2. IMDAMohammad Hamidi Esfahani
Introduction
Recommender systems are widely used on the Web
for recommending products and services to users.
Most e-commerce sites have such systems.
These systems serve two important functions.
They help users deal with the information overload by giving
them recommendations of products, etc.
They help businesses make more profits, i.e., selling more
products.
3. IMDAMohammad Hamidi Esfahani
Ecommerce Concerns
How to collect more users ???
How to settled the users in our website ???
Which product to sell or resell ???
Which users are important ????
7. IMDAMohammad Hamidi Esfahani
Collaborative Method
if user A and user B have a
purchase history that overlaps
strongly and user A has
recently bought an item that B
has not yet been, the basic
rationale is to propose this
item also to B
8. IMDAMohammad Hamidi Esfahani
Collaborative Method
Maintain a database of many users’ ratings of a variety
of items.
For a given user, find other similar users whose ratings
strongly correlate with the current user.
Recommend items rated highly by these similar users,
but not rated by the current user.
Almost all existing commercial recommenders use this
approach (e.g. Amazon).
9. IMDAMohammad Hamidi Esfahani
A 9
B 3
C
: :
Z 5
A
B
C 9
: :
Z 10
A 5
B 3
C
: :
Z 7
A
B
C 8
: :
Z
A 6
B 4
C
: :
Z
A 10
B 4
C 8
. .
Z 1
User
Database
Active
User
Correlation Match
A 9
B 3
C
. .
Z 5
A 9
B 3
C
: :
Z 5
A 10
B 4
C 8
. .
Z 1
Extract
Recommendations
C
10. IMDAMohammad Hamidi Esfahani
Collaborative Method
Weight all users with respect to similarity with the
active user.
Select a subset of the users (neighbors) to use as
predictors.
Normalize ratings and compute a prediction from a
weighted combination of the selected neighbors’
ratings.
Present items with highest predicted ratings as
recommendations.
11. IMDAMohammad Hamidi Esfahani
Typical Questions
How to find users with similar tastes to the user for whom we
need a recommendation?
How to measure similarity?
What should be done with new users, for whom a buying
history is not yet available?
How to deal with new items that nobody has bought yet?
What other techniques besides looking for similar users can
use for making a prediction about whether a certain user will
like an item?
12. IMDAMohammad Hamidi Esfahani
Content Based Method
if item A and item B have similar
attributes, the basic rationale is
to Recommend item B to whom
that buy item A
13. IMDAMohammad Hamidi Esfahani
Content Based Method (Conti..)
The key idea of content-based filtering approach is
based on similarity of item’s features.
The best idea in content-based recommendation
system, is to recommend similar items to user that
buy the first item previously.
14. IMDAMohammad Hamidi Esfahani
Content Based Method
Recommendations are based on information on the
content of items rather than on other users’ opinions.
Uses a machine learning algorithm to induce a profile of
the users preferences from examples based on a featural
description of content.
Some previous applications:
Newsweeder (Lang, 1995)
Syskill and Webert (Pazzani et al., 1996)
15. IMDAMohammad Hamidi Esfahani
Advantages of Content-Based Approach
No need for data on other users.
No cold-start or sparsity problems.
Able to recommend to users with unique tastes.
Able to recommend new and unpopular items
No first-rater problem.
Can provide explanations of recommended items by
listing content-features that caused an item to be
recommended.
15
16. IMDAMohammad Hamidi Esfahani
Disadvantages of Content-Based Method
Requires content that can be encoded as
meaningful features.
Users’ tastes must be represented as a learnable
function of these content features.
Unable to exploit quality judgments of other users.
Unless these are somehow included in the content
features.
16
17. IMDAMohammad Hamidi Esfahani
Typical Questions
How can systems automatically acquire and
continuously improve user profiles?
How to determine which items match, or are at least
similar to or compatible with, a user’s interests?
What techniques can be used to automatically
extract or learn the item descriptions to reduce
manual annotation?
18. IMDAMohammad Hamidi Esfahani
Knowledge
Based
(Rule Based)
Typical customers buy a new camera only once
every few years, so the recommender system
cannot construct a user profile or propose
cameras that others liked, which would result in
proposing only top-selling items.
19. IMDAMohammad Hamidi Esfahani
Knowledge Based Method
The recommender system typically makes use of
additional, often manually provided, information
about both the current user and the available items
User Interaction
Direct
Indirect
20. IMDAMohammad Hamidi Esfahani
Typical Questions
What kinds of domain knowledge can be represented in a knowledge
base?
What mechanisms can be used to select and rank the items based on the
user’s characteristics?
How to acquire the user profile in domains in which no purchase history is
available, and how to take the customer’s explicit preferences into
account?
Which interaction patterns can be used in interactive recommender
systems?
