Presentation on Recommender Systems

Survey Papers

COURSE CO-ORDINATOR PRESENTED BY

Prof. Sanjog Ray Anuj Sharma
Information Systems Area, FPM Participant
IIM Indore, Information Systems Area
Rau-Pithampur road
Indore (M.P.)-453331
# 0731- 2439 524 (O)
Session 1 & 2: Presentation on Recommender
Systems Survey Papers

Compulsory Readings:
1. J. Ben Schafer, Joseph A. Konstan, John Riedl (1999): Recommender systems in
e‐commerce. ACM Conference on Electronic Commerce
2. 2. G. Adomavicius, A. Tuzhilin (2005): Toward the Next Generation of
Recommender Systems: A Survey of the State‐of‐the‐Art and Possible
Extensions. IEEE Transactions on Knowledge and Data Engineering 17(6): 734‐
749

Additional Readings:
3. S. S. Anand, B. Mobasher (2005): Intelligent techniques for web
personalization. In Intelligent Techniques for Web Personalization, pages 1–36.
Springer
Outline
• Introduction to Recommender Systems
– Data model, properties and types of Recommender Systems
– Issues of Recommender Systems

• The Survey of Recommender Systems

– Content-Based Methods
– Collaborative Methods
– Hybrid Methods
– Summary and Conclusions

• Extending Capabilities Of Recommender Systems

– Comprehensive Understanding of Users and Items
– Extensions for Model-Based Recommendation Techniques
– Multidimensionality of Recommendations
– Multcriteria Ratings
– Nonintrusiveness
– Flexibility
– Effectiveness of Recommendations

• Conclusions
Outline
• Intelligent Techniques for Web Personalization
• Classifications of Approaches to Personalization
– Individual Vs Collaborative
– Reactive Vs Proactive
– User Vs Item Information
– Memory Based Vs Model Based
– Client Side Vs Server Side
• Personalization Techniques
– Content-Based Filtering
– Traditional Collaborative Filtering
– Model Based Techniques
–Item-Based Collaborative Filtering
–Clustering Based Approaches
–Graph Theoretic Approaches
• Issues
– The Cold Start and Latency Problem
– Data Sparseness
– Scalability
– Privacy
– Recommendation List Diversity
– Adapting to User Context
– Using Domain Knowledge
– Managing the Dynamics in User Interests
– Robustness
– Trust
• Evaluation of Personalization Systems
 Introduction to Recommender Systems

• What are Recommender Systems

• Why are they useful
• What are different types of them
• Relation with information architecture
• Limitations and possible improvements
 Information Overload
News items,
TV programs,
Books, Journals,
Music CDs,
Research
Movie titles
papers

Consumer
Web pages,
products, e-
Usenet articles,
commerce
e-mails
items,
Recommender systems are a technological
proxy for a social process

Which one should I

read?

Recommendations Recommendations
from friends from Online
Systems
 Basic interaction paradigm of recommender
systems

Input (ratings of books):

What should I read “I recently enjoyed Snow Crash, Seabiscuit,
next? The Soloist, and Love in a Cold Climate”

Output (Recommendations):
“Books you might enjoy are…”
 Recommender Systems (RS)
• Solution to information overload.
• Content-based
– RSs find items similar to the ones you liked in
past
• Collaborative Filtering
– Users give ratings to items
– RS finds users similar to you (User similarity)
– Suggest you items liked by them
CF is Simple and effective, BUT ...
Everyday Examples of Recommendations

• Bestseller Book lists

• Top 40 music lists
• The “recent returns” shelf at the library
• Many weblogs
• Movie
• Friends on Social Network sites....

• Common insight: personal tastes are correlated:

– If Alice and Bob both like X and Alice likes Y then Bob is more likely to like
Y
– especially (perhaps) if Bob knows Alice or their demographics, personality
traits matches
 Recommender Systems
• Given a set of users and items
– Items could be documents, products, other users
…
• Recommend items to a user based on
– Past behavior of this and other users
• Who has viewed/bought/liked what?
– Additional information on users and items
• Both users and items can have known attributes [age,
genre, price, …]
 Definitions
• A recommender system is any system which provides a
recommendation/prediction/opinion to a user on items
– Rule-based systems use manual rules to do this

