/literature-review

Notes about how to do literature review

MIT LicenseMIT

Literature Review

Notes about how to do literature review

Literature review is the first step in research. We conduct literature review to answer questions. It allows us to see further than others by standing upon the shoulders of giants. To conduct literature review efficiently, the following abilities are usually required:

  1. Search for relevant literature

  2. Filter out irrelevant literature by reading only title, abstract, introduction and conclusion

  3. Critical reading and thinking

  4. Innovate new design or methods

Workflow

  1. Use review paper to find 30-40 papers that are relevant (based on their title and abstract) to the topic you want to study. Here you need to find those most-cited methods (2A) and cluster them into groups (2B).

  2. If you can't tell their characteristics, strength and weakness (2C), then find those most-cited papers, and cluster them into the groups from 2B. Sort them by date, and read them one group at a time.

  3. By this far, you should have about 20 papers that are closely related to the topic. Read those groups of papers again, but this time, read main body until you can tell their assumption (under what conditions it works, 3A), and evaluate how hard/easy the assumption holds in reality (3B). what's the strength of the approach under the assumption. You also need to know what contributes to the strength (3C) in the equations.

  4. You may know the history of a group, its core assumption, supporting theory, and main achievement(s) by this point. Now you need to tell their main weakness.

  • Batch reading to optimize the objective function.

Presentation

  1. First slide: paper title, author, year

  2. One slide, one concept

  3. Show its main assumption, equations and applications

  4. Illustrate its potential

  5. Analyze/predict performance in various env

  6. Personal view point: creativity, main weakness and strength, when to use the ideas

Levels:

  • Find useful papers -> relevant papers -> closely related papers

  • Apply methods found -> extend methods found -> combine various methods to create better ones

  • Read papers in hands to have some ideas -> applicable ideas -> find all applicable ideas while limiting papers read

  • Pick papers to read, and tell their strength and weakness

Paper structure:

Abstract: problem, method, strength

Introduction: clustering of methods, strength and weakness

Main body: problem definition, assumption, induction

Results: quantify the performance of the proposed method

Conclusion