How to Write Literature Review with AI and Knowledge Graphs

You can use InfraNodus to write literature review for a specific domain or a collection of research papers. The advantage of using knowledge graphs and AI is that it helps identify the main themes, explore the relations between them, and — most importantly — discover the gaps and weak signals that can direct your attention to potentially interesting new areas of research.

In this tutorial, we will demonstrate various features of InfraNodus that you can use at each stage of the standard research literature review workflow:

1. Collecting and importing the data
2. Presenting the topic of research
3. Grouping and exploring existing studies by themes 
4. Critical analysis: revealing the content gaps and weak links  (what's missing?)
5. Conclusion (synthesis & transition)

We will also show how you can use AI and knowledge graphs to get additional insights at every step.

Step 1: Collecting and Importing the Data

First, you need to collect and import the data. The easiest way to do that is to create a spreadsheet with the titles of the papers, the abstracts, conclusions, and meta data you can later use for filtering (e.g. date and country of publication, journal, keywords, impact, etc). 

Once you have the spreadsheet created, you can import it in InfraNodus using the CSV import app: https://infranodus.com/import/csv 

In our example, we will use a collection of research papers on gut microbiome collected during 2025 by a research team in France. 

When we import the file, we first select the column we want to be analyzing — in our case, it's the "Conclusions" column, which has a human-written summary of the main findings from each research paper relevant for the team that collected the data. If the row has a URL link, you can add it to the analysis as well, so that later you can use the graph to click to the specific research paper you found using the graph.

At the second step, we select which columns we want to be using as tag filters later. These will be useful if we want to segment our insights by the country of  publication or the relevance of the journal:

Once the data is imported, InfraNodus will propose to generate a knowledge graph based on the data in the document. 

When you import the data, you can also select a special mode of analysis where InfraNodus performs automatic entity detection on your columns and extracts named entities that will be represented as nodes in the graph. This mode is recommended for building sparser graphs that you will be exploring manually:

Step 2: Presenting the Topic of Research

After the spreadsheet is imported, InfraNodus will visualize the data in the column you're analyzing as a graph. In our case, it's the research conclusions for each of the 300 papers in the collection.

As the first step, you might want to clean the graph and remove the words that appear too often. For instance, in the case above, we have papers on "gut" and "microbiota", so it makes sense to remove those words from the graph, as well as "microbiome", as those terms are mentioned in almost every conclusion and we're more interested in the context around. You can remove the terms by selecting them and then clicking the "trash" icon at the top right of the graph.

Once the graph is cleaned and the terms are removed, we can now see the main patterns and themes emerging in the graph. The topical clusters are identified using the algorithm outlined in the paper above and we use the AI to give names to those clusters based on the content of conclusions that belong to them:

Interestingly, we can now see that the majority of papers on gut microbiome selected for this literature review are focused on:

• microbial impact, 
• dietary strategies,
• metabolic health,
• nutritional quality
• fatty acids
• gut axis

Alternatively, if we used entity detection instead of the lemmas for analysis, we would get the following graph:

In this graph, we have similar insights but they are a bit more specific: 

• microbial metabolites, 
• inflammatory diseases,
• probiotic treatments,
• microbiota nutrition
• obesity management
• bacterial species
• mental health disorders

We can use this graph as a reference when writing an introduction to the topic of gut microbiome, pointing out that multiple studies focus on microbial metabolites and their impact on gut flora, relation of gut flora to inflammatory diseases, dietary strategies (with a focus on probiotics and obesity), metabolic health, nutritional quality, and relation of gut flora to neurological function, especially for children (anxiety, depression, alzheimer, etc).

While the topic names can be used as references, the interactive graph itself can be used to explore the relations between different concepts and patterns that emerge in most research papers analyzed.

We can also use the built-in AI to generate a summary for the whole graph by clicking the AI Summarize button:

You will see that AI picks up on the other important themes we will discover later: the relation of gut microbiome to cognitive function, Alzeimer, and strategies for neuroinflammation reduction.

Step 3: Grouping and Exploring Existing Studies by Themes

The next step is to explore each theme in detail and also discover some latent, less visible topical clusters that are important in this field of research. We can also group the existing studies by themes in order to be able to find relevant papers using the specific terms or topics we're interested in through the graph. 

