Analysis Engine: Document - 11791-04/project-team04 GitHub Wiki

Work Flow:

• reads the question in raw String from the Question Reader
• use a unigram model to wipeout some common terms that are not meaningful
• queries the PubMed API to obtain candidate documents
• process each document (title and abstract) and query by removing punctuation, stoppers and perform Krovetz Stemming
• rank documents based on their similarity to the query.
• Several rankers were tried, including Okapi BM25, Indri, Dirichlet and XQL.
• The results are written to jcas by Document.

Outline:

descriptorimpl.DocumentRetrieval_AE

• service : WebAPIServiceProxy
• stemmer : KrovetzStemmer
• outQuestions : PrintWriter
• baseline : boolean
• conceptSet : Set<String>
• initialize(UimaContext)
• qeWithConcept(String)
• process(JCas)
• collectionProcessComplete()

descriptorimpl.DocumentRetrieval_AE.DocScoreComparator

• compare(Pair<DocInfo, Double>, Pair<DocInfo, Double>)