Finally, in which dimensions can personalize the dialog to maximize the
precision of the preference elicitation process?
21. IMDAMohammad Hamidi Esfahani
Hybrid Approaches
One obvious solution is to
combine different techniques
to generate better or more
precise recommendations
22. IMDAMohammad Hamidi Esfahani
Typical Questions
Which techniques can be combined, and what are
the prerequisites for a given combination?
Should proposals be calculated for two or more
systems sequentially, or do other hybridization
designs exist?
How should the results of different techniques be
weighted and can they be determined dynamically?
23. IMDAMohammad Hamidi Esfahani
Sample Hybrid Method
Proposed hybrid recommendation system method
Knowledge-
basedpart
Collaborative
part
Content-basedpart
User
interaction-
basedpart
Item’s
features
User’s
properties
C-Means Clustering Method
Extract Weighted Map
Users
History
Extract Rules
Prioritize Rules
Users Interaction
Like/Dislike
24. IMDAMohammad Hamidi Esfahani
Proposed Method (Conti..)
First off all, Extract the clusters on items and users
based an their profile features and the belonging
factor for each item or user
Second, Extract the relations and relation weight
between clusters
25. IMDAMohammad Hamidi Esfahani
Proteins
Classifying VS Clustering
Objects characterized by one or more features
Classification
Have labels for some points
Want a “rule” that will accurately assign labels to new
points
Supervised learning
Clustering
No labels
Group points into clusters based on how “near” they
are to one another
Identify structure in data
Unsupervised learning
Genes
27. IMDAMohammad Hamidi Esfahani
K-Nearest Neighbor
Features
All instances correspond to points in an n-dimensional
Euclidean space
Classification is delayed till a new instance arrives
Classification done by comparing feature vectors of the
different points
Target function may be discrete or real-valued
30. IMDAMohammad Hamidi Esfahani
K-Nearest Neighbor
An arbitrary instance is represented by (a1(x), a2(x), a3(x),..,
an(x))
ai(x) denotes features
Euclidean distance between two instances
d(xi, xj)=sqrt (sum for r=1 to n (ar(xi) - ar(xj))2)
Continuous valued target function
mean value of the k nearest training examples
31. IMDAMohammad Hamidi Esfahani
The K-Means Clustering Method
Given k, the k-means algorithm is implemented in four
steps:
1. Partition objects into k nonempty subsets
2. Compute seed points as the centroids of the clusters of the current partition
(the centroid is the center, i.e., mean point, of the cluster)
3. Assign each object to the cluster with the nearest seed point
4. Go back to Step 2, stop when no more new assignment
37. IMDAMohammad Hamidi Esfahani
Fuzzy C-means Clustering
Fuzzy c-means (FCM) is a method of clustering
which allows one piece of data to belong to two or
more clusters.
This method (developed by Dunn in 1973 and
improved by Bezdek in 1981) is frequently used in
pattern recognition.
46. IMDAMohammad Hamidi Esfahani
Proposed Method (Conti..)
item 𝐼 𝑎 belongs to cluster 𝐶𝑖 by 𝐵(𝐶𝑖𝑎) degree
Item 𝐼 𝑏 belong to cluster 𝐶𝑖 by 𝐵(𝐶𝑖𝑏) degree
User 𝑈 𝑎 belongs to cluster 𝐶 𝑢 by 𝐵(𝐶 𝑢𝑎) degree
User 𝑈 𝑏 belongs to cluster 𝐶 𝑢 by 𝐵(𝐶 𝑢𝑏) degree
𝑈 𝑎 buy 𝐼 𝑎 and user 𝑈 𝑏 don’t buy any item and item
𝐼 𝑏 never buy
47. IMDAMohammad Hamidi Esfahani
Proposed Method (Conti..)
cluster 𝐶𝑖 on items has a relation to cluster 𝐶 𝑢 on users by 𝑀(𝐶𝑖𝑢)
𝑀 𝐶𝑖𝑢 =
𝑥
𝑖𝑡𝑒𝑚𝑠
𝑦
𝑢𝑠𝑒𝑟𝑠 𝐵 𝐶 𝑢𝑥 ∗ 𝐵(𝐶 𝑖𝑦)
𝑚∗𝑛
m => the count of users
n => the count of items.
Then, a rule generated for recommending items in cluster 𝐶𝑖 on items to
users in cluster 𝐶 𝑢 on users. So, user 𝑈 𝑏 have a relation with item 𝐼 𝑏
𝑅𝑒𝑐 𝑈 𝑏, 𝐼 𝑏 = 𝑀 𝐶𝑖𝑢 ∗ 𝐵 𝐶𝑖𝑏 ∗ 𝐵 𝐶 𝑢𝑏