How InfraNodus Works: AI-Enhanced Text Network Analysis

InfraNodus uses a combination of text mining, network analysis, data visualization, NLP, and artificial intelligence to provide insights about any discourse and enhance a research process. Network analysis is the secret sauce that sets InfraNodus apart from other text analysis and AI-based tools. Thanks to this approach, we can design highly relevant prompts for AI systems and use the most advanced language models to derive meaning from text and see how it can be developed further.

Another advantage of InfraNodus is its interactive data visualization, which enables the users to "touch" the texts and explore any narrative in a non-linear way, remove high-level ideas, reveal underlying concepts, and explore the periphery of any discourse. You can zoom in on any group of concepts and see the context where they are used.

InfraNodus text visualization example that uses AI and network analysis to provide insights

Finally, InfraNodus is the first tool that uses GPT-3 AI not only for text generation but also for text analysis. We strongly believe that AI systems are not there to replace humans, but to, rather, enhance our existing processes, increasing their efficiency and making them more scalable. That's why the use of GPT-3 AI in InfraNodus is geared towards helping derive insight through an iterative and interactive research process.

Step 1: Text Network Visualization

InfraNodus represents a text as a network, where the single concepts (or entities) are the nodes and their co-occurrences are the connections between them. Powerful algorithms from network science and graph theory are then applied to this text network to highlight the most influential concepts, topical clusters, and structural gaps between them. Advanced NLP AI algorithms are used to help users interpret the graph, derive additional insights from their data, and develop their ideas further.

To learn more about the algorithms used in InfraNodus and for citations, please, refer to this peer-reviewed paper: Paranyushkin, D (2019). InfraNodus: Generating Insight Using Text Network Analysis, Proceedings of WWW'19 The Web Conference, (ACM library, PDF) as well as the first paper we published on the subject: Paranyushkin, D (2011). Identifying the pathways for meaning circulation using text network analysis, Nodus Labs. (Google scholar), and other Nodus Labs publications.

Step 2: Revealing High-Level Ideas

After you add or import a text, it is transformmed into a sequence of lemmas, the stopwords are removed, and each lemma is represented as a node. The nodes are connected if they appear in a 4-gram window. The closer they are, the higher is the weight of the connection.

Based on this logic, we build a network graph where the nodes are ranked by the measure of betweenness centrality (BC). The higher the measure of BC, the bigger is the node on the graph.

The nodes that appear in the same context more often are aligned into clusters on a two-dimensional plane and have the same color. We use community detection algorithm based on the modularity measure to cluster the nodes into topical groups. These are also shown in the Analytics panel.

Text network visualization showing the most influencial nodes based on betweenness centrality and community detection

At this stage, you can see an overview of any discourse: the most important concepts, topics, and pathways for meaning circulation, helping you gain a better undertanding of any text and deconstruct its dispositif (to use the term coined by Michel Foucault), revealing its rhizomatic structure (Deleuze).

Step 3: Structural Gap Detection

Once we visualize the structure of the graph, we can quickly see what is missing. InfraNodus will then apply a special algorithm to detect the structural gaps: the parts of the discourse that could be better connected. These are identified from finding the topical clusters on the graph that are distinct from one another and have a high distance. We then highlight those gaps, bringing the user's attention to the blind spots in the discourse.

Structural gap detected in a network

While the structural gap is highlighted in the graph, we also reveal the two topical clusters that could be better connected. The users are then encouraged to think of a possible connection to bridge the structural gap and generate a new research question or an idea. Alternatively, you can also use the built-in GPT-3 AI to generate a research question or an idea for you.

Step 4: Underlying Ideas, Entrance Points, and Periphery

Most of the text analysis tools are too focused on the main concepts, topics, and entities found in text. InfraNodus, on the other hand, is a great tool for discovering the ideas that are not easily visible on the surface.

Part of the workflow that we propose is removing the most obvious concepts from the graph to see underlying ideas around them. This helps reveal hidden low-level ideas that make the discourse specific and different from the rest.

Another interesting feature is the detection of discourse entrance points: the concepts that have high influence per occurrence. These are the ideas that can provide an easy access to discourse as they function as pathways for meaning circulation but, unlike the most influential nodes, are not burdered by "meaning traffic".

Finally, network visualization also helps users reveal the periphery: these are the groups of ideas that are not central to the text itself but play an important role in connecting it to other discourses.

Revealing low-level ideas in discourse

More Information

To learn more about the approach behind InfraNodus, please, read about Ecological Thinking, Cognitive Variability, and see how you can benefit it in your research in the Research Framework article, which proposes a concrete workflow you can use in InfraNodus to embrace this heuristic.

The basic workflow outline is available on our support portal and directly in InfraNodus' interface


Try It Yourself

If you would like to try this approach, log in or sign up for InfraNodus below:

Sign Up     Log In