SAGE Rejected Article Tracker: Check if an article has been published elsewhere.

What is it?

The SAGE rejected article tracker This package will take metadata(title, authors etc...) for research articles ("query articles") and will find the published versions in CrossRef ("match articles").

The published versions are selected using very simple machine-learning.

1. Numerical values are calculated for the difference in the titles of each query article and each candidate match article returned by CrossRef.
2. ArXiv preprints with known DOIs are used to train a simple logistic regression to find the correct candidate for each query.

This process is approximately 95% accurate. However, the package includes code to recreate the training dataset using the ArXiv and CrossRef APIs. This training dataset can be used to test other approaches to this problem and similar problems (such as detecting duplicate articles).

Presently, the package takes JSON input, but using a library such as pandas to convert it, spreadsheet data can be used.

If you are a ScholarOne user, you can produce this easily. More detailed instructions are below.

Where to get it?

Although the source code is on Github, the latest release is always published to the Python package index.

pip install rejected_article_tracker

Dependencies

• pandas
• openpyxl
• xlsxwriter
• fuzzywuzzy
• requests
• sklearn
• numpy

Typical Usage

# Import package
from rejected_article_tracker import RejectedArticlesMatch

# Create a list of article data dicts.
# PLEASE NOTE: Due to reliance on 3rd party APIs, the more articles the longer it takes.
 
articles = [ # Fake values fabricated for the example
{
      "journal_name": "FakeJ2",
      "manuscript_title": "Multiplex Genome Engineering Using CRISPR/Cas Systems",
      "authors": "Cong, Le; Ran, F. Ann; Cox, David",
      "final_decision": "",
      "decision_date": "2012-02-15T13:29:58.999Z", 
      "submission_date": "2012-02-14T13:29:58.999Z",
      "manuscript_id": "ABC-18-070",
    }
]

# @see below for configuration details.
config = {
    "threshold": 0.5, # Filters out matches which have a score below this value
}


# The CrossRef API requires an email address for lookups.    
email = os.getenv('MY_EMAIL','')


# Define a 'results' list. 
# This is a little unconventional and not great practice. 
# However, injecting a results variable, allows us grab already processed articles
# in case of an issue with 3rd party APIs. 
# This would then mean only the remaining articles would need to be rerun.
results = []

# Run match
RejectedArticlesMatch(
            articles=articles,
            config=config,
            email=email,
            results=results
        ).match()

print(results)

---

Article format

This is how each article should be structured:

Property	Contents
journal_name	Simply the name of the journal e.g. ‘Proceedings of mysociety’
manuscript_title	The text of the title of the manuscript you are looking for. The article will not be found without this.
authors	This is a critical part of the input data and MUST be formatted ‘LastName1, FirstName1; LastName2, FirstName2 …’ So note that authors names appear with their lastname first and then a comma, followed by the first name. Then, when we have multiple authors on a paper, we separate them with semicolons.
final_decision	E.g. ‘Accept’ or ‘Reject’ (the application will simply ignore everything that is listed as ‘Accept’)
decision_date	The date of final rejection from your journal. Should be a date string format, like: “YYYY-MM-DD”, “YYYY-MM-DD hh:mm:ss” Either format is fine.
submission_date	Same format as decision date.
manuscript_id	This can be any unique id that you use to identify the article. If it is formatted in the way that ScholarOne format their ids, then revision numbers will be ignored so that each article is only searched for once. So ‘ABC-20-123’, ‘ABC-20-123.R1’ and ‘ABC-20-123.R2’ all become ‘ABC-20-123’

---

Usage with directly downloaded Scholar One data

Preparing the input:

If using Scholar One, under ‘Peer Review Details Reports’ select ‘Build Your Own Reports’. The report should have the following columns:

‘Journal Name', 'Manuscript ID', 'Manuscript Title', 'Author Names', 'Submission Date', 'Decision Date', 'Accept or Reject Final Decision'

It’s ok to include other columns in the Excel file, but they are not needed. IMPORTANTLY: remember to download your report as Excel 2007 Data Format.

If not using ScholarOne, then simply prepare your submission data with the above column headings ensuring that

• Author lists are correctly formatted last_name1, first_name1; last_name2, first_name2;...
• Date columns (Submission Date and Decision Date) are correctly formatted, either as datetime strings (as below) or as simple date strings, e.g. 2010-12-31 for 31 December 2010.

