ASReview Datatools

ASReview Datatools is an extension to ASReview LAB that can be used to: - Describe basic properties of a dataset - Convert file formats - Deduplicate data - Stack multiple datasets - Compose a single (labeled, partly labeled, or unlabeled) dataset from multiple datasets - Snowball a dataset to find incoming or outgoing citations - Sample old, random, and new papers in order to check if the terminology has changed over time.

Several tutorials are available that show how ASReview-Datatools can be used in different scenarios.

ASReview datatools is available for ASReview LAB version 1 or later. If you are using ASReview LAB version 0.x, use ASReview-statistics instead of ASReview datatools.

Installation

ASReview Datatools requires Python 3.7+ and ASReview LAB version 1.1 or later.

The easiest way to install the extension is to install it from PyPI:

bash pip install asreview-datatools

After installation of the datatools extension, asreview should automatically detect it. Test this with the following command:

bash asreview --help

The extension is successfully installed if it lists asreview data.

To make sure that you are working with the latest version of datatools you can use:

bash pip install asreview-datatools --upgrade

Getting started

ASReview Datatools is a command line tool that extends ASReview LAB. Each subsection below describes one of the tools. The structure is

bash asreview data NAME_OF_TOOL

where NAME_OF_TOOL is the name of one of the tools below (describe, convert, dedup, vstack, or compose) followed by positional arguments and optional arguments.

Each tool has its own help description which is available with

bash asreview data NAME_OF_TOOL -h

Tools

Data Describe

Describe the content of a dataset

bash asreview data describe MY_DATASET.csv

Export the results to a file (output.json)

bash asreview data describe MY_DATASET.csv -o output.json

Describe the van_de_schoot_2018 dataset from the benchmark platform.

bash asreview data describe synergy:van_de_schoot_2018 -o output.json { "asreviewVersion": "1.1", "apiVersion": "1.1.1", "data": { "items": [ { "id": "n_records", "title": "Number of records", "description": "The number of records in the dataset.", "value": 6189 }, { "id": "n_relevant", "title": "Number of relevant records", "description": "The number of relevant records in the dataset.", "value": 43 }, { "id": "n_irrelevant", "title": "Number of irrelevant records", "description": "The number of irrelevant records in the dataset.", "value": 6146 }, { "id": "n_unlabeled", "title": "Number of unlabeled records", "description": "The number of unlabeled records in the dataset.", "value": 0 }, { "id": "n_missing_title", "title": "Number of records with missing title", "description": "The number of records in the dataset with missing title.", "value": 5 }, { "id": "n_missing_abstract", "title": "Number of records with missing abstract", "description": "The number of records in the dataset with missing abstract.", "value": 764 }, { "id": "n_duplicates", "title": "Number of duplicate records (basic algorithm)", "description": "The number of duplicate records in the dataset based on similar text.", "value": 104 } ] } }

Data Convert

Convert the format of a dataset. For example, convert a RIS dataset into a CSV, Excel, or TAB dataset.

asreview data convert MY_DATASET.ris MY_OUTPUT.csv

Data Dedup

Remove duplicate records with a simple and straightforward deduplication algorithm. The algorithm first removes all duplicates based on a persistent identifier (PID). Then it concatenates the title and abstract, whereafter it removes all non-alphanumeric tokens. Then the duplicates are removed.

asreview data dedup MY_DATASET.ris

Export the deduplicated dataset to a file (output.csv)

asreview data dedup MY_DATASET.ris -o output.csv

By default, the PID is set to 'doi'. The dedup function offers the option to use a different PID. Consider a dataset with PubMed identifiers (PMID), the identifier can be used for deduplication.

asreview data dedup MY_DATASET.csv -o output.csv --pid PMID

Using the van_de_schoot_2018 dataset from the benchmark platform.

bash asreview data dedup synergy:van_de_schoot_2018 -o van_de_schoot_2018_dedup.csv Removed 104 records from dataset with 6189 records.

We can also choose to deduplicate based on the similarity of the title and abstract, instead of checking for an exact match. This way we can find duplicates that have small differences, but are actually the same record (for example, an additional comma or a fixed typo). This can be done by using the --drop_similar flag. This process takes about 4s on a dataset of about 2068 entries.

bash asreview data dedup neurips_2020.tsv --drop_similar Not using doi for deduplication because there is no such data. Deduplicating: 100%|████████████████████████████████████| 2068/2068 [00:03<00:00, 531.93it/s] Found 2 duplicates in dataset with 2068 records.

If we want to check which entries were found as duplicates, we can use the --verbose flag. This will print the lines of the dataset that were found as duplicates, as well as the difference between them. Any text that has to be removed from the first entry to become the second one is shown as red and has a strikethrough, and any text that has to be added to the first entry is shown as green. All text that is the same in both entries is dimmed.

bash asreview data dedup neurips_2020.tsv --drop_similar --verbose

The similarity threshold can be set with the --similarity flag. The default similarity threshold is 0.98. We can also choose to only use the title for deduplication by using the --skip_abstract flag.

bash asreview data dedup neurips_2020.tsv --drop_similar --similarity 0.98 --skip_abstract Not using doi for deduplication because there is no such data. Deduplicating: 100%|████████████████████████████████████| 2068/2068 [00:02<00:00, 770.74it/s] Found 4 duplicates in dataset with 2068 records.

Note that you might have to adjust your similarity score if you choose to only use the title for deduplication. The similarity score is calculated using the SequenceMatcher class from the difflib package. The similarity score is calculated as the ratio of the number of matching characters to the total number of characters in the two strings. For example, the similarity score between the strings "hello" and "hello world" is 0.625. By default, we use the realquickratio and quick_ratio methods, which are faster and usually good enough, but less accurate. If you want to use the ratio method, you can use the --strict_similarity flag.

