The principle when choosing research directions

When you are not sure about whether a research direction is good or not, you can check the quality of the related works.

(1) Whether the related works have solid definition of the problem?

(2) Whether the related works have clear motivation? (Whether useful in downstream tasks?)

(3) Whether the benchmarks (datasets, metrics) used in related works are open-source and clearly-established?

(4) Whether the related works have comprehensive and solid evaluation?