Semantic Types Indexing (Real-Time)

STRI (Semantic Type Real-Time Indexing) uses statistical associations between the words in a training set of MEDLINE citations and a small set of 135 categories in the Semantic Network in NLM's UMLS Metathesaurus. Similar to STI, this method use ST-documents are created comprised of UMLS Metathesaurus string belonging to the ST to calculate ST score on real-time base (instead of pre-calculate). The procedures are briefly described as follows:

• Calculate the JDI scores (Words|JD|Wc|Dc) on the input Text or MeSH,
• Read in St-Jd scores (ST|JD|Wc|Dc) from file
• Calculate the vector similarity between JDI score of Input (Words|JD|Wc|Dc) and St-Jd score (ST|JD|Wc|Dc) by cosine coefficients and get St scores (Words|ST|Wc|Dc).
• Sort and display the results

This STRI program along with MEDLINE Tokenizer are used to indexing MEDLINE records on:

• Text: phrase, titles, abstracts, combination of titles and abstracts
• MeSHs: Starred MeSH headings and Subheadings

I. Java Software Components:

• Journal Descriptors
• Semantic Types

• InputFilter
  • Word Extraction Filter
  • Legal Words
• OutputFilter

• St-Jd Table

II. Programs:

• STRI on text
• STRI on MeSH