We will show an example for one specific theme below, but you can run this analysis for every theme that's discovered in the graph. 

For instance, our thematic analysis above revealed that there's a cluster on the relation of gut flora to mental health disorders. Interestingly, it has only 3% of influence, which means that in this collection of research papers this topic is underrepresented in comparison to the studies on obesity and diabetes. Let's zoom into this topic to learn more about it:

We can click on this topic, and then InfraNodus will automatically filter the top papers that relate to this particular thematic cluster. We can scroll through these papers and find the specific studies that link gut microbiota to depression. The specific study we find shows that physical activity that induces changes in gut microbiota may help fight depression in women.

Tip #1: Finding a Needle in the Haystack

Consider how easy it is to use the graph to find a relevant research paper. We can simply select the topic of our interest (mental health), automatically filter the papers that talk about this topic, then open that paper  in another browser tab and explore it in more detail. 

Tip #2: How is the Influence of a Topic is Calculated?

Each topical cluster in InfraNodus has an influence measure. It is the relative betweenness centrality of all the nodes that belong to that cluster to the total influence of all nodes in the graph. The 3% value for the "mental health" cluster above indicates that the nodes that belong to that cluster are responsible only for 3% of the meaning circulation in this particular selection of research papers conclusions. We can also see in the statements panel that only 5% of the conclusions analyzed belong to that topical cluster. This is may be an indication of an underexplored theme and may be used for finding attractive research topics that may enrich the current body of research.

Saving Time: Using AI to Summarize the Themes

We can also use the AI to summarize the specific themes. In the example above, we can open the AI Insight panel and click the AI: Summarize Selected Topics button in the Analytics panel or by clicking the AI: Summarize Visible above the statements panel:

The InfraNodus AI will ingest both the content of the topical cluster and the related papers in order to summarize the conclusions of all the 15 papers filtered that belong to that particular subject. As you can see, the summary correctly synthesizes the conclusions from the papers indicating a close relation of mental disorders to anxiety and depression through gut microbiome with an increased risk for children and adolescents.  

Step 4: Critical Analysis: Revealing the Content Gaps and Weak Links

Knowledge graphs can be very useful for revealing the content gaps and weak links. One strong advantage of using graph theory metrics is that we can use the network to reveal the structural gaps between different themes in the discourse: topics that everyone is talking about but that are not so well connected. 

Using an analogy with social networks, these are the groups of concepts that could benefit from being connected because they are relevant to the field but they are not well-connected yet. This may happen, of course, due to the fact that there's no interesting relation between them, however, occasionally, those structural gaps can serve as the starting points for new explorations and interdisciplinary research. 

In order to reveal the content gap, we can go to the Analytics Menu > Content Gaps, then click Show Hightlight and Show Another Gap until we find the clusters at the opposite sides of the graph. This usually indicates that there are two topics that are not yet well connected and that could benefit from better integration:

For instance, in our case, it's the topic of "mental health disorders" and "microbiota nutrition" — for instance, what nutrition strategies can alleviate depression and anxiety. 

We can use the graph itself to generate interesting research questions or ideas that bridge this gap or use the built-in AI to generate the research question for us: click the AI: Question button and InfraNodus will generate a research question to link those clusters together — 

As you can see, the question that AI generated proposes us to think of:

"How can a diet rich in prebiotics and fermented fibers, targeting gut microbiota fermentation processes, act as both a preventive measure and therapeutic intervention to manage obesity-related depression symptoms in adolescents through modulation of the gut-brain axis?"

This is a very interesting topic for research that can bridge the existing gap in literature: linking the group papers that identify the relation between the gut and depression to specific nutrition strategies (rich in prebiotics and fermented fibers) that can help alleviate the symptoms and affect the gut microbiome in a positive way. 

Additionally, we can also use the graph to select the concepts at the periphery and explore the shortest paths between them. This exploration can either be done manually using the graph or with the help of AI that can generate interesting research questions that link those concepts in new ways:

In this case, we select "inflammatory", "probiotics" and "biomarkers" and AI generates a research question that links those concepts together:

Can specific probiotic strains modulate inflammatory biomarkers in diverse gut microbiota profiles to enhance therapeutic outcomes for metabolic diseases like T2DM or obesity?

