Parity Computing
Transforming Data Into Knowledge Networks
Company Solutions Products Clients News
Products
About Parity Products
Profiler System
Reference Processor
Parity Journal Sense™


Journal Sense™

 Journal Sense allows authors to automatically identify suitable journals for publishing their research. Users can submit keywords, an abstract, or the entire text of their unpublished manuscript, and obtain journal recommendations. Journal Sense achieves significantly higher accuracy than traditional out-of-the-box matching solutions, and can often recommend relevant journals that authors may not be familiar with or did not think would be suitable for their research.

Methodology

  • Generates term vectors from query terms and matches against journal term vectors
    • Optimized accuracy over choices of:
      • Features (e.g., phrases, concepts, words, subject metadata)
      • Term normalization methods (e.g., stemming, lemmatization)
      • Weighting methods (e.g., TFIDF, BM25)
        • Proprietary Parity weighting function used to create term vector
  • Applies threshold function to partition result set into groups
    • Journals in the same group are statistically similar and across groups are statistically different

Evaluation

  • “Roll back the clock” automated accuracy evaluation:
    • Hold out a paper from the corpus
    • Use its abstract as a query
    • What is the rank of the actual published journal?
    • Compare to a deployed baseline system:
      • SOLR-based journal recommendation system
      • Uses standard techniques (stemming, TF-IDF weighting)
  • Expert manual accuracy evaluation of final system:
    • Performed by PhD-level domain experts
    • Relevance assessment of publications returned by query
    • Focus evaluation: queries where published journal is not top-ranked

Evaluation results from automated "roll back the clock" evaluation of the system

Measure

Baseline

Journal Sense

Number of queries

22,023

22,023

% of queries for which published journal is:

    #1 ranked

11.6%

44.5%

    In top-ranked set (average set size=3.3)

20.9%

66.7%

Average rank of published journal

68.9

9.8


Evaluation results from expert manual accuracy evaluation of final system

Measure

Journal Sense

When published journal is not #1 ranked, manually-evaluated precision of #1 ranked journal (50 queries)

80%

When published journal is not in the top-ranked set (average set size=3.3), manually evaluated precision of top-ranked set (50 queries)

91%


© Parity Computing, Inc. Privacy Policy


Request Literature
Parity Journal Sense™