UPDATE_pos vs UPDATE_all - Githubissues

ai-se / ML-assisted-SLR

Automated Systematic Literature Review

2 stars 2 forks source link

UPDATE_pos vs UPDATE_all #51

Open azhe825 opened 7 years ago

azhe825 commented 7 years ago

What

In UPDATE, we import previous knowledge to boost current review. Two options for the type of knowledge being imported:

UPDATE_all: Every labeled examples
UPDATE_pos: Only labeled examples that are "positive" ("relevant").

Why

UPDATE_pos saves more on storage and communications.

How

Presumptive non-relevant examples made it happen to train a learner with only "positive" ("relevant") examples in imbalanced data sets.

Result

Hall data set:

Wahono data set:

Danijel data set:

Conclusion

UPDATE_pos performs quite similarly with UPDATE_all. It is reasonable to use UPDATE_pos and save a lot in storage and communications.

timm commented 7 years ago

UPDATE_pos!