Coverage of open citations in DOAJ journals - Protocol

We download the public data dump of DOAJ articles in .tar.gz format and the public data dump of OpenCitations' COCI in zipped files containing several .csv files. The other index of OpenCitation, CROCI, is not available as public data dump, so for this research we consider only COCI.

Downloading from DOAJ was not problematic. A few issues had to be addressed with COCI.

The first problem we faced is how to obtain these citations directly from the platform. The possibilities are essentially reduced to two: either through the API service or through the direct download of the entire dataset, made free and accessible by the platform itself.

The first one, after several attempts, has been discarded. The DOIs of our interest, coming from DOAJ, are approximately 5 million. Considering that each request containing a DOI addressed to the OC API takes an average of 1.2 seconds to return a response, it would be impossible to quickly conclude the collection of all the data needed for research purposes.

OC implements beyond the API a SPARQL endpoint, immediately discarded for the same reasons just explained. In fact this system, more dated than the API, has response times much longer than the latter, despite allowing greater flexibility in research.

The only solution that could be reconciled with the search times is the direct download of the entire dataset . The main problem with this option is the amount of data to be preserved on the hard drive. Especially if you try to read the data by unzipping the content.

The choice to focus on the latter method, then, is mainly to be found in the timing to complete the data collection.

From the DOAJ dump we create a DOAJ_journals.json. Inside this JSON file we create a key for each journal, selecting: the issn (if it is present), eissn (if it is present), the title of the journal, the list of all the articles' DOIs.

We also create another file containing all articles's DOIs from DOAJ ( articles_DOIs.json), which has just the aim to semplify next working steps.

Then we manage the OC dump . We unzip the first level of files, obtain a series of repository. In this way, we avoid extending the required memory for storing files.

Then we iterate over all the CSV files inside the zip directory and we obtain from them a series of data frames.

On each dataframe, we select only rows that have a DOAJ DOI in the cited or citing column. To accomplish this work we use the DOIs stored inside the articles_DOIs.json .

Once we apply this filter on dataframe we saved for each iteration two new CSV: one with only the DOAJ DOIs in the cited ( references.json ) column and another one with citing ( citations.json ) ones.

Then for convenience, we transform these latter CSV into a JSON format. In this way, we can reorganize all the content divided as the DOAJ_journals.json , by journals.

We then count the number of elements in references.json and citations.json and update the DOAJ_journals.json with the fields: reference_count and citation_count.

Then, we check if the citing DOI in the citations of references.json and the cited DOI in the citations of citations.json are open by looking at the oa_link.

The result will be open_citations.json and open_references.json with all of the citations for each journal and DOAJ_journals.json will be updated with a counter of the open citations coming and done by the journals ( open_citations_count , open_references_count ).

We combine open_citations.json and open_references.json. We group the citations based on the creation_date, taking just the year.

The result will be open_access_citations_by_year . json of all of these citations for every year that was found.

We visualize our results with python libraries.

We specifically visualize open_access_citations_by_year . json in a graph that specifies the number of citations in the y axis and the list of years in the x axis.

We publish all the CSV and JSON data that we gathered on our Data Management Plan in Zenodo and also in our Github repository.