Tagommenders: Connecting Users to Items through Tags

1
Tagommenders Connecting Users to Items through
Tags

• Written by Shilad Sen, Jesse Vig, and John Riedl
  (2009)
•  
• Presented by Ken Hu and Hassan Hattab

2
Overview

• Introduction
• Tagommender
• Philosophy
• Dataset
• Tag Preference Inference
• Approach
• Methods

• Recommenders
• Implicit
• Explicit
• Results
• Conclusion

3
First, Recommenders. 

• What is Recommender system?
• Two Main tasks
• Recommend.
• Predict.

4
Recommender Systems 

• Types of recommender systems
• User-based decides according to the user's
  previous choices
• Item-based decides according to related items to
  a selected item
• SVD
• Problem These methods don't consider the content
  of the item. 
• Solution Content-based Recommenders 

5
Overview

• Introduction
• Tagommender
• Philosophy
• Dataset
• Tag Preference Inference
• Approach
• Methods

• Recommenders
• Implicit
• Explicit
• Results
• Conclusion

6
Tagging Systems   

• Uses tags to address (categorize) items to users
• Tags are created by general users (More
  meaningful )

7
Tagommenders 

• Basically, they combine Recommenders
  (content-based) and tagging systems. 
• Two main parts for Tagommenders
• They infer users preferences for tags based on
  their interactions with tags and movies
• and they infer users preferences for movies
  based on their preferences for tags.

8
Tagommender's data set

• These are collected from the MovieLens website.
• Movie Rating 
• Movie clicks
• Tag applications
• Tag Searches 
•  Tag Preference Ratings

9
Tagommender's data set

•  

10
Overview

• Introduction
• Tagommender
• Philosophy
• Dataset
• Tag Preference Inference
• Approach
• Methods

• Recommenders
• Implicit
• Explicit
• Results
• Conclusion

11
Tagommender's  Cycle

•  

12
Inferring Tag Preference

• Inferring Preference using Tag Signals (Direct) 

13
Inferring Tag Preference

• Inferring Preference using Item Signals
  (indirect) 

14
Inferring Preference using Item Signals

• Sigmoid transformation is used to calculate the
  weight of movie m to tag t

15
Inferring Preference using Item SignalsMethods

• Movie-Clicks
• Movie-log-odds-clicks
• Movie-r-Clicks
• Movie-r-log-odds-clicks
• Movie-Rating
• Movie Bayes

16
1- Movie-Clicks

•  

set of movies clicked by user u
17
2- Movie-log-odds-clicks

•  

18
3- Movie-r-Clicks4- Movie-r-log-odds-clicks

• The only difference is Movie-rating is counted
  rather than movie clicks 

19
5- Movie-Rating

• A users preference for a tag is the average
  rating for a movie under that tag. 

user u's rating for movie m
20
6- Movie-bayes

• A bayesian generative model for users rating for
  a certain tag.
• if the tag is relevant to a rating then the
  rating will be chosen from the user-tag-specific
  distribution
• Else, it will be chosen from the user background
  rating  distribution

21
Which one is better?

•  

22
Overview

• Introduction
• Tagommender
• Philosophy
• Dataset
• Tag Preference Inference
• Approach
• Methods

• Recommenders
• Implicit
• Explicit
• Results
• Conclusion

23
Recommenders

• Implicit
• Tag data only
• Recommend only
• 2 algorithms
• Implicit-tag
• Implicit-tag-pop

• Explicit Algorithms
• Use users' movie ratings
• Recommend and predict
• 3 algorithms
• Cosine-tag
• Linear-tag
• Regress-tag

24
Implicit Implicit-tag

• Vector Space Model
• Inferred tag preference
• Relevance weight

25
Implicit Implicit-tag-pop

• Implicit-tag with movie popularity
•  
•  
•  
•  
•  
•  
•  
•  
• Tag gt clicks, clicker count gt click count
• Linear estimation of log function

26
Recommenders

• Implicit
• Tag data only
• Recommend only
• 2 algorithms
• Implicit-tag
• Implicit-tag-pop

• Explicit Algorithms
• Use users' movie ratings
• Recommend and predict
• 3 algorithms
• Cosine-tag
• Linear-tag
• Regress-tag

27
Explicit Cosine-tag

• Cosine similarity rating vs tag preference

28
Explicit Linear-tag

• Least-square fit linear regression

29
Explicit Regress-tag

• Linear-tag with similarity between tags 
•  
•  
•  
• SVM was best to estimate h
• Robustness against overfitting

30
Overview

• Introduction
• Tagommender
• Philosophy
• Dataset
• Tag Preference Inference
• Approach
• Methods

• Recommenders
• Implicit
• Explicit
• Results
• Conclusion

31
Results Background

• Comparisons
• Top-5
• Compare top five recommendations
• MAE
• Average error of prediction

• Competitors
• Overall-avg
• User-avg
• User-movie-avg
• Explicit-item
• Implicit-item
• Funk-svd
• Hybrid
• Regress-tag funk-svd

32
Results Top-5
33
Results MAE
34
Overview

• Introduction
• Tagommender
• Philosophy
• Dataset
• Tag Preference Inference
• Approach
• Methods

• Recommenders
• Implicit
• Explicit
• Results
• Conclusion

35
Conclusion

• Introduced recommender algorithms based on user
  suggested tags (Tagommenders)
• Best at recommendation tasks
• Adds value at prediction tasks
• Hybrid predictors does very well
• Other advantages
• Ease to explain
• Algorithmic evaluation of tag quality

36
Questions?

•