Finding related work can take forever if you do not understand the trick. You can keep on googling with several keywords and you will find a thousand entries. This is not efficient. There is a better way. I called it "snowballing and backtracking" technique:
First, find out which papers are the current state-of-art in your area. If you are unsure whether certain papers are good, ask your supervisor.
Second, take note (1) what are the related work they commonly cited (snowballing), and (2) what are the related work that commonly cites them (backtracking). Most digital library (e.g, ACM, IEEE) allows you to quickly find those two.
Third, go inside each (1) and (2), and read their related work, and repeat the second steps around two to three times. Through this process, slowly build a network of what is important in your field and how most experts relate their own work with these related work. In most of the times, you will find around ~1-5 key papers, while the rest 10+ are more like supporting papers. By now, you should a clear idea what should be written in your Related Work section.
This technique is borrowed from social computing, i.e., by slowing propagating from one paper to another paper, you slowly get a clear sense of the entire network related to the original node. It is a simple technique commonly used and will save you a lot of confusion and time.
One more tip is to also learn to take note the common authors appearing in these papers. They are the “experts” or “gurus” in the field, so they know what is important. So it is good to also look up their website, their publication list, and see if anything matches your work.
Another point about reading papers is to first start with top-tier venues (e.g., NIPS, ICML, ACL). In this way, you will learn from the best, as well as the "culture" of that community, what they know and they do not. Knowing your audience is perhaps one most important step for your successful publication.