This sounds like a very specific and interesting research topic that we could explore further. We can use the AI chat to ask if any paper in our collection looks into this topic already and if there is none, we could perform wider search in scientific databases to find if there's any research in this topic.

5. Conclusion: Synthesis and Transition

In the last step of analysis, we can combine all the different insights obtained in steps 2 through 4, revealing the main themes, the current shortcomings, and potential new areas for research. 

Bonus: Exploring The General Public's Knowledge of Your Topic of Study

If you are writing research you might want it to reach wider audience. Most of the people find their content online using search engines and LLMs. InfraNodus can be used to analyze typical search queries for specific topics in order to get the picture of what people are thinking about when they search for a certain topic. 

For instance, in our case, we can use the Search Intent analysis app in InfraNodus to analyze search intent for "gut microbiome" — https://infranodus.com/import/keywords — then when the graph is generated we will see some interesting patterns emerging:

Most of the people are interested in "health supplements" when they search for "gut microbiome". Specifically, they want to know what supplements they can use to lose weight and to avoid bloating. 

While this correlates with some of the topics we discovered in research papers analyzed, we can also see that most people are not aware of the more serious links of the gut microbiome to obesity, diabetes, depression, anxiety, and Alzheimer. This could be an interesting pathway for us to explore. On the one side, we could get more people interested in gut microbiome by addressing their interest (bloating) and explaining how it can be a symptom / precursor of other, more serious illnesses that can happen later in lie. On the other side, we can also see that there's no search intent on mental disorders and its link to gut, which means that general public could really benefit from more research and better visibility of this subject

FAQ

Here we list some of the common questions people have when running literature review analysis using knowledge graphs. If your question is not covered here, please, contact us or leave a comment below.

What is the Advantage of Using Knowledge Graphs to Standard AI Analysis?

The advantage of using this approach to running a summary in a standard LLM is that InfraNodus uses peer-reviewed algorithm for topical clustering, so you are 100% sure that there will be no hallucinations and that the results you get are reproducible and won't change depending on the LLM model or version you use. 

Knowledge graphs also let you explore weak signals and gaps, something that LLMs are not so good at because they are designed as prediction machines to provide the most likely next token, not the most interesting research idea.

Which method is used to analyze and represent the data?

By default, the words are converted to lemmas, auxiliary stopwords are removed (you can customize this step), when the words appear in the same statement they will be connected (a sliding window of 4-grams is used), the closer the words, the stronger is the weight of connection. Based on this representation, a text network is built, Force Atlas and modularity algorithms are applied for community detection and the nodes (lemmas) are ranked by betweenness centrality. This approach helps identify the groups of concepts that frequently co-occur together in the same context and identify the concepts that serve as the crossroads for meaning linking distinct clusters together. For the exact description of the method used for analysis, please, refer to our peer-reviewed paper on the ACM portal: InfraNodus: Generating Insight Using Text Network Analysis (free access). 

Should I Choose Words (Lemmas) or Detected Entities for Analysis?

At the first stage, when we select the column to analyze, in the advanced settings, we can choose whether we want to use single words or automatic entity detection as the nodes in the graph. 

Below we will demonstrate the results for both, so you can see the different affordances each type of analysis provides. 

Personally, we prefer to use single words to avoid adding additional ontology as an interpretative layer. The graph of the concepts is useful enough to reveal the main themes and also tends to be more granular and precise, especially for AI insights. However, in some instances, especially in the contexts where technical language is used and where you plan to rely on manual graph analysis and exploration, you might want to turn on entity detection in order to have a sparser graph with less granularity and higher-level concepts more visible.

How to Run this Analysis with My Data?

To try this with your collection of documents, go to https://infranodus.com 

If you feel uncomfortable using the graphs, you can run similar analysis using our MCP server in any LLM: ChatGPT, Claude, or an IDE of your choice via https://infranodus.com/mcp — it has access to all the workflows described above, so you can simply ask your LLM (after connecting the MCP server) to run content gap analysis or reveal the main topics and our MCP server will do it for you.