Usage:

from rejected_article_tracker import ScholarOneRejectedArticlesMatch
import pandas as pd

df = pd.read_excel("/path/to/file")
allowed_cols = [
    'Journal Name',
    'Manuscript ID',
    'Manuscript Title',
    'Author Names',
    'Submission Date',
    'Decision Date',
    'Accept or Reject Final Decision'
]
articles = df[allowed_cols].to_dict('records')

# Which might look like:
"""  
articles = [ # Fake values fabricated for the example
{
      "Journal Name": "FakeJ1",
      "Manuscript Title": "Cosmological Consequences of a Rolling Homogeneous Scalar Field",
      "Author Names": "Ratra, Bharat; Peebles, P J E",
      "Accept or Reject Final Decision": "",
      "Decision Date": "1987-06-14T13:29:58.999Z", 
      "Submission Date": "1987-06-14T13:29:58.999Z",
      "Manuscript ID": "ABC-18-070",
    }
]
"""

# @see below for configuration details.
config = {
    "threshold": 0.5, 
}

# The CrossRef API requires an email address for lookups.    
email = "someome@example.com"

# Define a 'results' list.
results = []

# Run match
ScholarOneRejectedArticlesMatch(
    articles=articles,
    config=config,
    email=email,
    results=results
    ).match()

print(results)

Example output

Example output when match found:

[
  {
  "manuscript_id": "ABC-18-070", 
  "raw_manuscript_id": "ABC-18-070", 
  "journal_name": "FakeJ1", 
  "manuscript_title": "Cosmological Consequences of a Rolling Homogeneous Scalar Field", 
  "submission_date": "1987-06-14", 
  "decision_date": "1987-06-14", 
  "authors": "Bharat+Ratra, P J E+Peebles", 
  "text_sub_date": "1987-06-14", 
  "final_decision": "", 
  "match_doi": "10.1103/physrevd.37.3406", 
  "match_type": "journal-article", 
  "match_title": "Cosmological consequences of a rolling homogeneous scalar field", 
  "match_authors": "Bharat+Ratra, P. J. E.+Peebles", 
  "match_publisher": "American Physical Society (APS)", 
  "match_journal": "Physical Review D", 
  "match_pub_date": "1988-6-15", 
  "match_earliest_date": "2002-07-27", 
  "match_similarity": 89, 
  "match_one": 1, 
  "match_all": 1, 
  "match_crossref_score": 84.5133, 
  "match_crossref_cites": 2881, 
  "match_rank": 1, 
  "match_total_decision_days": 5521
  }
  ]

Example out when NO match found:

[
  {
    "manuscript_id": "ABC-18-070",
    "raw_manuscript_id": "ABC-18-070",
    "journal_name": "The International Journal of Robotics Research",
    "manuscript_title": "Learning hand-eye coordination for robotic grasping with deep learning and large-scale data collection",
    "submission_date": "2018-10-01",
    "decision_date": "2019-01-01",
    "authors": "Levine, Sergey; Pastor, Peter; Krizhevsky, Alex; Ibarz, Julian; Quillen, Deirdre",
    "text_sub_date": "2018-07-20",
    "final_decision": "",
    "match_doi": "No Match",
    "match_type": "No Match",
    "match_title": "No Match",
    "match_authors": "No Match",
    "match_publisher": "No Match",
    "match_journal": "No Match",
    "match_pub_date": "No Match",
    "match_earliest_date": "No Match",
    "match_similarity": "No Match",
    "match_one": "No Match",
    "match_all": "No Match",
    "match_crossref_score": "No Match",
    "match_crossref_cites": "No Match",
    "match_rank": "No Match",
    "match_total_decision_days": "No Match"
  }
]

To rebuild the training dataset and train a new model

Note that rebuilding the training dataset relies on external APIs and can be a very slow process (a few days depending on response times). However, once acquired, model training and testing takes seconds.

from rejected_article_tracker.src.ML.Train import LogReg

LogReg().best_model_to_file()

---

Configuration

Configuration is set using a dictionary. The following values can be set:

Name	Description	Example
threshold	A float in the range (0.0-1.0) value which determines the minimum "cut off" for scoring matching articles. Any matching articles below this score will not be returned. Default: 0.5. Raising this value improves precision, lowering it improves recall.	0.5
max_results_per_article	Integer (1-10) for the maximum number of results returned for each article. Note that some articles are published twice, or with supplementary data. Default: 10	10
article_types	There are various article types available. Some rejected articles end up as book chapters, preprints or proceedings. List the article-types you will accept. Default: returns all article types	['journal-article','book-chapter']

---

License

MIT

Contributing

All contributions, bug reports, bug fixes, documentation improvements, enhancements, and ideas are welcome.

Please see the contributing for instructions