Using Knowledge Graphs to create dynamic learning paths for neurodiverse learning using the semantic tree concept - part two

This post continues our work from the previous post on using knowledge graphs to create dynamic learning paths for neurodiverse learners

To recap to our objective is to use Knowledge Graphs to create dynamic learning paths for neurodiverse learning using the semantic tree concept

The semantic tree concept as popularised by Elon Musk 

So, we see it as

Given a set of documents

We find the nodes, the edges and the weights using multiple perspectives

We then create the semantic tree as

• trunk
• branches
• leaves

We chose Fast fashion for this task to learn

We then identified 

1. primary themes (trunk)

2. sub themes (branches)

3. list of bullet points (up to 5 each for every sub theme ie leaves)

We used two perspectives (personas/ experts)

And got weights for these perspectives

We summarised weights in a table for the themes, sub themes and bullet points 

Next steps:

1. We need to extend this to mixture of experts and determining weights for knowledge graphs. Perhaps we should combine weights across perspectives?
2. We also need to upload and visualise the graph in Neo4j
3. We need to co-relate this work against the literature review for learning challenges for neurodiverse people
4. We chose two perspectives for fast fashion. We then can show how a variety of personas can be used to view the same topic in different ways. The first perspective considers the environmental and ethical implications of fast fashion, focusing on sustainability, worker exploitation, and long-term environmental impact. The second perspective is that of a young, fashion-conscious, cost-conscious consumer who is not environmentally conscious. 
5. We also created question templates. Thus, a possible workflow could be that the learner is engaged in actually creating the knowledge graph as an artefact.Or the leaner engages with multiple perspectives based on a variety of personas. Or the learner learns from question templates
6. Relate back to the pedagogy where we started with the Inverse bloom method of learning wit Dr. Yogita Khedkar
7. Computing weights is currently by prompts. We could automate it much more - through the OpenAI API but also by considering complex mechanisms. Also, the relationships in the document itself do not suffice to calculate weights. While, the LLM can detect strength of relationship between concepts but may have to assign weights in terms of context.

the full chat is HERE

we will present these ideas in the The Oxford Artificial Intelligence Summit: Towards AGI By Ajit Jaokar Anjali Jain Dr Kaouter Karboub Aditya Jaokar Dr. Yogita Khedkar

(full table of weights in chat)

with David Stevens

Image source:

https://pixabay.com/illustrations/ai-generated-pastoral-flowers-8690509/