23. • Supervised Learning
Someone gives us examples and the right answer
• Unsupervised Learning
We see examples but get no feedback
We need to find patterns in the data
• Semi-supervised Learning
Given a small number of examples with the right answers, we need to
find patterns in the data, so that we can predict the right answer for
unseen examples
• Reinforcement Learning
We take actions and get rewards
Have to learn how to get high rewards
Different Kinds of Learning
24. Make learning set with results preview
for judging
whether it is related to restaurant review or not
We need to classify them!
34. p(Is | s)
p(s | Is) *p(Is)
p(s)
p(ls) : possibility of choosing Is_Restaurant(assume as 0.5)
p(s | ls) : In Is_Restaurant set
possibility of choosing the given sentece
35. p(Is | s)
p(s | Is) * p(Is)
p(s)
p(s | ls) : In Is_Restaurant set,
possibility of choosing the given sentence
p(s | Is) = p(word | Is) * p(word | Is) * p(word | Is)…
We assume that each word cannot affect to each other...
36. Is_Restaurant
“It was good to eat delicious pasta in this restaurant!”
p(Is) = 0.5
p(restaurant | Is) =
1000
250
p(delicious | Is) =
1000
300
p(good | Is) =
1000
250
p(eat | Is) =
1000
0(1)
restaurant
delicious
good
200
250
300
250
word count
atmosphere
37. “It was good to eat delicious pasta in this restaurant!”
0.5 * 0.001 * 0.3 * 0.25 * 0.25 =
9.375 * 10^-6
Is_Restaurant
restaurant
delicious
good
200
250
300
250
word count
atmosphere
38. 0.001 ^ 4 = 1 * 10^-12
“It was good to eat delicious pasta in this restaurant!”
Is_Not_Restaurant
politic
hello
economy
atmosphere
350
200
150
300
word count
39. This is related to restaurant review!
with upper table...
“It was good to eat delicious pasta in this restaurant!”
45. Title Date Content Reply URL
Database for Blogs
Point for one blog review
= (date * 0.5) + (the number of reply * 0.5)
Point for one restaurant
= the average of reviews’ point
54. What we want to develop...
Improve filtering blog performance
On Ruby on Rails framework, using BlackLight
served as library to use Solr in Ruby on Rails
Front-End Design using HTML5, CSS3, Javascript, Jquery...
New Ranking Algorithm
55. Point of Restaurant =
Points of Blogger = Recency, Frequency, Density
Recency, Frequency, People
Based on Blogger’s action
New Ranking Algorithm
61. 서가앤쿡 환여횟집 빕스 설빙 뚝배기 이탈리아
삼겹살 맛집 냉면 맛집 샤브샤브 맛집 보리밥 맛집 초밥 맛집
99% 96% 93% 100% 98%
100% 100% 100% 100% 100%
100 blogs related to Is_Restaurant
Performance
For one keywords => 10 blogs
62. 정치 경제 문화 사회 한동대학교
소녀시대 중앙일보 한겨례 성경 기독교
59% 66% 45% 65% 42%
81% 67% 62% 54% 58%
100 blogs related to Is_Not_Restaurant
Performance
For one keywords => 10 blogs