classifier

README.md

@@ -1,4 +1,19 @@
-adj_supersense_classifier
-=========================
-
 Adjective supersense classifier
+=====================
+
+
+
+Supersense classifier described in the paper
+
+Augmenting English Adjective Senses with Supersenses. 
+---------
+**Yulia Tsvetkov, Nathan Schneider, Dirk Hovy, Archna Bhatia, Manaal Faruqui and Chris Dyer.**
+In Proc. LREC'14.
+
+<i class="icon-share"></i> http://www.cs.cmu.edu/~ytsvetko/
+
+You can run the classifier like this:
+
+    ./adj_supersense_tagger.sh
+
+

adj_supersense_tagger.sh

@@ -0,0 +1,83 @@
+#!/bin/bash
+
+export ROOT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+export DATA_DIR=${ROOT_DIR}/data
+export WORK_DIR=${ROOT_DIR}/work # for temporary files
+export OUT_DIR=${ROOT_DIR}/predicted_supersenses # for classifier outputs
+export SRC_DIR=${ROOT_DIR}/src
+
+LABELS=${DATA_DIR}/seed.txt
+FEATURES=${DATA_DIR}/VSM/eacl14-faruqui-en-svd-de-64.adj.txt
+
+mkdir -p ${WORK_DIR}
+mkdir -p ${OUT_DIR}
+
+if [ -a ${FEATURES}.gz ] ; then
+  gunzip -f ${FEATURES}.gz
+fi 
+
+echo "Extract seed annotations"
+${SRC_DIR}/extract_seed.py --seed_dir ${DATA_DIR}/annotations \
+    --out_seed_file ${LABELS} 
+
+echo "Split Test and Train"
+${SRC_DIR}/split_train_test.py --seed_file ${LABELS} \
+    --features ${FEATURES} \
+    --out_train ${WORK_DIR}/train_seed.txt \
+    --out_test ${WORK_DIR}/test_seed.txt
+
+echo "Expand by WN synonyms and antonyms"
+${SRC_DIR}/expand_labeled_data.py --labeled_data ${WORK_DIR}/train_seed.txt \
+    --out_file ${WORK_DIR}/expanded.txt --expand 
+
+${SRC_DIR}/build_training_sets.py --in_file ${WORK_DIR}/train_seed.txt \
+    --out_feat ${WORK_DIR}/train_seed.feat \
+    --out_labels ${WORK_DIR}/train_seed.labels \
+    --test_set ${WORK_DIR}/expanded.txt \
+    --out_test_feat ${WORK_DIR}/expanded.feat \
+    --out_test_labels ${WORK_DIR}/expanded.labels \
+    --features ${FEATURES}
+
+echo "Run multi-way classifier. " #Default - Random Forest with 300 trees
+${SRC_DIR}/classify.py --train_features ${WORK_DIR}/train_seed.feat \
+    --train_labels ${WORK_DIR}/train_seed.labels \
+    --test_features ${WORK_DIR}/expanded.feat \
+    --test_predicted_labels_out ${WORK_DIR}/expanded.predicted \
+    --write_posterior_probabilities 
+
+echo "Selecting best expanded words"
+${SRC_DIR}/filter_expanded.py --predictions ${WORK_DIR}/expanded.predicted \
+    --orig_seed ${WORK_DIR}/train_seed.txt \
+    --out_file ${WORK_DIR}/expanded_seed.txt
+
+echo "Expand by WN synonyms and antonyms"
+${SRC_DIR}/expand_labeled_data.py --labeled_data ${WORK_DIR}/expanded_seed.txt \
+    --out_file ${WORK_DIR}/expanded.txt --expand 
+
+${SRC_DIR}/build_training_sets.py --in_file ${WORK_DIR}/expanded_seed.txt \
+    --out_feat ${WORK_DIR}/train_seed.feat \
+    --out_labels ${WORK_DIR}/train_seed.labels \
+    --out_test_feat ${WORK_DIR}/vocab.feat \
+    --features ${FEATURES} \
+    --include_training
+
+echo "Run multi-way classifier" #Default - Random Forest with 300 trees
+${SRC_DIR}/classify.py --train_features ${WORK_DIR}/train_seed.feat \
+    --train_labels ${WORK_DIR}/train_seed.labels \
+    --test_features ${WORK_DIR}/vocab.feat \
+    --test_predicted_labels_out ${WORK_DIR}/vocab.predicted \
+    --write_posterior_probabilities \
+
+echo "Accuracy-at-k:"
+${SRC_DIR}/eval.py --predicted_results ${WORK_DIR}/vocab.predicted \
+    --held_out_seed ${WORK_DIR}/test_seed.txt
+echo "(Results may vary for different runs)."
+echo ""
+
+${SRC_DIR}/soft_voting.py --predicted_results ${WORK_DIR}/vocab.predicted \
+    --out_lemma_file ${OUT_DIR}/words.predicted \
+    --out_synset_file ${OUT_DIR}/synsets.predicted
+
+echo "See classifier predictions for lemmas at ${OUT_DIR}/words.predicted"
+echo "See classifier predictions for synsets at ${OUT_DIR}/synsets.predicted"
+

