def __init__(self):
self.MIN_SCORE = .1
self.dbConnection = HelpConnection()
self.filt = NormalTokFilter()
self.title_weight = 1
self.tag_weight = .5
self.rating_weight = .3
self.view_weight = .1
self.body_weight = 1
self.max_views = 1
self.max_rating = 1
self.stop_words = ['i', 'the', 'of', 'to', 'and', 'a', 'in', 'is', 'it', 'you', 'that', 'he', 'was', 'for', 'on',
'are', 'with', 'as', 'I', 'his', 'they', 'be', 'at', 'one', 'have', 'this', 'from', 'or',
'had', 'by', 'hot', 'but', 'some', 'we', 'can', 'out', 'other', 'were',
'all', 'there', 'when', 'up', 'use', 'your', 'said', 'an', 'each', 'she', 'which',
'do', 'their', 'will', 'way', 'about', 'many', 'then', 'them',
'would', 'like', 'so', 'these', 'her', 'thing', 'see', 'him',
'has', 'look', 'more', 'day', 'could', 'go', 'come', 'did', 'no',
'most', 'people', 'my', 'over', 'know', 'water', 'who', 'may',
'down', 'object', 'side', 'been', 'now', 'any', 'work', 'part', 'take',
'place', 'made', 'live', 'back', 'little', 'only', 'round', 'man']
class HInferenceEngine():
def __init__(self):
self.MIN_SCORE = .1
self.dbConnection = HelpConnection()
self.filt = NormalTokFilter()
self.title_weight = 1
self.tag_weight = .5
self.rating_weight = .3
self.view_weight = .1
self.body_weight = 1
self.max_views = 1
self.max_rating = 1
self.stop_words = ['i', 'the', 'of', 'to', 'and', 'a', 'in', 'is', 'it', 'you', 'that', 'he', 'was', 'for', 'on',
'are', 'with', 'as', 'I', 'his', 'they', 'be', 'at', 'one', 'have', 'this', 'from', 'or',
'had', 'by', 'hot', 'but', 'some', 'we', 'can', 'out', 'other', 'were',
'all', 'there', 'when', 'up', 'use', 'your', 'said', 'an', 'each', 'she', 'which',
'do', 'their', 'will', 'way', 'about', 'many', 'then', 'them',
'would', 'like', 'so', 'these', 'her', 'thing', 'see', 'him',
'has', 'look', 'more', 'day', 'could', 'go', 'come', 'did', 'no',
'most', 'people', 'my', 'over', 'know', 'water', 'who', 'may',
'down', 'object', 'side', 'been', 'now', 'any', 'work', 'part', 'take',
'place', 'made', 'live', 'back', 'little', 'only', 'round', 'man']
"""
takes list of strings as input, returns string
"""
def infer(self, msg):
qid = self.find_best_answer(msg)
response = self.findAnswer(qid)
return response
def find_best_answer(self, msg):
# Find first 100 question/answer pairs that contain
# the specified keywords in either question title/test or
# answer text
tmp = list(set(msg).difference(self.stop_words))
stripped = tmp if tmp else list(set(msg))
qDict = dict()
q = 'select q.qid, q.creator, q.editor, q.title, q.text, ' \
'q.rating, q.num_views, q.favorited, q.created, q.edited from question q where %s order by q.rating desc'
q = q % ' or '.join(["q.title ilike '%%%s%%'" % k for k in stripped])
qs = self.dbConnection.query(q)
if not qs:
return None
self.max_views = float(max(qs, key=lambda x: x[6])[6])
self.max_rating = float(qs[0][5])
for qu in qs:
(qid, creator, editor, title, body, rating, num_views, favorited, created, edited) = qu
rank = self.rankQuestion(qu, msg)
qDict[qu] = (rank, qid)
best = max(qDict.values(), key=lambda x: x[0])
return best[1] if best[0] >= self.MIN_SCORE else None
def get_body_score(self, body, keywords):
temp1 = set(self.filt.filter(body)).difference(self.stop_words)
temp2 = set(keywords).difference(self.stop_words)
body_set = temp1 if temp1 else set(self.filt.filter(body))
keyword_set = temp2 if temp2 else set(keywords)
return (len(body_set.intersection(keyword_set))/float(len(body_set))) * self.body_weight
def get_tag_score(self, tags, keywords):
temp = set(keywords).difference(self.stop_words)
keyword_set = temp if temp else set(keywords)
tag_set = set(tags)
return (len(tag_set.intersection(keyword_set))/float(len(tag_set))) * self.tag_weight
def rankQuestion(self, question, msg):
(qid, creator, editor, title, body, rating, num_views, favorited, created, edited) = question
tags = map(lambda x: x[0], self.dbConnection.query(('select distinct tag from tag_question where qid = %s', [qid])))
title = self.filt.filter(title)
# Tag score
tag_score = self.get_tag_score(tags, msg)
# Body score
body_score = self.get_body_score(body, msg)
#Title score
titleScore = self.getTitleScore(title, msg)
#Views score
viewScore = self.getViewScore(num_views)
#Rank score
ratingScore = self.getRatingScore(rating)
#Sum the weights here
rank = titleScore + viewScore + ratingScore + body_score + tag_score
return rank
def getRatingScore(self, rating):
return self.rating_weight * (rating / self.max_rating)
def getViewScore(self, num_views):
return self.view_weight * (num_views / self.max_views)
def getTitleScore(self, title, msg):
return self.title_weight * self.calcDist(title, msg)
def calcDist(self, title, message):
title_features = set([])
title_features = title_features.union(title).union(message)
titleVector = [0]*len(title_features)
messageVector = [0]*len(title_features)
for i, word in enumerate(title_features):
if word in title:
titleVector[i] = 1
if word in message:
messageVector[i] = 1
return ((len(title_features) -
sum([(x - y)**2 for x, y in zip(titleVector, messageVector)]))) / float(len(title_features))
#Finds the most likely answer in a very basic way
def findAnswer(self, qid):
# Find the answer associated with the best qid
try:
answers = self.dbConnection.query(
("SELECT answer.text " \
"FROM answer " \
"WHERE answer.qid = %s " \
"ORDER BY answer.rating DESC", [str(qid)]))
print qid
return answers[0]
except:
self.dbConnection.connection.commit()
return "I could not find an answer"
