Issue with using QueryParser to parse TREC topics about anserini HOT 4 CLOSED

Here are some example of differences:

Oprah Winfrey half-sister gets interpreted as Lucene query:

text:oprah text:winfrei (text:half text:sister)

And release of "The Rite" as

text:releas text:of text:"the rite"

Super Bowl, seats

text:super (text:bowl text:) text:seat

from anserini.

Interestingly:

TREC 2011
QueryParser   MAP = 0.3577  P30 = 0.4000
BagOfWords   MAP = 0.3532  P30 = 0.3993

TREC 2012
QueryParser   MAP = 0.2091  P30 = 0.3311
BagOfWords   MAP = 0.2090  P30 = 0.3311

So the weird QueryParser interpretation of query operators helps a tiny bit for TREC 2011!

from anserini.

Bottom line: better to do it "right" rather than depend on something completely idiosyncratic.

from anserini.

Resolved in commit 36cafb7

from anserini.