</h2><h4>Mounting Google Drive</h4><p>This code mounts Google Drive to Colab, allowing access to files stored in Drive. The mounted directory is /content/drive, enabling seamless file handling.</p>

<h5 class="">Installing Necessary Libraries</h5><p>The code installs Python packages, which are essential for various tasks. The installation process adds langchain-community along with langchain-openai packages for language model usage and rank_bm25 alongside pymupdf and <strong>pypdf</strong> for PDF processing and <strong><a href="https://docs.pydantic.dev/latest/" target="_blank">Pydantic</a></strong><a href="https://docs.pydantic.dev/latest/" target="_blank"> </a>implements data validation and faiss-cpu enables quick similarity searches on extensive datasets.</p>


<h4>Changing Directory in Google Colab</h4><p>The command %cd /content/drive/MyDrive/New 90 Projects/generative_ai_project/GraphRAG changes the working directory to the specified folder in Google Colab.</p>

<h4>Importing Libraries for the Project</h4><p>The code imports fundamental libraries which are necessary for the project implementation. The program loads natural language processing libraries such as NLTK and spaCy alongside the machine learning tools sklearn and numpy and the graph processing capabilities of networkx. LangChain fuels the system through document loading and embedding operations and retrieval tasks and the application benefits from thread pooling features and environment variable management functionalities.</p>

<h4>DocumentProcessor Class</h4><p>The <strong>DocumentProcessor</strong> class serves as an interface for processing documents with embedding generation and document chunk similarity analysis.</p><p><strong>__init__:</strong> During initialization, DocumentProcessor receives RecursiveCharacterTextSplitter to divide documents into chunks while using an OpenAIEmbeddings instance to embed these chunks.</p><p><strong>process_documents:</strong> A document list is input into this method, which splits the documents into chunks while generating vectorized FAISS storage from embedded sections. When processing begins, it returns the separated documents together with the generated vector store.</p><p><strong>create_embeddings_batch:</strong> This method embeds multiple text items in separate batches through OpenAI's framework. The function generates arrays containing embeddings that correspond to the provided input texts.</p><p><strong>compute_similarity_matrix:</strong> This function computes cosine similarity between embedding vectors of input texts before returning a similarity matrix that depicts text-to-text relationships.</p>

<h4>KnowledgeGraph Class: Building and Analyzing a Knowledge Graph</h4><p>The KnowledgeGraph class creates and manages a graph of concepts derived from document content. It adds nodes for document splits, computes embeddings and extracts concepts using both <a href="https://spacy.io/" target="_blank">spaCy</a> and a large language model (LLM). The class also establishes connections (edges) between nodes based on the similarity of embeddings and shared concepts, with edge weights determined by both factors. This class helps structure and analyze information from a knowledge base, enabling efficient querying and exploration.</p>


<h4>QueryEngine Class: Handling Queries and Expanding Context</h4><p><strong>__init__:</strong> Initializes the QueryEngine with a vector store, knowledge graph, LLM and a chain to check answers.</p><p><strong>_create_answer_check_chain:</strong> Creates a chain that checks if the context provides a complete answer to the query.</p><p><strong>_check_answer:</strong> Checks if the given context fully answers the query, returning a boolean and the answer.</p><p><strong>_expand_context:</strong> Expands the context by traversing the knowledge graph and updating it with the most relevant nodes.</p><p><strong>query:</strong> Processes the query by retrieving relevant documents, expanding the context and generating a final answer.</p><p><strong>_retrieve_relevant_documents:</strong> Retrieves documents relevant to the query using a vector store and a contextual compression retriever.</p>

<h4>Visualizer Class: Graph Traversal Visualization</h4><p>The Visualizer class provides methods for visualizing the traversal of a knowledge graph. The visualize_traversal method highlights the traversal path, nodes and edges, with special markers for the start and end nodes and adds labels and colorbars for better clarity. It also displays a color gradient for edge weights and shows the filtered content of the nodes in the traversal path through print_filtered_content, printing the first 200 characters of each node's content.</p>

<h4>GraphRAG Class</h4><p>The GraphRAG class manages the entire workflow for document processing, knowledge graph construction, querying and visualization.</p><ul><li>_<strong>_init__:</strong> Initializes the GraphRAG system with necessary components like a large language model (LLM), embedding model, document processor, knowledge graph, query engine (set to None initially) and a visualizer.  </li><li><strong>process_documents:</strong> Processes a list of documents by splitting them into chunks, embedding the chunks and building a knowledge graph. It then initializes the query engine for handling future queries.  </li><li><strong>query:</strong> Handles queries by retrieving relevant information from the knowledge graph using the query engine. It also visualizes the traversal path through the knowledge graph and returns the final response to the query.</li></ul>

<h4>Defining the Document Location</h4><p>The variable path stores the file path to a PDF document named "Understanding_Climate_Change.pdf" located in the specified directory on Google Drive.</p>

<h4>Loading and Selecting Documents from PDF</h4><p>The code loads the "Understanding_Climate_Change.pdf" file using the PyPDFLoader class. It reads the content of the PDF and stores it in the documents variable. Then, it selects the first 10 documents (or chunks) from the loaded content for further processing or analysis.</p>

