Frequently Asked Questions¶
How to check which version of ASReview I have installed?¶
Open asreview LAB and click on the left menu. The version number is displayed on top. You can also check the version of ASReview by running the following in your command line:
ASReview LAB & systematic reviewing¶
What parts of my Systematic Review project does ASReview support?¶
ASReview LAB is designed to optimize the Screening phase of your systematic review, e.g. screening titles and abstracts of a large number of publications. For the Identification phase, we recommend you to use your current databases and reference managers,or to distribute the work among your colleagues. You can import your records into ASReview, screen them, and import the results back into your own tools.
Can we use ASReview LAB for things other than Systematic Reviews?¶
ASReview LAB was originally designed for screening records with a title and an abstract. However, we try to develop a tool that is not exclusively fit for screening titles and abstracts in systematic reviews. In fact, ASReview LAB is intended to be applicable to all cases where a small number of relevant “textual items” have to be selected from an enormous pile of texts, such as patents, jurisdiction, historical newspapers, company reports, or keeping track of relevant research in an information overload environment.
Can we use ASReview LAB as a stand-alone screener?¶
(e.g. put in the prior information and let ELAS select the relevant records automatically)
No. We believe that machine learning algorithms are not good enough yet to replace human reviewers completely in deciding which records are relevant. Classification techniques are simply not good enough for this purpose. In systematic reviews, all relevant publications should be seen by the researcher. We refer to this as Researcher-in-the-loop as described in our paper introducing ASReview https://arxiv.org/abs/2006.12166.
Has the use of ASReview in systematic reviews been validated?¶
See our preprint for details on the validation of our software. Our team has been working on a paper that is currently under peer review. ASReview can lead to far more efficient reviewing than manual reviewing, while exhibiting adequate quality. Also ASReview can be used in combination with traditional approaches to systematic reviewing.
Has ASReview been used in any peer-reviewed article?¶
Multiple researchers have started using ASReview. At the moment, no systematic reviews have been published using the tool, but several reviews are work in progress. Our team has been working on a paper that is currently under peer review.
How can I refer to ASReview in my paper?¶
If you want to refer to ASReview project and research study in your paper, please cite our pre-print on ArXiv. We ask users of the software to cite the used version of our software. See https://doi.org/10.5281/zenodo.3345592 for a persistent link to our software.
Can the tool be used in combination with any (academic) database?¶
Yes it can. You will have to export data from a database yourself, and import these into ASReview. For supported databases and formats, please read the documentation on creating datasets.
How to deal with records that do not have abstracts?¶
Some records simply do not have abstracts. In case of missingness, we advise you to (quickly) screen unseen records with missing abstracts manually once you have finished screening with ASReview. However, it is very important for the performance of ASReview to have as little missing data as possible. We have written a blogpost on how the absence of abstracts impacts your review and how you can retrieve missing abstracts.
How does the tool handle quality and standardisation of abstracts? Is this accounted for in any way in training the machine learning model?¶
The texts of the documents are handled as is, there is no attempt to differentiate between e.g. different parts of abstracts. This could be done with standardized abstracts - but not all abstracts are standardized.
What happens if I have records from different languages?¶
The texts of the records are handled as is. ASReview does not differentiate between records that use different languages. Therefore, ASReview will have difficulty with identifying a relevant record when it is written in a language that is different from the rest of the records in your dataset.
Why did you choose a license that allows commercial reuse for the software?¶
We believe that free and open source software is important in advancing research. In the field of machine learning and systematic reviews, transparency is very important to give a better understanding of the process.
What do you mean with a dataset?¶
A datasets is a file that contains information such as the title, abstract, authors, doi etc. of all articles that are or have been screened.
What do you mean with a model?¶
A model (sometimes also referred to as a classifier) is a machine learning model that is used to predict the relevance of the records.
How do we decide when to stop?¶
At this moment, there is limited guidance on this; the decision of when to stop is left to the user. An example stopping rule can be:
- stop after screenings 25% of the records in the dataset
- 250 irrelevant records in a row (this number can be found in the statistics panel)
Can we use ASReview LAB with multiple screeners?¶
Currently, we do not support collaboration of multiple users within one project. We recommend multiple users to screen their records independently in separate projects. Afterwards, the results can be easily exported and combined to compare their screening decisions.
Can we use ASReview LAB also to screen full text?¶
ASReview LAB was originally designed for screening records with a title and an abstract. Viewing the full text can be accomplished by including a link to the original source of the publication by adding a Digital Object Identifier (DOI) to your dataset (column with name ‘doi’ in tabular data), which will be shown during screening. Note that the full text will not be used to train the model. Alternatively, you are free to put the full text into the abstract field of your dataset. When you put full-text in the abstract field, the full-text is used for display and training purposes.
Which classifier should I choose in ASReview LAB?¶
In ASReview, you need to choose which classifier you want to use to predict relevancy of your documents. Currently, we always advise to use the Naive Bayes classifier since it performs very well and needs little computation time. We have performed several simulation studies to evaluate performance of different classifiers on several datasets. We do not advise specific classifiers for specific jobs because we’ve not found enough evidence (yet) to make such recommendations.
I had already started labeling before I came across ASReview. How can I keep my former screening decisions when starting a new project in ASReview?¶
You can keep your former labeling decisions by adding an extra column in your dataset called ‘included’ or ‘label_included’. In this column, you can indicate previous screening decisions on records with 0s (irrelevant) and 1s (relevant). ASReview will use this information to train the model.
Is it possible to get the inclusion likelihood for unlabelled papers?¶
Unfortunately, no. Getting unbiased estimates for inclusion probabilities is a hard problem, especially in combination with active learning. Internally, we have scores that signify which papers are more likely included, but to avoid confusion, we do not put these in the export file. They are however available in the state files.
How can I make my previously labeled records green, like in the example datasets?¶
You can explore a previously labeled dataset in ASReview LAB by adding an extra column called ‘debug_label’, indicating the relevant and irrelevant records with ones and zeroes.
How do I remove duplicate publications?¶
ASReview LAB works best with deduplicated datasets. One can use software like EndNote to remove duplicates. See the following article for examples. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4915647/