Semantic Similarity Analysis

Leveraging WordNet for semantic similarity calculations with word disambiguation. The semantic similarity is based on maximum word similarity between one word and another sentence. In addition, word order similarity can be also considered in the semantic similarity calculations.

Setup the path and load DACKAR modules for similarity analysis

import os
import sys

cwd = os.getcwd()
frameworkDir = os.path.abspath(os.path.join(cwd, os.pardir, 'src'))
sys.path.append(frameworkDir)

import time
from dackar.similarity import synsetUtils as SU
from dackar.similarity import simUtils

Example

# sents = ['The workers at the industrial plant were overworked',
#       'The plant was no longer bearing flowers']
sents = ['RAM keeps things being worked with', 'The CPU uses RAM as a short-term memory store']
# sents = ['The pump is not working due to leakage', 'There is a leakage in the pump']

Compute sentences similarity without disambiguation

• delta \in [0, 1] is used to control the similarity contribution from semantic and word order similarity

similarity = simUtils.sentenceSimilarity(sents[0], sents[1], infoContentNorm=True, delta=.8)
print('Similarity Score: ',similarity)

Similarity Score:  0.5366827832678444

similarity = simUtils.sentenceSimilarity(sents[0], sents[1], infoContentNorm=False, delta=.8)
print('Similarity Score: ',similarity)

Similarity Score:  0.45519466650531865

Compute sentences similarity with disambiguation

similarity = simUtils.sentenceSimilarityWithDisambiguation(sents[0], sents[1], delta=.8)
print('Similarity Score: ',similarity)

Warming up PyWSD (takes ~10 secs)...

Similarity Score:  0.4897634312885952

took 1.7747421264648438 secs.

Convert sentences into synsets list, and then compute similarity

sentSynsets = simUtils.convertSentsToSynsets(sents)
similarity = SU.synsetListSimilarity(sentSynsets[0], sentSynsets[1], delta=.8)
print('Similarity Score: ',similarity)

Similarity Score:  0.2612493301234213

#### Using disambiguation method to create synsets

sentSynsets = simUtils.convertSentsToSynsetsWithDisambiguation(sents)

similarity = SU.synsetListSimilarity(sentSynsets[0], sentSynsets[1], delta=0.8)
print('Similarity Score: ',similarity)

Similarity Score:  0.4992052446929406

Timing for performance

st = time.time()
for i in range(100):
    sentSynsets = simUtils.convertSentsToSynsets(sents)
print('%s second'% (time.time()-st))

14.189666271209717 second

st = time.time()
for i in range(1000):
    similarity = SU.synsetListSimilarity(sentSynsets[0], sentSynsets[1], delta=.8)
print('%s second'% (time.time()-st))

5.876173257827759 second

st = time.time()
sentSynsets = []
for i in range(1000):
    for j in range(len(sents)):
        _, synsetsA = simUtils.sentenceSenseDisambiguationPyWSD(sents[j], senseMethod='simple_lesk', simMethod='path')
        sentSynsets.append(synsetsA)
print('%s second'% (time.time()-st))

1.1626660823822021 second