Unlocking Entities and Relations: Creating Knowledge Graphs with AI

GraphRAG is something which is picking up recently, in this article we will try to get to the basics of GraphRag, Knowledge Graphs and go through steps to extract few basic entities & relations graphs using LLMs. Building meaningful knowledge graphs from textual data is a powerful way to reveal hidden connections and relationships. Knowledge graphs let us visualize entities (people, places, organizations) and the links between them leading to deeper insights. At high level it consists of Nodes and Edges, where nodes represent the entities and edges represent relationships. Once we have Knowledge graph in place, we can further use it for GraphRag. Let’s understand more about GraphRag and possible use cases and challenges. ?

Here is an example visual of all words used and relationships in a podcast. For certain questions doing a GraphRag based search could yield better responses.

GraphRAG is a smarter way to find information because it doesn’t just rely on matching words, it understands how things are connected. Unlike normal RAG which searches based on text similarity. ?Instead of looking at isolated facts, it explores entire networks, helping to uncover hidden patterns and deeper insights. This ability to connect related information efficiently makes GraphRAG ideal for solving complex problems where relationships matter. Whether tracking interactions, predicting patterns, or retrieving linked data, GraphRAG ensures meaningful results. ?

[Use Cases]?

[1] Finding the Right Medicines Faster -?Connect diseases, medicines, and case facts. Can help practitioners find better treatments.?

[2] Understanding Social Networks - People with similar interests form groups online (like on Instagram or Twitter). GraphRAG helps find connections between people to suggest friends or trending topics.?

[3] Making Better Plans - Some tasks must happen in a specific order (like cooking or building a robot). Can help figure out what step comes next in a process.?

[4] Finding the Right Products Online - Online stores like Amazon use GraphRAG to recommend products based on categories. It helps shoppers find what they need easily.?

[5] Managing City Traffic - Cities can use this to study traffic flow and predict jams. This helps reduce waiting time and improve roads.?

[Challenges]?

[1] Different Types of Data?- Normal RAG works with text in a simple way, like a list of words or chunks of text. But GraphRAG has to handle many types of information, like graphs, images, and molecules. Some data comes in triples (like subject, action, object) and some in network connections (like social media links). This means one single method won’t work, different tools required to search and understand each type of data.?

[2] Information in Graphs is Linked - Normal RAG stores information in small, independent pieces (like sentences or paragraphs in a document). But in GraphRAG, everything is connected, a single piece of information depends on many others.?

[3] Graphs Work Differently - One of the problems is that graph data changes depending on the field. Unlike normal RAG, which can easily transfer knowledge from one area to another, GraphRAG is specialized and domain dependent.?

Now let’s see a practical example and see how we can use LLMs to extract Knowledge Graphs. We will not go into detail implementations of GraphRag, for now just creating Knowledge Graph and visualization of entities and relations we will try to cover.??

[1] Perform installs and add required imports to use?

[2] Initialize a Gemini 1.5 Pro model instance using ChatGoogleGenerativeAI. Temperature is set to 0.0 to get deterministic responses.? ?

[3] Define Prompt for Knowledge Triple Extraction?- For Subject, Predicate, Object format.

[4] Define functions to extract triples and then function to parse extracted triples?

[5] Pass the text and perform the chat API calls using Gemini Pro model and store the subject, object predicate details in a Dataframe?

[6] Observe – Subject, Object, predicate details in a Dataframe?

[7] Write a code to visualize the data with Networkx & Pyvis libraries?

[8] Analyze the Knowledge Graph and see the relations. ?

Looking at this we can’t not deny that this approach provides different levels of insights than normal text similarity will do. ?

Summary?

We covered the basics of GraphRAG a way of retrieving information that goes beyond basic text matching. Instead of just looking for similar words like traditional RAG, GraphRAG understands how things are connected. This makes it smarter and more efficient. The foundation of GraphRAG lies in Knowledge Graphs, which organize data into nodes?and edges. By structuring information this way GraphRAG uncovers hidden patterns and deeper insights. Some good example we saw like in healthcare, it can help doctors and practitioners link diseases, medicines and case facts - leading to better treatments. In social networks, it can identify meaningful connections between people, making recommendations more personalized.?

Using GraphRAG comes with its own set of challenges. Unlike simple text-based searches, GraphRAG has to work with different data types like graphs, networks, and even molecular structures. Since information is deeply interconnected, extracting the right data efficiently is more complex.?