def idf():
IDF = {}
numDocs = 0
# Get all the products from the database
dat = data.getDatabase()
for table in dat:
#print '.'
# Go through each product in each table
for product in dat[table]:
item = product[3]
# Get their reviews
revs = GetReviews.readReview(item)["Reviews"]
try:
for r in revs:
#print r
# Tokenize and Stem reviews
con = Tokenizer.stemming(Tokenizer.tokenize(r['Cons']))
pro = Tokenizer.stemming(Tokenizer.tokenize(r['Pros']))
comment = Tokenizer.stemming(Tokenizer.tokenize(r['Comments']))
#print 'Before:',r['Cons'],'\n\nAfter:',con
# Count unique tokens in the document
for token in list(set(con) | set(pro) | set(comment)):
if token in IDF: IDF[token] = IDF[token] + 1
else: IDF[token] = 1
numDocs = numDocs + 1
# Increment the number of documents
except: pass
# Calculate and return the idf score
for term in IDF:
IDF[term] = math.log(float(numDocs)/float(IDF[term]))
pickle.dump(dict(IDF),open('../data/idf.pickle','wb')) # Pickling saves SOOO much time
return IDF
def tf_idf():
TF_IDF = {}
# Load the inverse document frequencies
#IDF = idf()
IDF = dict(pickle.load(open('../data/idf.pickle','rb')))
dat = data.getDatabase() # get all of the products
for table in dat:
print '.' # progress marker
for product in dat[table]: # For each product in each table
item = product[3] # Item number is [3] in the tuple
revs = GetReviews.readReview(item)["Reviews"] # we want to read the actual reviews
product_review = []
try:
for r in revs: # for each review
tf = {}
# Tokenize and stem the entire review
con = Tokenizer.stemming(Tokenizer.tokenize(r['Cons']))
pro = Tokenizer.stemming(Tokenizer.tokenize(r['Pros']))
comment = Tokenizer.stemming(Tokenizer.tokenize(r['Comments']))
# combine pros, cons, and comments sections
for token in list(con+pro+comment): # calculate the term frequencies
if token in tf: tf[token] = tf[token] + 1
else: tf[token] = 1
for t in tf:
tf[t] = float(1+math.log(tf[t]))*IDF[t] # calculate tf-idf score
product_review.append(tf) # add to list of reviews
except: pass
TF_IDF[item] = product_review # add list of reviews to the dictionary
return TF_IDF
def BuildTrainingSet():
# initialize variables
database = data.getDatabase()
IDF = dict(pickle.load(open('../data/idf.pickle','rb')))
numReviews = 0
posReview = {}
numPos = 0
negReview = {}
numNeg = 0
# For each product in each subcategory
for table in database:
for product in database[table]:
item = product[3]
revs = GetReviews.readReview(item)["Reviews"]
# get the review
try:
for r in revs:
if (numReviews%37)==0: # Analyze every 37th review
tf = {}
# Get reviews for you to read
con = r['Cons']
pro = r['Pros']
comment = r['Comments']
# Read the reviews
print pro,' :: ',con,' :: ',comment
# set up to add to training set
con = Tokenizer.stemming(Tokenizer.tokenize(r['Cons']))
pro = Tokenizer.stemming(Tokenizer.tokenize(r['Pros']))
comment = Tokenizer.stemming(Tokenizer.tokenize(r['Comments']))
# Treat all parts as one review
for token in list(con+pro+comment):
if token in tf: tf[token] = tf[token] + 1
else: tf[token] = 1
for t in tf:
# tf-idf formula
tf[t] = float(1+math.log(tf[t]))*IDF[t]
# hopefully you have had time to read, now decide
Q = int(raw_input('\n1 for good.... 0 for bad.....\n').rstrip('\n'))
if Q==1: # Good
posReview[numPos] = tf # add to training set
numPos = numPos + 1
elif Q==0: # Bad
negReview[numNeg] = tf # add to training set
numNeg = numNeg + 1
else: print 'FAIL!!!!!!'
numReviews = numReviews + 1 # increase number of reviews
except: pass
saveSet(posReview,negReview) # Save the training sets
return (numPos, numNeg)
