Science Concierge: A fast content-based recommendation system for scientific publications

Can a few examples of relevant papers guide scientific recommendations?

Science Concierge is a content-based recommender that learns from papers a researcher marks as relevant or irrelevant. It uses their text to find related work and updates recommendations as the researcher provides feedback.

What the study found

• In an evaluation using Society for Neuroscience conference abstracts, the method outperformed the tested keyword-based baseline.
• Positive and negative feedback can be combined to refine a researcher's interest profile.

How the study works

The pipeline represents article text numerically, reduces its dimensionality with latent semantic analysis, and uses Rocchio relevance feedback to update a query profile. Nearest-neighbor search retrieves matching abstracts. Human-assigned conference topics provide a relevance proxy for evaluation.

Scope and limitations

• Conference topic similarity is a proxy for usefulness, not a direct measure of scientific novelty or a reader's satisfaction.
• The evaluation concerns one conference collection and does not establish performance across all research fields.

Using this work

Use the linked Python repository to explore the recommendation pipeline with a corpus you are permitted to analyze. The conference dataset used in the paper is not publicly downloadable; access requires arrangements with the Society for Neuroscience.

Implementation and documentation

Cite this work

Titipat Achakulvisut, Daniel E Acuna, Tulakan Ruangrong, Konrad Kording (2016). Science Concierge: A fast content-based recommendation system for scientific publications. PloS ONE. https://doi.org/10.1371/journal.pone.0158423

@article{achakulvisut2016science,
  title = {Science Concierge: A fast content-based recommendation system for scientific publications},
  author = {Achakulvisut, Titipat and Acuna, Daniel E and Ruangrong, Tulakan and Kording, Konrad},
  year = {2016},
  publication_date = {2016-07-06},
  journal = {PloS ONE},
  volume = {11},
  number = {7},
  pages = {e0158423},
  publisher = {Public Library of Science San Francisco, CA USA},
  doi = {10.1371/journal.pone.0158423},
  url = {https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0158423}
}