import networkx as nx
import re
import sys
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.feature_extraction import text
import ext.common_english_words as common_english_words
import ext.extend_stop_words as custom_stop_words
from nltk.stem.wordnet import WordNetLemmatizer
import numpy as np
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.cluster import KMeans, MiniBatchKMeans
from time import time
from sklearn.decomposition import TruncatedSVD
from sklearn.preprocessing import Normalizer
from sklearn.pipeline import make_pipeline
from scipy.spatial.distance import cdist
import matplotlib.pyplot as plt
import util
sys.path.append('../lib')
import config
import util
sys.path.append('../..')
import ext
import ext.common_english_words as common_english_words
import ext.extend_stop_words as custom_stop_words
[docs]def nick_change_graph(log_dict, DAY_BY_DAY_ANALYSIS = False):
""" creates a graph which tracks the nick changes of the users
where each edge has a time stamp denoting the time
at which the nick was changed by the user
Args:
log_dict (str): Dictionary of logs created using reader.py
Returns:
list of the day_to_day nick changes if config.DAY_BY_DAY_ANALYSIS=True or else an aggregate nick change graph for the
given time period.
"""
rem_time = None #remembers the time of the last message of the file parsed before the current file
nick_change_day_list = []
aggregate_nick_change_graph = nx.MultiDiGraph() # graph for nick changes in the whole time span (not day to day)
for day_content_all_channels in log_dict.values():
for day_content in day_content_all_channels:
day_log = day_content["log_data"]
today_nick_change_graph = nx.MultiDiGraph() #using networkx
current_line_no = -1
for line in day_log:
current_line_no = current_line_no + 1
if(line[0] == '=' and "changed the topic of" not in line): #excluding the condition when user changes the topic. Search for only nick changes
nick1 = util.correctLastCharCR(line[line.find("=")+1:line.find(" is")][3:])
nick2 = util.correctLastCharCR(line[line.find("wn as")+1:line.find("\n")][5:])
earlier_line_no = current_line_no
while earlier_line_no >= 0: #to find the line just before "=="" so as to find time of Nick Change
earlier_line_no = earlier_line_no - 1
if(day_log[earlier_line_no][0] != '='):
year, month, day = util.get_year_month_day(day_content)
util.build_graphs(nick1, nick2, day_log[earlier_line_no][1:6], year, month, day, today_nick_change_graph, aggregate_nick_change_graph)
break
if(earlier_line_no == -1):
today_nick_change_graph.add_edge(nick1, nick2, weight=rem_time)
aggregate_nick_change_graph.add_edge(nick1, nick2, weight = rem_time)
count = len(day_log) - 1 #setting up the rem_time for next file, by noting the last message sent on that file.
while(count >= 0):
if(day_log[count][0] != '='):
rem_time = day_log[count][1:6]
break
count = count-1
nick_change_day_list.append(today_nick_change_graph)
if DAY_BY_DAY_ANALYSIS:
return nick_change_day_list
else:
return aggregate_nick_change_graph
[docs]def top_keywords_for_nick(user_keyword_freq_dict, nick, threshold, min_words_spoken):
"""
outputs top keywords for a particular nick
Args:
user_keyword_freq_dict(dict): dictionary for each user having keywords and their frequency
nick(str) : user to do analysis on
threshold(float): threshold on normalised values to seperate meaningful words
min_words_spoken(int): threhold on the minumum number of words spoken by a user to perform analysis on
Returns:
null
"""
keywords = None
for dicts in user_keyword_freq_dict:
if dicts['nick'] == nick:
keywords = dicts['keywords']
break
total_freq = 0.0
for freq_tuple in keywords:
total_freq += freq_tuple[1]
top_keywords = []
top_keywords_normal_freq = []
if total_freq > min_words_spoken:
if keywords:
for keyword in keywords:
if keyword[2] >= threshold:
top_keywords.append(keyword[0].encode('ascii', 'ignore'))
top_keywords_normal_freq.append(keyword[2])
if len(top_keywords) == 0:
if config.DEBUGGER:
print "No word's normalised score crosses the value of", threshold
top_keywords = None
else:
if config.DEBUGGER:
print "No message sent by nick", nick
pass
else:
if config.DEBUGGER:
print "Not enough words spoken by", nick, "; spoke" ,int(total_freq), "words only, required", min_words_spoken
pass
return (top_keywords, top_keywords_normal_freq)
[docs]def keywords(log_dict, nicks, nick_same_list):
"""
Returns keywods for all users
Args:
log_dict (str): Dictionary of logs data created using reader.py
nicks(List) : list of nickname created using nickTracker.py
nick_same_list :List of same_nick names created using nickTracker.py
Returns
keywords_filtered: filtered keywords for user
user_keyword_freq_dict: dictionary for each user having keywords and their frequency
user_words_dict: keywods for user
nicks_for_stop_words: stop words
"""
user_words_dict = []
user_keyword_freq_dict = []
keywords_filtered = []
no_messages = 0
for day_content_all_channels in log_dict.values():
for day_content in day_content_all_channels:
day_log = day_content["log_data"]
for line in day_log:
flag_comma = 0
if(util.check_if_msg_line (line)):
m = re.search(r"\<(.*?)\>", line)
var = util.correctLastCharCR((m.group(0)[1:-1]))
for d in range(len(nicks)):
if var in nick_same_list[d]:
nick_sender = nick_same_list[d][0]
break
else:
nick_sender = var
nick_receiver = ''
for i in nicks:
rec_list = [e.strip() for e in line.split(':')] #receiver list splited about :
util.rec_list_splice(rec_list)
if not rec_list[1]: #index 0 will contain time 14:02
break
rec_list = util.correct_last_char_list(rec_list)
for z in rec_list:
if(z == i):
if(var != i):
for d in range(len(nicks)):
if i in nick_same_list[d]:
nick_receiver = nick_same_list[d][0]
break
else:
nick_receiver = i
if "," in rec_list[1]: #receiver list may of the form <Dhruv> Rohan, Ram :
flag_comma = 1
rec_list_2 = [e.strip() for e in rec_list[1].split(',')]
rec_list_2 = util.correct_last_char_list(rec_list_2)
for j in rec_list_2:
if(j == i):
if(var != i):
for d in range(len(nicks)):
if i in nick_same_list[d]:
nick_receiver = nick_same_list[d][0]
break
else:
nick_receiver = i
if(flag_comma == 0): #receiver list can be <Dhruv> Rohan, Hi!
