Identifying Structural Gaps in a Discourse

Structural gaps are the parts of the discourse that could be connected but are not (yet). Making connections between those parts will generate insight and potentially interesting ideas.

The rationale behind this approach is based on the theory of structural holes used in the social sciences: the parts of a network that could be connected but are not (yet). This is where there is the highest potential for innovation and new social capital. We apply the same analogy to text networks: what are the gaps in the discourse that could be bridged by the new ideas? 

InfraNodus has a set of tools that allows you to identify the structural gaps in a discourse and to formulate the questions you can ask to bridge those gaps and generate insight.

It also reveals the latent topical brokers: the nodes that bridge the existing communities and indicate interesting narrative turns.

Here is how you do this kind of analysis step by step:

1. Visualize a text as a network. In order to do it, go to the Apps and choose what data you have, e.g.:

• Your own text
• Your notes and ideas
• News of the day
• Google search results 
• Twitter search for a hashtag 

In our case will use the most popular tweets on coronavirus as of 4 May 2020 for this case study. The interactive graph is available on InfraNodus if you want to use it while you're reading this tutorial.

2. Click on the Analytics > Insight panel to see the information about the structural gap and the topical brokers.

3. In the Structural Gap field you will see the two topics that are not yet well connected but could be. Click "Reveal the Gap" button under to see the actual topics on the graph and the statements that contain the highest concentration of keywords that are found in those topics:

4. Ask yourself a question: What would be the relation between the first and the second topic? In our example, what's the relation between the "nursing homes" in the context of coronavirus and the  discussion about Trump? 

5. Use the information provided in the Statements to Link column (top left) to better understand the context of this possible new connection. In our case, we are talking, specifically, about the discussion on Twitter on the nursing homes in New York and other states that were forced to take in coronavirus patients even though they couldn't cope, on the one side, and the criticism of Trump's response on the other. 

Actionable insight: if we propose an idea that links those two statements together, we could make a contribution to this discourse that would be relevant in the context of the current discussion. For example, tweeting something like this  — "The government authorities have failed us on multiple levels: both the federal government and Trump who did not respond on time and the state governments that forced nursing homes to take #Coronavirus patients beyond their capacities."

6. We used the example of Tweets to demonstrate the approach, but the most interesting results will be produced on your own texts: something that you are writing or simply your notes and ideas. Perform the same sequence of actions as above to find interesting gaps and to formulate new research questions that will bridge those gaps to foster innovative thinking. 

For example, let's take as an example a research paper we wrote about the methodology used in InfraNodus, presented at the WWW Conference, available at the ACM library.

In order to do that we visualize the PDF file as a graph and get the following result:

7. You will see that there is a gap between the two topics. Let's use the context provided and form a research question. The first topic is on the feature of InfraNodus method that allows us to identify the structural properties of the discourse. The second topic is on the LDA functionality and how InfraNodus method could be used to enhance topic modeling done using LDA. 

Actionable insight: If we ask: "Is there a way we could enhance LDA using discourse structure analysis?" — we could explore an interesting direction in the research and development of InfraNodus as a tool that provides structural analysis of discourse to improve data obtained using LDA and other topic modeling tools. 

8. Another interesting feature is the Latent Topical Brokers field under the structural gap. It shows us the nodes / words that are not very frequent in the text but that link the important topical clusters together. In other words, these are the nodes that already bridged the existing structural gaps in the text. Discovering the context where these nodes are used (by clicking on them and finding the excerpts) will reveal some interesting ideas present inside the text. These topical brokers may also indicate points of narrative shifts: where one topic connects to another.

In the context of our research paper, we find the part which explains the actual algorithm used to encode data in InfraNodus. This part does not contain too many frequently mentioned words, but it connects to them and plays a very important part in this particular discourse as it provides the important technical details on the method.

9. Once you generate an idea, you can always reiterate by removing the nodes / words you already worked with to reveal latent topics behind:

Try it yourself:

Coronavirus Tweets graph (4 May 2020):
https://infranodus.com/support/coronatweets?background=default&most_influential=bc&maxnodes=150&currentapp=twitter&stopwords=coronavirus

InfraNodus Research Paper graph:

https://infranodus.com/support/infranoduspaper?background=default&most_influential=bc&maxnodes=150&panel=insight&stopwords=journal+infranodus