Recommendation Algorithms:

Collaborative Filtering

CSE 6111 Presentation Advanced Algorithms Fall. 2013

Presented by: Farzana Yasmeen

2013.11.29

2

Contents

● What are recommendation algorithms?

– Recommendations

– Recommender Systems

– Recommender Algorithms

● Focus: Collaborative Filtering – Nearest Neighbour

● Conclusions

– Pros and Cons

● Discussion

3

Recommendations

● What are recommendations?

– Alternative to ‘search’

– Relevant content : movies, online shopping, radio stations….

●

•Customer is looking for a product

•Receive personal offerings

•Receive tips

4

Recommendations

● Benefits?

5

Recommender Systems

● What are recommender systems?

– predict the opinion of users based on prior knowledge

Models

•Data collection and processing •Relevance & preference ordering

•Display recommendations

•Self-learning & improving capabilities 1

2

3

4

Recommender System

6

Recommendation Algorithms (1)

The best fitting algorithms are selected – after careful analysis of the data – to

the given recommendation problem and the corresponding optimization task.

given (input) = recommendation problem

(task) = corresponding optimization

required (output) = a recommendation

7

Recommendation Algorithms (2)

● Content-based Filtering (CBF):

– item triggered (user, item metadata)

– keyword matching

– key problem: learn and apply cross-content (i.e. decision trees)

● Collaborative Filtering (CF):

– event triggered (vod purchase, live channel watching)

– finds similarities on users and items (vod content, live schedule)

– key problem: how to combine and weight the preferences of user

neighbors (i.e. nearest neighbor)

● Hybrid Model:

– combining the watching and searching habits of similar users

(collaborative filtering)

– offering content with shared characteristics that a user has rated

highly (content-based filtering).

because you saw most of the major horror movies, here is minor horror movies

Content-based Filtering

A user

User’s Favorite Movies

Already Saw it... Nothing Special... Well…

Collaborative Filtering

Collaborative Filtering

Collaborative Filtering

Collaborative Filtering

Collaborative Filtering

CF Algorithms

Collaborative Filtering

Recommender Space

• Ratings or Vote data = m x n sparse binary matrix

– n columns = “products”, e.g., books for purchase or movies for viewing

– m rows = users

– Interpretation:

• Implicit Ratings: v(i,j) = user i’s rating of product j (e.g. on a scale of 1 to 5)

• Explicit Purchases: v(i,j) = 1 if user i purchased product j

• entry = 0 if no purchase or rating

Implicit Ratings

Users Items

Observed preferences

The Recommender Space as a Bipartite Graph

Item-Item

Links

User-User

Links

Links derived from similar

attributes, similar content,

explicit cross references

Links derived from

similar attributes,

explicit connections

(Ratings, purchases, page

views, play lists,

bookmarks, etc)

Near-Neighbor Algorithms for Collaborative Filtering

ri,k = rating of user i on item k

Ii = items for which user i has generated a rating

Mean rating for user i is

Predicted vote for user i on item j is a weighted sum

weights of K similar users Normalization constant

(e.g., total sum of weights)

Value of K can be optimized on a validation data set

Near-Neighbor Weighting

• K-nearest neighbor

• Pearson correlation coefficient (Resnick ’94, Grouplens):

Sums are over items rated by both users

How do I find someone similar

• Manhattan Distance

•Euclidean Distance

How do I find someone similar – N dim

• Minkowski

How do I find someone similar – N dim

How do I find someone similar – N dim

Problem – 1: entry matching

Problem – 2 : blame the users

Pearson’s Correlation Co-efficient

Pearson’s Correlation Co-efficient

Pearson’s Correlation Co-efficient

Pearson’s Correlation Co-efficient (1 to -1)

Pearson’s Correlation Co-efficient (1 to -1)

Approximation:

(x)

(y)

Pearson’s Correlation Co-efficient (1 to -1)

What to Use?

Comments on Neighbor-based Methods

• Here we emphasized user-user similarity

– Can also do this with item-item similarity, i.e.,

– Find similar items (across users) to the item we need a rating for

• Simple and intuitive

– Easy to provide the user with explanations of recommendations

• Computational+ Issues • In theory we need to calculate all n2 pairwise weights

• So scalability is an issue (e.g., real-time)

• Cold start – data sparsity

• For recent advances in neighbor-based approaches see Y. Koren, Factor in the neighbors: scalable and accurate collaborative filtering, ACM

Transactions on Knowledge Discovery in Data, 2010

33

Technology thrives simply because we create the

‘supply’ as opposed to the ‘demand’ first…

34

References

● Corporate presentations of Random company executives

● Lots of Google’ing and YouTube

● Wikipedia

● Programmers Guide to Data Mining

35

s