<h4>Initializing the GraphRAG System</h4><p>The graph_rag object is created as an instance of the <a href="https://microsoft.github.io/graphrag/" target="_blank">GraphRAG </a>class. This initializes all the components of the GraphRAG system, including the large language model (LLM), embedding model, document processor, knowledge graph, query engine and visualizer. The system is now ready to process documents and handle queries.</p>


<h4>Processing Documents with GraphRAG</h4><p>The process_documents method is called on the graph_rag object, passing the first 10 documents (documents) as input. This method splits the documents into chunks, embeds them and builds the knowledge graph using the DocumentProcessor and KnowledgeGraph classes. It also initializes the QueryEngine with the vector store and knowledge graph, preparing the system for querying.</p>

<h4>Querying the GraphRAG System</h4><p>The query method is called on the graph_rag object with the query "what is the main cause of climate change?". The method retrieves relevant information from the knowledge graph, visualizes the traversal path and generates a response based on the available data. The final answer is stored in the response variable.</p>

<h2>Conclusion</h2><p>The <strong>GraphRAG</strong> project successfully integrates advanced techniques in document retrieval, knowledge graph construction and query expansion to provide an efficient and accurate system for answering complex queries. By combining <strong>vector search</strong>, <strong>BM25</strong> and <strong>knowledge graph traversal</strong>, it ensures contextually relevant and precise responses. The system's ability to process, embed and organize document chunks into a knowledge graph enhances its retrieval capabilities. Additionally, the use of <strong>large language models</strong> ensures that incomplete answers are supplemented, while <strong>visualization</strong> tools provide transparency in the query resolution process. Overall, GraphRAG offers a robust solution for applications requiring deep document understanding, making it an effective tool for intelligent information retrieval.</p><h2>Challenges New Coders Might Face</h2><ul><li><p><strong>Challenge: Handling Large PDF Files</strong><br /><strong>Solution:</strong> Use batch processing, split documents into smaller chunks and utilize Colab's high-RAM runtime for better performance.  </p></li><li><p><strong>Challenge:  Slow Embedding Generation</strong><br /><strong>Solution:</strong> Batch processing of text embeddings in smaller chunks (using a batch_size) helps manage resources efficiently. Additionally, embeddings can be precomputed and stored in FAISS for faster access during query handling.  </p></li><li><p><strong>Challenge: Knowledge Graph Construction</strong><br /><strong>Solution:</strong>  Focus on <strong>cosine similarity</strong> for edge creation, which is computationally efficient and easy to implement. Using <strong>spaCy</strong> for named entity recognition and leveraging <strong>LLMs</strong> for concept extraction allows for better and more automated graph enrichment.  </p></li><li><p><strong>Challenge: Query Mismatch in Retrieval</strong><br /><strong>Solution:</strong> Use LLM-generated hypothetical documents to refine search queries and improve retrieval accuracy.  </p></li><li><p><strong>Challenge: Dependency Installation Issues</strong><br /><strong>Solution:</strong> Ensure Python 3.8+, install dependencies with !pip install --upgrade and use virtual environments for package management</p></li></ul><h2>FAQ</h2><p><strong>Question 1: What is GraphRAG?</strong><br /><strong>Answer:</strong> GraphRAG is an advanced retrieval-augmented generation (RAG) system that integrates knowledge graphs with large language models (LLMs) to enhance document retrieval and generation. By leveraging the relational structure of graphs, GraphRAG improves the accuracy and contextual relevance of AI-generated insights.</p><p><strong>Question 2: Is GraphRAG suitable for real-time applications?</strong><br /><strong>Answer:</strong> GraphRAG can be adapted for real-time applications, though it may require optimization to meet the performance demands of such environments. The system's modular design allows for adjustments to balance accuracy and response time, making it feasible for real-time use cases.</p><p><strong>Question 3: How does GraphRAG improve document retrieval?</strong><br /><strong>Answer:</strong> GraphRAG enhances document retrieval by utilizing the interconnectedness of data within knowledge graphs. This approach allows for more precise and contextually appropriate responses to complex queries, addressing the limitations of traditional vector-based retrieval methods.<br /><strong>Question 4. How does GraphRAG handle user queries?</strong><br /><strong>Answer</strong>: GraphRAG interprets user queries, retrieves relevant information from the knowledge graph and generates comprehensive responses using a large language model.</p><p><strong>Question 5. Where can I learn more about GraphRAG?</strong><br /><strong>Answer</strong>: For more detailed information and resources on GraphRAG, you can visit the official GraphRAG website. <a href="https://microsoft.github.io/graphrag/?utm_source=chatgpt.com" target="_blank">Microsoft GitHub</a></p><p>Additionally, the Neo4j blog offers an in-depth guide on GraphRAG patterns and implementations. <a href="https://neo4j.com/developer-blog/graphrag-field-guide-rag-patterns/?utm_source=chatgpt.com" target="_blank">Neo4j</a></p><p>These resources provide comprehensive insights into the concepts, applications and advancements related to GraphRAG.</p>