🎥 Text Mining and Topic Modeling with Network Analysis and Visualization

InfraNodus can represent the text as a network and use powerful graph analysis algorithms to identify and visualize the main keywords, topics, and their relations.

Online Tutorial (4 minutes):

Step 1: Getting a General Overview

1. Find a text you would like to analyze (e.g. a research paper, your writing, notes, customer feedback, Tweets or news). You can also use any interconnected data or sequence (e.g. RNA, music notes, etc.)

2. Open Infranodus Apps and go to Add a new text app

3. Copy and paste your text and click Save. 

4. Once the text is visualized, the words are represented as the nodes on the graph and their co-occurrences are the edges between them. These co-occurrences are calculated on the basis of 4-grams where the words that are next to each other have the weight of 3, the words separated by 1 word have the weight of 2, the words separated by 2 words have the weight of 1.

5. You can use the graph to analyze the discourse qualitatively. The bigger nodes (words) are the ones that have a higher influence (betweenness centrality) in the text network. The nodes (words) that are closer to each other and have the same color belong to the same network community (topical cluster). This way you can see the most important terms in your texts as well as the semantic groups that they form together. 

6. Open the Analytics panel at the bottom right to see the most influential terms (based on betweenness centrality) and the most important topics (network communities) in the text.

If you're interested to learn more about the science behind InfraNodus approach, please, read this peer-reviewed paper InfraNodus: Generating Insight Using Text Network Analysis or check our network science knowledgebase.

Step 2: Revealing the Meaning Brokers and Structural Gaps

7. Remove some more obvious words from the graph to reveal the context behind them. You can do that by selecting those words on the graph or in the analytics panel and clicking the "Hide" button (here's how). All the graph's metrics are recalculated, so the new topical clusters will come up.

8. See the network structure statistics in the analytics panel. This may help you see how connected / disrupted your text is, which has implications for the narrative.

9. Click the Insight tab in analytics to find the structural gap: parts of the text that could be connected but are not yet — that is the place where the new innovative ideas may be.

10. Also take a look at the latent meaning brokers: the nodes that have a high proportion of influence to frequency. This may be important terms that can be missed from the first sight that connected the important parts of the discourse together.

11. Open the Stats panel to see the full list of the most influential words and the terms that 
12. Open the Relationships panel to see the main bigrams and to explore relations between the concepts (by clicking the nodes in the graph)

13. Open the Sentiment panel in Analytics to analyze the text's sentiment. We recommend to use the BERT AI model as it works with multiple languages and is more advanced. It takes longer to load though.