rec = util.correctLastCharCR(line[line.find(">")+1:line.find(", ")][1:])
if(rec==i):
if(var != i):
for d in range(len(nicks)):
if i in nick_same_list[d]:
nick_receiver = nick_same_list[d][0]
break
else:
nick_receiver = i
#generating the words written by the sender
message = rec_list[1:]
no_messages += 1
correctedNickReciever = util.correct_nick_for_(nick_receiver)
if correctedNickReciever in message:
message.remove(correctedNickReciever)
lmtzr = WordNetLemmatizer()
#limit word size = 3, drop numbers.
word_list_temp = re.sub(r'\d+', '', " ".join(re.findall(r'\w{3,}', ":".join(message).replace(","," ")))).split(" ")
word_list = []
#remove punctuations
for word in word_list_temp:
word = word.lower()
word_list.append(word.replace("'",""))
word_list_lemmatized = []
try:
word_list_lemmatized = map(lmtzr.lemmatize, map(lambda x: lmtzr.lemmatize(x, 'v'), word_list))
except UnicodeDecodeError:
pass
fr = 1
for dic in user_words_dict:
if dic['sender'] == nick_sender:
dic['words'].extend(word_list_lemmatized)
fr = 0
if fr:
user_words_dict.append({'sender':nick_sender, 'words':word_list_lemmatized })
nicks_for_stop_words = []
stop_word_without_apostrophe = []
for l in nick_same_list:
nicks_for_stop_words.extend(l)
for dictonary in user_words_dict:
nicks_for_stop_words.append(dictonary['sender'])
nicks_for_stop_words.extend([x.lower() for x in nicks_for_stop_words])
for words in common_english_words.words:
stop_word_without_apostrophe.append(words.replace("'",""))
stop_words_extended = text.ENGLISH_STOP_WORDS.union(common_english_words.words).union(nicks_for_stop_words).union(stop_word_without_apostrophe).union(custom_stop_words.words).union(custom_stop_words.slangs)
count_vect = CountVectorizer(analyzer = 'word', stop_words=stop_words_extended, min_df = 1)
for dictonary in user_words_dict:
try:
matrix = count_vect.fit_transform(dictonary['words'])
freqs = [[word, matrix.getcol(idx).sum()] for word, idx in count_vect.vocabulary_.items()]
keywords = sorted(freqs, key = lambda x: -x[1])
total_freq = 0.0
for freq_tuple in keywords:
total_freq += freq_tuple[1]
for freq_tuple in keywords:
freq_tuple.append(round(freq_tuple[1]/float(total_freq), 5))
user_keyword_freq_dict.append({'nick':dictonary['sender'], 'keywords': keywords })
except ValueError:
pass
for data in user_keyword_freq_dict:
keywords, normal_scores = top_keywords_for_nick(user_keyword_freq_dict, data['nick'], config.KEYWORDS_THRESHOLD, config.KEYWORDS_MIN_WORDS)
if config.DEBUGGER:
print "Nick:", data['nick']
print "Keywords with normalised score > 0.01\n", keywords
print "Their Normal scores\n", normal_scores
print "\n"
if keywords:
keywords_filtered.append({'nick': data['nick'], 'keywords': keywords})
return keywords_filtered, user_keyword_freq_dict, user_words_dict, nicks_for_stop_words
[docs]def keywords_clusters(log_dict, nicks, nick_same_list):
"""
Uses `keywords` to form clusters of words post TF IDF (optional).
