This tutorial shows how the active learning software and algorithms can be tested. Because it is not possible to test the software by reading everything yourself. Therefore, ASReview implements a mode in which the relevant articles are displayed in green. This make decision making straightforward.
This tutorial assumes you have already installed Python and ASReview. If this is not the case, please check out the installation page.
Create a project¶
Create a project and give the project a name in step 1. The name is not relevant, but is adviced to have a test-prefix.
Upload a dataset¶
Select one of the three test-datasets available by default. The datasets available are PTSD, Hall and AceInhibitors. See a description of the datasets below.
- The PTSD data containing the results of a systematic search for longitudinal studies that applied unsupervised machine learning techniques on longitudinal data of self-reported symptoms of posttraumatic stress assessed after trauma exposure (https://doi.org/10.1080/00273171.2017.1412293). The total number of papers found was 5,782 of which only 38 were included (0.66%);
- Results for a systematic review by Hall et al. of studies on fault prediction in software engineering (10.1109/TSE.2011.103 ) with 8,911 papers of which 104 inclusions (1.17%);
- Results for a systematic review on the efficacy of Angiotensin-converting enzyme (ACE) inhibitors, from a study collecting various systematic review datasets from the medical sciences (https://doi.org/10.1197/jamia.M1929) with 2,544 papers of which 41 inclusions (1.61%).
In the next step, Step 3, you are asked to add prior inclusions. Select 2 papers of your dataset of choice and copy-paste the title in the search bar.
- Latent trajectories of trauma symptoms and resilience: the 3-year longitudinal prospective USPER study of Danish veterans deployed in Afghanistan
- A Latent Growth Mixture Modeling Approach to PTSD Symptoms in Rape Victims
- Peace and War: Trajectories of Posttraumatic Stress Disorder Symptoms Before, During, and After Military Deployment in Afghanistan
- The relationship between course of PTSD symptoms in deployed U.S. Marines and degree of combat exposure
- Trajectories of trauma symptoms and resilience in deployed US military service members: Prospective cohort study
- Predicting Defect-Prone Software Modules at Different Logical Levels
- Quantitative analysis of faults and failures in a complex software system
- A Comprehensive Empirical Study of Count Models for Software Fault Prediction
- Predicting fault prone modules by the Dempster-Shafer belief networks
- Robust prediction of fault-proneness by random forests
- Quinapril in patients with congestive heart failure: controlled trial versus captopril.
- Clinical effects of early angiotensin-converting enzyme inhibitor treatment for acute myocardial infarction are similar in the presence and absence of aspirin: systematic overview of individual data from 96,712 randomized patients. Angiotensin-converting Enzyme Inhibitor Myocardial Infarction Collaborative Group.
- Efficacy of different drug classes used to initiate antihypertensive treatment in black subjects: results of a randomized trial in Johannesburg, South Africa.
- Long-term mortality in patients with myocardial infarction: impact of early treatment with captopril for 4 weeks.
- Comparison of perindopril versus captopril for treatment of acute myocardial infarction.
Mark all five papers in Step 4 as irrelevant.
Start reviewing the first 50, 100 or even 200 papers. Abstracts in red are relevenant papers and abstracts in black are irrelevant. This is based on a fully labeled dataset. Hint: open the statistics panel on the top right.
For the PTSD dataset we expect you to find about 7 out of 38 relevant papers (in red) after screening 50 papers, 19 after screening 100 papers and 36 after 200 papers.
For the Hall dataset we expect you to find 25 out of 104 relevant papers (in red) after screening 50 papers, 48 after screening 100 papers and 88 after 200 papers.
For the ACE dataset we expect you to find 16 out of 41 relevant papers (in red) after screening 50 papers, 27 after screening 100 papers and 32 after 200 papers.
Export the results (in the menu; top left) and open the file in excel. The papers are now presented starting with your inclusions – unseen papers papers ordered from most to least relevant according to the last iteration of the software – your exclusions.
For all three datasets, the animated plots below show how fast you can find the relevant papers by using ASReview LAB compared to random screening papers one by one. These animated plots are all based on a single trial per dataset in which only one paper was added as relevant and one as irrelevant.
PTSD (38 inclusions out of 5,782 papers)
Hall (104 inclusions out of 8,911 papers)
ACE (41 inclusions out of 2,544 papers)