Literature review

Field overview

To do an overview of a field, try to locate two types of papers: (i) the seminal papers that everyone cites and (ii) recent review papers written by the experts in the field. From the former, we can identify important papers that cite the seminal papers. From the latter, we can get a high-level view of the field and identify key papers to follow next. This “sandwiching” makes use of both directions of citations.

Finding relevant papers

Usually the first step is to search on Google scholar with several keyword combinations to identify keywords that points to the area that I want, unless I have a specific paper that I already know.

Then go through the search results to identify relevant papers. Open them in new tabs. Use the venue & number of citations as useful signals. If it is published in highly regarded venues or cited a lot, it is more likely to be well-written and important (not always!). Scheme through the title and abstract to determine the relevance of the papers.

Once I identify some highly relevant papers, I can explore the past and future of each by going through the references (past) and the papers that cite the focal paper (future).

Probably the most critical thing to remember is that you can’t read every relevant paper and thus we should balance the activity of casting the net widely and catching actual fish. Most papers will not need a careful study because of their low-quality or relative irrelevance. Some papers will require a quick scheming to figure out what they are about. Few papers will need a lot of attention to details because they are highly relevant.

Processing relevant papers

Common ways

Once we identify some relevant papers, we need to process & file them. Otherwise, the knowledge we learn will get completely lost.

The most brute-force method will be printing out the papers. Although there’s some benefits of having physical copies, this doesn’t really scale and it becomes cumbersome quickly.

A common way is to use tools like Zotero to save the paper. Although this digital method allows you to create a searchable database, I think it is still not the best way because (i) the incentive to process and digest papers is weak and (ii) it does not create connections between papers, concepts, and topics.

I believe that creating connections and facilitating digestion is critical. This is the reason why I use Wiki (or whatever you call it, Ideas of connected ideas).

How to use Wiki/Obisidian

Using Obsidian (i.e., YYiki), I make pages for relevant papers in the following format: Paper/Lastname2020titlefirstword. This format doesn’t have much ambiguity, which facilitates the discoverability, and both succinct and informative enough.

Each paper page has a link to the paper, list of authors (and sometimes the link to the author’s local wiki page), links to other relevant pages, and notes.

I try to write a quick take-away (ideally using my own words) in the paper page, as well as creating links from/to relevant pages so that the paper is linked and discoverable and connected.

Tools