research process - How do we know if something relevant is already published?

I keep reading European Conference on Computer Vision, International Conference on Pattern Recognition, and others, and I find so much in common in those papers.

For example, half the papers seem like: X Algorithm combined with Y algorithm tested on Z database and was better than A in this category but worse than A in that category.

There are many top journals and top conferences with > 1000 papers, in a year, in this field. In addition to this we would have to go through all the tier 2, tier 3 conferences for lack of a similar piece of work.

While working on organic solar cells I found the same problem. I saw 20 different papers on extracting parameters from a single-diode/double-diode/XYZ model.

So how do we know if something relevant is already published?

Answer

That title summarises a large fraction of all research ;-)

You need to spend a significant amount of time browsing through scientific literature. This illustration from PhD comics illustrates the problem:

My strategy is as follows:

1. 
   

   I start with an arbitrary paper with a title that seems relevant. Let's call this paper A.

   
   



2. 
   

   Using an online service such as Scopus (payment or subscription required), I look for:

   
   
   
   
   1. 
      

      All references, e.g. papers cited by A.

      
      
   
   
   2. 
      

      All citations, e.g. papers that cite A.

      
      
   
   
   
   
   


3. 
   

   Both the references and citations can be sorted by the number of citatons, so you get the most highly cited papers first. Select the papers that seem relevant (at this stage, reading the abstract is usually good enough).

   
   


4. 
   

   For all relevant papers left in step 3, repeat step 2. You should find considerable overlap already. By now, you should have already found the important papers as well as one or more review articles.

   
   


5. 
   

   Iterate steps 2–4 until your hard disk is full ;-)

   
   

Of course, if you iterate this process indefinitely, you will soon have downloaded all scientific research ever published. But by being selective but not too selective, this process should lead you to any relevant publications. Even if nobody has cited the relevant publications, the relevant publications themselves should cite other publications, so via step 2.1 you will still find it.

The only caveat here is that databases like Scopus are incomplete, probably in particular when it comes to conference proceedings (in my field, those are rarely relevant). Maybe other databases work better in your field. Hopefully, there is a scientific database that is reasonably complete for relevant publications in your area of research!