Now, if we want to discard stopwords for deduplication (for a more strict check on the important words), we can use the --discard_stopwords flag. The default language for the stopwords is english, but that can be set with the --stopwords_language flag. The list of supported languages for the stopwords are the same supported by the nltk package. To check the list of available languages, you can run the following commands on your python environment:

python from nltk.corpus import stopwords print(stopwords.fileids()) ['arabic', 'azerbaijani', 'basque', 'bengali', 'catalan', 'chinese', 'danish', 'dutch', 'english', 'finnish', 'french', 'german', 'greek', 'hebrew', 'hinglish', 'hungarian', 'indonesian', 'italian', 'kazakh', 'nepali', 'norwegian', 'portuguese', 'romanian', 'russian', 'slovene', 'spanish', 'swedish', 'tajik', 'turkish']

Data Vstack (Experimental)

Vertical stacking: combine as many datasets in the same file format as you want into a single dataset.

❗ Vstack is an experimental feature. We would love to hear your feedback. Please keep in mind that this feature can change in the future.

Stack several datasets on top of each other: asreview data vstack output.csv MY_DATASET_1.csv MY_DATASET_2.csv MY_DATASET_3.csv Here, three datasets are exported into a single dataset output.csv. The output path can be followed by any number of datasets to be stacked.

This is an example using the demo datasets:

bash asreview data vstack output.ris dataset_1.ris dataset_2.ris

Data Compose (Experimental)

Compose is where datasets containing records with different labels (or no labels) can be assembled into a single dataset.

❗ Compose is an experimental feature. We would love to hear your feedback. Please keep in mind that this feature can change in the future.

Overview of possible input files and corresponding properties, use at least one of the following arguments:

| Arguments | Action | |----------------------|--------------------------------------------| | --relevant, -r | Label all records from this dataset as relevant in the composed dataset. | | --irrelevant, -i | Label all records from this dataset as irrelevant in the composed dataset. | | --labeled, -l | Use existing labels from this dataset in the composed dataset. | | --unlabeled, -u | Remove all labels from this dataset in the composed dataset. |

The output path should always be specified.

Duplicate checking is based on title/abstract and a persistent identifier (PID) like the digital object identifier (DOI). By default, doi is used as PID. It is possible to use the flag --pid to specify a persistent identifier other than doi. In case duplicate records are detected, the user is warned, and the conflicting records are shown. To specify what happens in case of conflicts, use the --conflict_resolve/-c flag. This is set to keep_one by default, options are:

| Resolve method | Action in case of conflict | |----------------|-----------------------------------------------------------------------------------------| | keep_one | Keep one label, using --hierarchy to determine which label to keep | | keep_all | Keep conflicting records as duplicates in the composed dataset (ignoring --hierarchy) | | abort | Abort |

In case of an ambiguously labeled record (e.g., one record with two different labels), use --hierarchy to specify a hierarchy of labels. Pass the letters r (relevant), i (irrelevant), and u (unlabeled) in any order to set label hierarchy. By default, the order is riu meaning that relevant labels are prioritized over irrelevant and unlabeled, and irrelevant labels are prioritized over unlabeled ones.

Asume you have records in MY_DATASET_1.ris from which you want to keep all existing labels and records in MY_DATASET_2.ris which you want to keep unlabeled. Both datasets can be composed into a single dataset using:

bash asreview data compose composed_output.ris -l DATASET_1.ris -u DATASET_2.ris --hierarchy uir -c abort Because of the flag -c abort in case of conflicting/contradictory labels, the user is warned, records with inconsistent labels are shown, and the script is aborted. The flag --hierarchy uir results in the following hierarch if any duplicate ambiguously labeled records exist: unlabeled is prioritized over irrelevant and relevant labels, and irrelevant labels are prioritized over relevant labels.

Snowball

ASReview Datatools supports snowballing via the asreview data snowball subcommand. It can perform both backwards (outgoing citations) and forwards (incoming citations) snowballing. The tool works by searching the OpenAlex database for citation data. An example usage would be:

bash asreview data snowball input_dataset.csv output_dataset.csv --forward

This performs forwards snowballing on input_dataset.csv and writes the results to output_dataset.csv. For this to work it is necessary that the input dataset contains a column with DOI's or a column called openalex_id containing OpenAlex work identifiers. The output dataset will contain the columns id, doi, title, abstract, referenced_works and publication_date. In the case of forward snowballing it will contain all works in OpenAlex that have a reference to one of the included works in the input dataset. In the case of backward snowballing it will contain all works in OpenAlex with referenced by one of the included works of the input dataset.

If you want to find references for all records in your dataset, instead of just the included works, you can include the flag --all, so for example:

bash asreview data snowball input_dataset.csv output_dataset.csv --backward --all

One thing to note is that OpenAlex will handle data requests faster if the sender sends along their email with the request (see OpenAlex Polite Pool), you can to this using the --email argument. An example would be:

bash asreview data snowball input_dataset.csv output_dataset.csv --backward --email my_email@provider.com

Sample

This datatool is used to sample old, random and new records from your dataset by using the asreview data sample command. The sampled records are then stored in an output file. This can be useful for detecting concept drift, meaning that the words used for certain concepts change over time. This script assumes that the dataset includes a column named publication_year. An example would be:

bash asreview data sample input_dataset.xlsx output_dataset.xlsx 50 This samples the 50 oldest and 50 newest records from input_dataset.xlsx and samples 50 records randomly (without overlap from the old and new partitions!). The resulting 150 records are written to output_dataset.xlsx.

License

This extension is published under the MIT license.

Contact

This extension is part of the ASReview project (asreview.ai). It is maintained by the maintainers of ASReview LAB. See ASReview LAB for contact information and more resources.