Submit a Task

A list of PubMed IDs, one per line, to serve as relevant training examples (the rest of Medline is used to approximate term frequencies in irrelevant articles). Any number of PubMed IDs may be submitted, but at least 10 is recommended. Broad topics may need hundreds or thousands of examples.

Usually, the classifier is trained on MeSH terms and journal of publication of each Medline record. Select this box to include title/abstract/author as well, which slows down the query from 90 seconds to 3-5 minutes. Feel free to select it if you are prepared to wait a bit longer for somewhat improved results.

The name of the task, to make it easier to locate on the results page (the name is restricted to letters, numbers and .,;:- punctuation)

A short code that will be required to delete the output (defaults to an empty code). You may select "hide output" to prevent the results from being listed on the output page, in which case you need to bookmark the location of the results to find them later.

Upper limit on the number of citations to return (may be set between 100 and 10000).

Limit Medline to records completed after this date. This feature is useful for monitoring for new literature. This adds 30 seconds to normal filtering, and adds 2 minutes if the use all features box is selected.

Advanced option: Only returns articles that score at least this much. You may want to use a negative threshold when the sample data includes very few articles it is more probable than irrelevance.

In cross validation we select this many random Medline records to serve as irrelevant training examples. To perform cross validation you will also need to paste at least 30 relevant examples in the "Input Citations" box.