How it works

The idea

A researcher is writing a paper and wants to publish it in an open access journal. As there is an increasing number of journals, it can be hard to find a suitable and trustworthy one.
They enter title, abstract and references into B!SON or ask a librarian who can consult B!SON for suggestions. Not all fields have to be filled out.

B!SON compares (1) title and abstract with articles listed in the DOAJ and (2) references with the citation database OpenCitations. Both DOAJ and OpenCitations are open data sets. B!SON does not store any information the users enter or collects any other usage data.

The researcher or librarian checks out the results and selects an appropriate journal. Please note: we highly recommend to double-check critical info, such as publishing costs, directly on the journal's website. B!SON relies on the data the journals have submitted to DOAJ about themselves and we cannot guarantee their correctness. At the same time, as all suggested journals had to go through the quality assurance process of DOAJ, the risk of choosing a predatory publisher is minimized.

How the recommendation works

First, title, abstract and references are analyzed separately to find similar articles in the DOAJ data set.

  • For title and abstract, the user input is checked against the title or abstract of articles in the DOAJ via Elasticsearch. At the beginning of the comparison, stop words, which are common words with little relevance like "the" or "a" in English, are removed. Afterwards, the text similarity determined using Okapi BM 25. This algorithm checks whether the same words appear in both texts and takes the relevance of words into account by checking their frequency. A rare, medical term has a higher importance than a generic word like "method". This process assign a numerical score for title and abstract of matching articles.
  • The combination of title and abstract is fed into a custom neural network which predicts journals and outputs a probability for each hit.
  • For the reference similarity, DOIs are extracted from the user input. In the next step, the OpenCitations data set is used to determine which articles are citing the given DOIs. If an article cites more than a (dynamic) threshold of input DOIs, it gets a score assigned.

Finally, the corresponding articles are matched with their journal and and a neuronal network weighs the scores and calculates a total score.

The language detection uses the lingua-py software and the LCC subject detection uses a customly trained neural network.

B!SON is open source, so you can look behind the scenes, file bugs, improve the code or host your own version. You can find the repository on GitLab.

Why B!SON?

Why should I use B!SON? How does B!SON differ from other journal search services on the web?

B!SON recommends open access journals in which research similar to a manuscript has been published. It uses title, abstract and references of the manuscript. The main feature of B!SON is the recommendation component based on semantic and bibliometric methods.

Further characteristics of B!SON are:

  • It includes only quality-assured open access journals.
  • It provides publisher-independent recommendations.
  • The algorithm on which the recommendations are based is transparently documented.
  • It is built exclusively on open data sources that are continuously updated (DOAJ, OpenCitations Meta, Journal Checker Tool/cOAlition S).
  • It does not collect usage data.
  • It is developed open source and without any commercial focus.
  • It is operated by large, scientific information infrastructure institutions (TIB and SLUB Dresden).