data/MTurk/README

@@ -0,0 +1,15 @@
+adjectives+annotations+MACE.txt file contains SemCor adjectives 
+annotated by 3-5 Amazon Mechanical Turk workers. 
+
+File format 
+
+word \t annotations \t aggregated label
+
+Responses for each lemma are aggregated with the MACE tool, which takes 
+annotator-specific patterns into account to better determine the true label. 
+
+http://www.isi.edu/publications/licensed-sw/mace/
+
+Dirk Hovy, Taylor Berg-Kirkpatrick, Ashish Vaswani,
+and Eduard Hovy. 2013. Learning Whom to trust
+with MACE. In Proceedings of NAACL HLT.

data/VSM/README

@@ -0,0 +1,16 @@
+Word vectors created by Faruqui & Dyer (2014) are a variation on 
+traditional latent semantic analysis that uses multilingual information 
+to produce representations in which synonymous words have similar vectors.
+
+
+Before running the adjective classifier run this command:
+
+gunzip eacl14-faruqui-en-svd-de-64.adj.txt.gz
+
+Publication: 
+
+Manaal Faruqui and Chris Dyer. 2014. 
+Improving vector space word representations using multilingual correlation. 
+In Proceedings of the 14th Conference of the European Chapter of the 
+Association for Computational Linguistics,
+EACL ’14. Association for Computational Linguistics.

data/annotations/germanet_behavior.txt

@@ -0,0 +1,94 @@
+active
+adept
+aggressive
+alienated
+amiable
+amicable
+amoral
+ardent
+argumentative
+awkward
+bossy
+brave
+calm
+calmer
+caring
+charming
+cheerful
+compulsive
+considerate
+consistent
+cooperative
+courageous
+cordial
+crazy
+cruel
+deceitful
+despised
+despicable
+determined
+disciplined
+domineering
+egotistic
+energetic
+enthusiastic
+evaluative
+evil
+excitable
+faithful
+friendly
+funny
+furious
+furiouser
+generous
+greedy
+harmonious
+helpful
+hostile
+hypoactive
+impulsive
+inconsiderate
+inveterate
+irritating
+jovial
+kind
+lazy
+loving
+loyal
+manic
+manipulative
+moody
+moral
+mute
+organized
+panicked
+placid
+pleasing
+plodding
+poised
+polite
+popular
+queer
+reconciling
+reluctant
+rough
+rude
+screaming
+secluded
+selfish
+sincere
+sociable
+skeptical
+spiteful
+still
+strict
+sympathetic
+tactful
+talkative
+thoughtful
+thoughtless
+timid
+used_to
+vicious
+wild
+yielding

data/annotations/germanet_body.txt

@@ -0,0 +1,100 @@
+achy
+ailing
+albino
+alive
+athletic
+attractive
+bald
+beautiful
+beefy
+blind
+blonde
+blue-eyed
+breathing
+brown-eyed
+brunette
+chubby
+crippled
+curvy
+cute
+dead
+deaf
+debilitated
+defective
+dehydrated
+dirty
+disabled
+diseased
+dizzy
+drymouth
+exhausted
+fat
+feeble
+female
+feminine
+feverish
+fleshy
+gorgeous
+hairy
+handicapped
+handsome
+healthy
+heavy
+hungry
+ill
+impaired
+incapacitated
+infected
+irritated
+itching
+lame
+large
+lean
+lifeless
+lined
+living
+male
+masculine
+muscular
+mutilated
+numb
+obese
+ordinary
+pain
+paralyzed
+pettite
+plump
+pregnant
+pretty
+quivering
+redhead
+scarred
+scrawny
+scruffy
+shapely
+short
+shuddering
+sick
+sickly
+skinny
+slender
+slobbery
+small
+sound
+starving
+stocky
+stout
+strong
+sturdy
+sweaty
+tall
+tattooed
+thick
+thin
+thirsty
+tired
+trembling
+ugly
+undiseased
+unhealthy
+weak