FAQ: PubMed® Enhanced Automatic Term Mapping
Question: Why was PubMed's Automatic Term Mapping (ATM) changed?
PubMed Automatic Term Mapping (ATM) was originally developed to identify author names, journal titles, and MeSH terms in search queries without search tags. In the old schema, author name and journal title fields were not searched when the search terms also matched a MeSH term. This caused thousands of citation-type searches to fail every day. The new ATM addresses this problem by removing those field restrictions. Under the new schema, query terms can be found in both MeSH terms and every PubMed citation field.
In conjunction with the new ATM, we have introduced the PubMed Citation Sensor, an engine designed to identify queries that contain citation-type information and match those terms to specific citation fields. The results retrieved when the Citation Sensor is invoked are a subset of the full result set retrieved with the new default ATM. Had we not changed ATM, the results from the Citation Sensor would have been a different result set from those produced by ATM - a somewhat confusing behavior for our users to understand.
The kinds of searches that have been affected by the new ATM are those in which terms are a journal or author name as well as a concept. In these cases, the number of results retrieved by the new ATM has changed to include the results from the searches hitting citation fields. For example, previously a search for "fox nature" would not have retrieved citations from the journal Nature by the author Fox; "gills lung neoplasms" retrieved nothing with the old mapping; however, the new mapping retrieves a number of records by broadening the search for "gills."The new ATM now retrieves results in cases where none were retrieved previously, for example, "wang science 1998 kcnq3" and "nature 7196 763".
Before releasing the new ATM to our live site, we ran tests on a set of thousands of queries that people had actually performed on our system, and found the overall statistical difference in the number of results returned to be very low overall (about 10%). In user tests that ran over a period of several weeks, no significant problems were discovered, nor complaints received. The Citation Sensor is now used in several thousand searches per day. From all our available indicators, the enhanced ATM has considerably improved searching PubMed. However, if you have any specific cases of searches that now cause a significant problem, please let us know and we would be glad to look at these more closely.
For more information, see:
- Searching PubMed (in PubMed Help) http://www.ncbi.nlm.nih.gov/books/bv.fcgi?rid=helppubmed.section.pubmedhelp.Searching_PubMed.
- PubMed: MEDLINE® Retrieval on the World Wide Web Fact Sheet at http://www.nlm.nih.gov/pubs/factsheets/pubmed.html