There is no general answer possible because it always depend on the data and how they in the end should be displayed. Furthermore quite often you will need to combine all of these approaches (and some more) to build a suitable datamodel.
If you have already read all this and you understand each of these techniques, you should share with us your tables structures (sample data) and the exptected output, MAYBE you will have to choose between two techniques and maybe only one technique will be suitable for you problem.
I would let you know based on my exp and approach :
APPLYMAP : IF i need to use the MAPPING of that columns across many areas in Dashboard rather than JOIN which helps at that instance only .
JOIN : This is mostly used when you have fields in common and you need to reduce the data set of 1 table based on the values of the fields in other table.
KEEP : If i have DIMENSION tables and have a key to associate to FACT Table, then using the KEEP helps to have only keys of the field in DIMENSION Table that exists in the FACT table , ( Also this gives better performance than JOIN and also if you need to use the DIM table in other areas in Model , by using KEEP we can have the table in Model whereas in JOIN when we join DIM to FACT , DIM would be dropped as result is only 1 table).
Hope this helps, please let me know if any help is needed.
Most important will be to load only those data which are really needed within the QlikView environment and then much more reduced within for certaines purposes specialized applications. Quite probably the number of tables will now decrease and even more significantely the number of fields and records.
Keep it simple and starts small until you understand the data and how Qlik worked especially the differences to the SQL logics. SQL means often that the data are normalized which is the opposite from the Qlik logic (although it would mostly work). This meant your various dimension-tables could probably just be joined together. But by the fact-tables it depends like above mentioned ...