LAILAPS-QSM

LAILAPS-QSM applies machine learning to reconstruct linguistic context and generate alternative keyword suggestions to improve precision and completeness of full-text searches in life-science databases such as PubMed and UniProt.

Key Features:

• Machine Learning-Driven Context Reconstruction: Employs machine learning to reconstruct potential linguistic contexts for keyword queries using text records from PubMed and UniProt.
• Preprocessing of Text Records: Preprocesses and extracts relevant text data from databases to serve as input for context reconstruction and vector training.
• Customized Distributed Word Vectors: Computes customized distributed word vectors that capture semantic relationships between keywords for suggesting alternatives.
• Query Suggestion Using Word Vectors: Uses distributed word vectors to propose alternative keyword queries that enhance search precision and completeness in full-text life-science searches.
• Implementation and Optimized Performance: Implemented in JAVA with optimized data structures and streamlined code for fast, scalable computation.
• Evaluation Metrics: Achieved a mean information content similarity of 0.70 for 15 representative queries with 34% scoring above 0.80 compared to a human expert benchmark of 0.90.

Scientific Applications:

• Plant Science Quality Assessment: Evaluated in plant science use cases to support cost-efficient quality assessments across trait, biological entity, taxonomy, affiliation, and metabolic function categories by leveraging ontology term similarities.
• Enhanced Life-Science Database Search: Improves precision and completeness of keyword-based retrieval in full-text searches of life-science databases such as PubMed and UniProt.

Methodology:

Preprocessing text records from databases, computing distributed word vectors from those records, using machine learning to reconstruct query contexts, and suggesting alternative queries based on the vectors.