search

knowitall / relgrams

Relgrams -- Tool for computing relational co-occurrences.
2 stars 0 forks source link

Not many relgrams with distinct verbs in relations. #6

Open niranjanb opened 11 years ago

niranjanb commented 11 years ago

Not many rel-grams with different verbs in the first and second tuples.

niranjanb commented 11 years ago

Attempting the following fixes:

1) Dealing with redundant extractions:

When two extractions were redundant, the extraction with a longer span was retained. e.g.,

(bomb, exploded this weekend in Fallujah pushing death toll to, 23)

would be favored over

(bomb, exploded, this weekend) (bomb, exploded this weekend in, Fallujah)

Fix involves outputting all extractions but not computing rel-grams between the redundant rel-grams.

2) Filtering out relations with communicative/reporting verbs (e.g., say, tell).

niranjanb commented 11 years ago

Other filtering possibilities include:

a) Extractions with confidences < 0.1 -- There are so few of these and in most cases these extractions are bad.

b) Extractions for which we cannot detect a noun head in either argument -- These are very few as well, and in most cases are non-sensical extractions.

c) Extractions where the arg2 occurs before rel or arg1 in the sentence of rel occurs before arg1 -- Some of these are useful but most appear bad. Needs further investigation.

d) Extractions with imposed prepositions -- Most of these are bad. We should definitely ignore these.

niranjanb commented 11 years ago

Fix 1 and 2 have been made resulting in a doubling of the number of rel-grams.