Args:
log_dict (str): Dictionary of logs data created using reader.py
nicks(List) : list of nickname created using nickTracker.py
nick_same_list :List of same_nick names created using nickTracker.py
Returns
null
"""
'''
AUTO TFIDF FROM JUST SENTENCES
'''
#http://scikit-learn.org/stable/auto_examples/text/document_clustering.html
#BUILDING CORPUS
keyword_dict_list, user_keyword_freq_dict, user_words_dict_list, nicks_for_stop_words = keywords(log_dict, nicks, nick_same_list)
corpus = []
for user_words_dict in user_words_dict_list:
corpus.append(" ".join(map(str,user_words_dict['words'])))
print "No. of users", len(corpus)
#TF_IDF
stop_word_without_apostrophe = []
for words in common_english_words.words:
stop_word_without_apostrophe.append(words.replace("'",""))
stop_words_extended = text.ENGLISH_STOP_WORDS.union(common_english_words.words).union(nicks_for_stop_words).union(stop_word_without_apostrophe).union(custom_stop_words.words).union(custom_stop_words.slangs)
vectorizer = TfidfVectorizer(max_df=0.5, min_df=2, stop_words=stop_words_extended,
use_idf=True)
print "Extracting features from the training dataset using TF-IDF"
t0 = time()
tf_idf = vectorizer.fit_transform(corpus)
print("done in %fs" % (time() - t0))
print "n_samples: %d, n_features: %d \n" % tf_idf.shape
# LSA
if config.ENABLE_SVD:
print("============USING SVD==========")
print("Performing dimensionality reduction using LSA")
t0 = time()
# Vectorizer results are normalized, which makes KMeans behave as
# spherical k-means for better results. Since LSA/SVD results are
# not normalized, we have to redo the normalization.
svd = TruncatedSVD(100) #recommened value = 100
normalizer = Normalizer(copy=False)
lsa = make_pipeline(svd, normalizer)
tf_idf = lsa.fit_transform(tf_idf)
print("done in %fs" % (time() - t0))
explained_variance = svd.explained_variance_ratio_.sum()
print("Explained variance of the SVD step: {}%".format(
int(explained_variance * 100)))
if not config.ENABLE_ELBOW_METHOD_FOR_K:
# CLUSTERING
km = KMeans(n_clusters=config.NUMBER_OF_CLUSTERS, init='k-means++',
random_state=3465, max_iter=100, n_init=8)
print("Clustering sparse data with %s" % km)
t0 = time()
km.fit(tf_idf)
print("done in %0.3fs" % (time() - t0))
print("Top terms per cluster:")
if config.ENABLE_SVD:
original_space_centroids = svd.inverse_transform(km.cluster_centers_)
order_centroids = original_space_centroids.argsort()[:, ::-1]
else:
order_centroids = km.cluster_centers_.argsort()[:, ::-1]
np.set_printoptions(threshold=np.nan)
terms = vectorizer.get_feature_names()
for i in range(config.NUMBER_OF_CLUSTERS):
print("Cluster %d:" % i)
for ind in order_centroids[i, :config.SHOW_N_WORDS_PER_CLUSTER]:
print terms[ind]+"\t"+str(round(km.cluster_centers_[i][ind], 2))
print ""
else:
print "============ELBOW METHOD ============="
sum_squared_errors_list = []
avg_sum_squared_errors_list = []
for i in xrange(1, config.CHECK_K_TILL + 1):
print "\n===>> K = ", i
km = KMeans(n_clusters=i, init='k-means++', max_iter=100, n_init=8)
t0 = time()
km.fit(tf_idf)
if config.ENABLE_SVD:
original_space_centroids = svd.inverse_transform(km.cluster_centers_)
order_centroids = original_space_centroids.argsort()[:, ::-1]
else:
order_centroids = km.cluster_centers_.argsort()[:, ::-1]
distance_matrix_all_combination = cdist(tf_idf, km.cluster_centers_, 'euclidean')
# cIdx = np.argmin(distance_matrix_all_combination,axis=1)
distance_from_nearest_centroid = np.min(distance_matrix_all_combination, axis=1)
sum_squared_errors = sum(distance_from_nearest_centroid)
avg_sum_squared_errors = sum_squared_errors/tf_idf.shape[0]
print "Sum Squared Error =", sum_squared_errors
print "Avg Sum Squared Error =", avg_sum_squared_errors
sum_squared_errors_list.append(sum_squared_errors)
avg_sum_squared_errors_list.append(avg_sum_squared_errors)
print("Top terms per cluster:")
terms = vectorizer.get_feature_names()
for i in range(i):
print("Cluster %d:" % i)
for ind in order_centroids[i, :config.SHOW_N_WORDS_PER_CLUSTER]:
print(' %s' % terms[ind])
print()
plt.plot(range(1, config.CHECK_K_TILL+1), sum_squared_errors_list, 'b*-')
# ax.plot(K[kIdx], avgWithinSS[kIdx], marker='o', markersize=12,
# markeredgewidth=2, markeredgecolor='r', markerfacecolor='None')
plt.grid(True)
plt.xlabel('Number of clusters')
plt.ylabel('Average sum of squares')
plt.title('Elbow for KMeans clustering')
plt.show()
#NOTE RANDOM OUTPUTS BECAUSE OF RANDOM INITIALISATION
print "NOTE RANDOM OUTPUTS BECAUSE OF RANDOM INITIALISATION"