Welcome to RobotSearch, software for filtering RCTs from a search result as described in our paper in the Journal of Research Synthesis Methods.
We offer RobotSearch free of charge, but we'd be most grateful if you would cite us if you use it. We're academics, and thrive on links and citations! Getting RobotSearch widely used and cited helps us obtain the funding to maintain the project and make RobotSearch better.
It also makes your methods transparent to your readers, and not least we'd love to see where RobotSearch is used! :)
For most people, we encourage you to use RobotSearch via our website.
No need to install anything, simply upload your RIS files (Ovid or PubMed format), and instantly download the filtered version containing RCTs only. N.B. To ensure the highest accuracy, please make sure that your RIS files are exported with all the fields needed (see below for instructions for Ovid and Pubmed).
For those who are particularly technically minded, or have a pressing need to run the software on their own machines, read on...
Currently this software runs from the Command Prompt (in Windows), or the Terminal (in Mac, or Linux).
-
Before installing RobotSearch, you will need to install Python 3. We recommend that you use the MiniConda Python distribution (N.B. choose version 3.6 or higher). You can download this here.
-
Open up the Terminal (or Command Prompt in Windows). This is also how you will interact with RobotSearch when you use it.
-
Install RobotSearch using the following command (easiest to copy/paste):
pip install -U https://github.com/ijmarshall/robotsearch/archive/master.zip -
RobotSearch should be automatically downloaded and installed on your machine. The software is >200MB in size, so this process may take some time depending on how fast your internet connection is.
-
You should be ready to go!
We include the code for the online version also in this repository. The easiest way to run is via Docker.
From within the code directory run:
docker build -t robotsearch
If the build is successful, you can then start the website locally by running:
./start.sh
You can then access the website on any webbrowser on your local machine at: http://localhost:5050.
To stop the websever, run:
docker stop robotsearch
RobotSearch uses machine learning as an alternative to string-based study design filters.
It works with MEDLINE searches exported from either PubMed or Ovid.
-
You should conduct your search as normal (NB do not use any search terms or filters to restrict to RCTs at this stage!).
-
Then export your results in RIS format (see more detailed instrutions below on how to do this in PubMed and Ovid)
-
From the Command Prompt/Terminal, run the robotsearch command:
robotsearch my_file.ris -
Your search results will be saved as
my_file_robotreviewer_RCTs.ris
By default, RobotSearch runs a sensitive search (i.e. very high likelihood that all RCTs will be retrieved, at expense of sometimes including non-RCTs) - this is suitable for a systematic review.
To run a precise search (i.e. the retrieved articles have a very high likelihood of being RCTs, but at the expense of missing a tiny proportion), run with an extra -p flag, e.g.:
robotsearch my_file.ris -p
- Select 'Send to' (located in the upper left of the search results)
- Under 'Choose Destination' select File
- Under 'Format', select 'MEDLINE' from the pulldown menu
- Click the 'Create File' button
(Endnote format includes the 'publication type' field, the RIS format does not)
- Click the tickbox next to 'All' above the search results on the left hand side.
- Click on Export
- In the 'Export to' pulldown menu, select 'RIS'
- Under 'Select Fields to Display' select 'Custom Fields', then click the 'Select Fields' button
- In the Select Fields box, select the following fields: ab: Abstract, pt: Publication Type, and ti: Title. You may deselect any others.
- Click 'Save' in the bottom left
- Click 'Export Citation(s)' in the bottom right
- The citations will be exported, but there may be a wait of a minute or two.
RobotSearch has a optional test mode, which runs through a standardised search result, and double checks that the software is returning the same results as in the validation study. NB this takes around 5--10 minutes to run, depending on the speed of your machine.
To run this, type:
robotsearch -t
We would love software developers, and people who make biblographic databases to integrate this method into their systems. All the software is open source under the GNU GPL v3. Please contact us ([email protected]) to discuss further.
We also welcome contributions from the community; please contact us (or submit an issue via Github) if you are interested in improving the software.
RobotSearch may be called as a Python module, from within the root directory. See the example IPython notebook for how to do this.
This work is dependent on the work of the Cochrane Crowd, for whom we'd like to express enormous gratitude for their work, and in making their data available for training our machine learning models.
This work is supported by: National Institutes of Health (NIH) under the National Library of Medicine, grant R01-LM012086-01A1, "Semi-Automating Data Extraction for Systematic Reviews", and by NIH grant 5UH2CA203711-02, "Crowdsourcing Mark-up of the Medical Literature to Support Evidence-Based Medicine and Develop Automated Annotation Capabilities", and the UK Medical Research Council (MRC), through its Skills Development Fellowship program, grant MR/N015185/1
If you use RobotSearch, we'd love it if you could cite our paper, and also let us know! (you can email [email protected])