nlp-lab/Project/naive_approach/naive_approach.py

162 lines
5.1 KiB
Python
Raw Normal View History

2018-06-26 14:55:42 +02:00
# coding: utf-8
# In[1]:
import pandas as pd
from IPython.display import clear_output, Markdown, Math
import ipywidgets as widgets
import os
import unicodedata as uni
import numpy as np
from nltk.stem import PorterStemmer
from nltk.tokenize import sent_tokenize, word_tokenize
from nltk.corpus import wordnet
import math
import pprint
from gensim.models import Word2Vec, KeyedVectors
2018-06-26 14:55:42 +02:00
# # Naive Approach
table = pd.read_csv('../Tools/emoji_descriptions_preprocessed.csv')
2018-06-26 14:55:42 +02:00
2018-06-27 16:01:10 +02:00
##Store table in the format:
## { index: [emoji, description]}
tableDict = {}
for index, row in table.iterrows():
tableDict.update({index: [row['character'], row['description']]})
2018-06-26 14:55:42 +02:00
#######################
# Helper functions
#######################
def stemming(message):
2018-06-26 14:55:42 +02:00
ps = PorterStemmer()
words = word_tokenize(message)
sm = []
for w in words:
sm.append(ps.stem(w))
stemmed_message = (" ").join(sm)
return stemmed_message
2018-06-26 14:55:42 +02:00
# * compare words to emoji descriptions
def evaluate_sentence(sentence, description_key = 'description', lang = 'eng', emojis_to_consider="all",\
stem=True, use_wordnet=True):
# assumes there is a trained w2v model stored in the same directory!
__location__ = os.path.realpath(os.path.join(os.getcwd(), os.path.dirname(__file__)))
if use_wordnet==False:
wv = KeyedVectors.load(str(__location__)+"/word2vec.model", mmap='r')
2018-06-26 14:55:42 +02:00
if (stem):
sentence = stemming(sentence)
2018-06-26 14:55:42 +02:00
tokenized_sentence = word_tokenize(sentence)
n = len(tokenized_sentence)
matrix_list = []
2018-06-27 16:01:10 +02:00
for index in tableDict.keys():
emoji_tokens = word_tokenize(tableDict[index][1])
2018-06-26 14:55:42 +02:00
m = len(emoji_tokens)
mat = np.zeros(shape=(m,n))
for i in range(len(emoji_tokens)):
for j in range(len(tokenized_sentence)):
if use_wordnet:
syn1 = wordnet.synsets(emoji_tokens[i],lang=lang)
if len(syn1) == 0:
continue
w1 = syn1[0]
#print(j, tokenized_sentence)
syn2 = wordnet.synsets(tokenized_sentence[j], lang=lang)
if len(syn2) == 0:
continue
w2 = syn2[0]
val = w1.wup_similarity(w2)
if val is None:
continue
else:
try:
val = wv.similarity(emoji_tokens[i], tokenized_sentence[j])
except KeyError:
continue
2018-06-26 14:55:42 +02:00
mat[i,j] = val
matrix_list.append(mat)
return matrix_list
###########################
#Functions to be called from main script
###########################
# load and preprocess data
# emojis_to_consider can be either a list or "all"
def prepareData(stem=True, lower=True):
if(stem):
2018-06-27 16:01:10 +02:00
for index in tableDict.keys():
tableDict[index][1] = stemming(tableDict[index][1])
if(lower):
for index in tableDict.keys():
tableDict[index][1] = tableDict[index][1].lower()
2018-06-26 14:55:42 +02:00
#collect the emojis
lookup = {}
emoji_set = []
2018-06-27 16:01:10 +02:00
for index in tableDict.keys():
lookup[index] = tableDict[index][0]
emoji_set.append(tableDict[index][0])
2018-06-26 14:55:42 +02:00
emoji_set = set(emoji_set)
2018-06-26 14:59:24 +02:00
return lookup
2018-06-26 14:55:42 +02:00
# make a prediction for an input sentence
# use_wordnet=True --> use wordnet similarites, otherwise use Word2Vec
def predict(sentence, lookup, emojis_to_consider="all", criteria="threshold", lang = 'eng',\
use_wordnet=True, n=10, t=0.9):
2018-06-26 14:55:42 +02:00
result = evaluate_sentence(sentence, lang, emojis_to_consider=emojis_to_consider, use_wordnet=use_wordnet)
2018-06-26 14:55:42 +02:00
try:
if(criteria=="summed"):
resultValues = [-np.sum(x) for x in result]
elif (criteria=="max_val"):
resultValues = [-np.max(x) for x in result]
elif(criteria=="avg"):
resultValues = [-np.mean(x) for x in result]
else:
resultValues = [-len(np.where(x>t)[0]) / (x.shape[0] * x.shape[1]) for x in result]
indexes = np.argsort(resultValues)
results = np.sort(resultValues)
if (emojis_to_consider != "all" and type(emojis_to_consider) == list):
indexes2 = []
results2 = []
for i in range(len(indexes)):
if lookup[indexes[i]] in emojis_to_consider:
indexes2.append(indexes[i])
results2.append(results[i])
indexes = indexes2
results = results2
2018-06-27 16:01:10 +02:00
indexes = indexes[0:n]
results = results[0:n]
# build a result table
2018-06-27 16:01:10 +02:00
table_array = [lookup[indexes[i]] for i in range(n) ]
2018-06-27 16:01:10 +02:00
#table_frame = pd.DataFrame(table_array, columns=[criteria, 'description'])
#display(table_frame)
2018-06-27 16:01:10 +02:00
return table_array, results
2018-06-26 14:55:42 +02:00
except ZeroDivisionError as err:
print("There seems to be a problem with the input format. Please enter a nonempty string")
2018-06-27 16:01:10 +02:00
return [], []
2018-06-26 14:55:42 +02:00
#predict("I like to travel by train", description_key='description' , lang='eng')