Emati

Emati recommends biomedical research articles using content-based machine learning to prioritize newly published literature for relevance.

Key Features:

• Content-Based Recommendation Approach: Emati analyzes document text directly and operates independently of the number of users.
• TF-IDF with Multinomial Naïve Bayes: Documents are converted into TF-IDF-weighted features used by a Multinomial Naïve Bayes classifier that outputs probability scores for relevance.
• BERT Language Model: Emati uses BERT (Bidirectional Encoder Representations from Transformers) fine-tuned for text classification to leverage deep contextual language understanding.
• Weekly Ranking by Probability Scores: The system produces weekly-updated lists of article recommendations ranked by classifier probability scores.
• Personalized Search of PubMed and arXiv: Emati performs searches of PubMed and arXiv and sorts the returned results according to classifier probability scores.

Scientific Applications:

• Literature discovery and triage: Automates identification and prioritization of relevant biomedical articles from newly published literature to support researchers, clinicians, and academics.
• Relevance-based monitoring of research developments: Ranks candidate articles to reduce time required for staying abreast of new biomedical research.

Methodology:

Content-based recommendation using TF-IDF vectorization and a Multinomial Naïve Bayes classifier that provides probability scores, BERT (Bidirectional Encoder Representations from Transformers) fine-tuned for text classification, queries to PubMed and arXiv, and weekly re-ranking of articles by classifier probability scores.