Dynamically create tables vs key-value which is best?

[Cedric Le Varlet]

I am currently working on my thesis for my last year of Bachelor and am running into design issues.

Let me explain the problem:

Users will be expected to input datasets which format is not standardized. The datasets come from different tools and will not always present the same data as the datasets represent tests such as AB testing, Email/Social network conversion data, Google analytics ect. Furthermore some of the tools change their exported dataset formats from one version to another and therefore cannot be directly ported into a simple table.

I am therefore left with three options:

1. Migrate to a Graph Database such as neo4j or a key-value database
2. Allow the application to dynamically create MySQL tables and relationships and map those table names in a separate table for reference
3. Create a key-value table which effectively will store ALL the datasets in one table and use a template table to map the structure of the dataset.

The issue is compounded by the fact that the app which will interact with the database will need to be able to make sense of the data provided. Whilst it is possible to generalize some test cases this would also require me to keep a separate table simply to "tag" the individual fields to give them meaning.

My question is therefore the following:

Which would you recommend and why?