• An item similarity/clustering system uses item links to

recommend items like ones you like

• A classic collaborative filtering system uses the links between

users and items as the basis of recommendations

• Commonly one has hybrid systems which use all three kinds of
links in the previous picture
 What do RSs achieve?
• Help people make decisions
– Examples:
• Where to spend attention
• Where to spend money
• Help maintain awareness
– Examples:
• New products
• New information
 Recommender Systems
• RS – problem of information filtering
• RS – problem of machine learning
• Enhance user experience
– Assist users in finding information
– Reduce search and navigation time
• Increase productivity
• Increase credibility
• Mutually beneficial proposition
 Recommendation types
• User attributes-based Recommendation
– Male, 18-35: Recommend The Matrix
• Content Similarity
– You liked The Matrix: recommend The Matrix
Reloaded
• Collaborative Filtering
– People with interests like yours also liked Forrest
Gump
This title is a textbook-style exposition on the topic, with its
information organized very clearly into topics such as compression,
indexing, and so forth. In addition to diagrams and example text
transformations, the authors use "pseudo-code" to present algorithms
in a language-independent manner wherever possible. They also
supplement the reading with mg--their own implementation of the
techniques. The mg C language source code is freely available on the
Web.
Personalized
Recommendation
 Inputs to Recommender Systems
• Past transactions from users:
– which docs viewed
– content/attributes of documents
– which products purchased
– pages bookmarked
– explicit ratings (movies, books … )
• Current context:
– browsing history
– search(es) issued
• Explicit role/domain info:
– Role in an enterprise
– Document taxonomies
– Interest profiles
 Sample Applications
• Ecommerce
– Product recommendations - amazon
• Corporate Intranets
– Recommendation, finding domain experts, …
• Digital Libraries
– Finding pages/books people will like
• Medical Applications
– Matching patients to doctors, clinical trials, …
• Customer Relationship Management
– Matching customer problems to internal experts
 Recommender Data Model (I)
• Set C={c1, …, cn} of users
• Set S={s1, …, sm} of items (e.g. products)
• Elements from C and S can be described by a vector
respectively
– (a1, …, as)  attributes of user profile
– (b1, …, bt)  description of items (meta data, features, …)
• Goal of recommendation process: recommend new items for
an active user u
• Overview of process
1. User modelling (explicit or implicit, e.g. user rates items)
2. Personalization, generate list of recommended items
 Recommender Data Model (II)
• Let u be a utility function that measures the usefulness of
item s to user c, i.e., u : C X S → R, where R is a totally
ordered set
• Then, for each user c ε C, we want to choose such item s’ ε S
that maximizes the user’s utility. More formally-

• In recommender systems, the utility of an item is usually

represented by a rating, which indicates how a particular user
liked a particular item
 Recommender Data Model (III)
• Recommendation often based on ratings of an item
ij by a user uk:
– Rating rj,k: I  [0,1] ˘ ø
• Other range of values possible, e.g. {*, **, ***, ****, *****}
• ø := no rating for Item (or “0”)

• Example user-item matrix of ratings

[Ado./Tuz. (2005)]
 RS as a Intelligent Techniques for Web Personalization

• Classifications of Approaches to Personalization

– Individual Vs Collaborative
– Reactive Vs Proactive
– User Vs Item Information
– Memory Based Vs Model Based
– Client Side Vs Server Side
 Properties of Recommender Systems (I)
• Individual vs. collaborative
– Recommend item(s) for active user
– Collaborative: Take also information about other users into account
(often their ratings for items)
• Works on user-item matrix
– Individual: Based on user model for active user only
• I.e. user-item matrix is reduced to match user vector with item set
• Reactive vs. proactive
– Explicit interaction of user with system
• E.g. user query for search engine or dialogue system
– System generates recommendations even with explicit user
interaction
• Example: Amazon may send out email if a new item is available which is
similar to already bought one
 Properties of Recommender Systems (II)

• User- vs. Item-based

– Focus on ratings, interaction history of user, explicitly entered preferences etc.
– Focus on item descriptions (e.g. categories), related/similar items etc.

• Memory- vs. Model-based

– Interpretation of raw data, e.g. user-item matrix
– Generation of a dedicated model, e.g. item-item matrix or Bayesian Network

• Client- vs. Server-based

– Most systems are server-based so far, i.e. one centralized server stores all the
information (for example Amazon and most other online shops)
– Decentralized systems manage information about users and generate
recommendation client-side
The recommendation process from the user’s
perspective

User inputs Time and effort

to input
preferences…
Privacy concerns

…receives Time and

recommendations… effort to
review recs

…and decides if he/she will

sample recommendation

In the end, a user benefits only

if recommendations turn out to
be good ones.
 What Users Want

New to me
Fast Engaging

RECOMMENDATIONS
PROCESS

Easy Good

To succeed, collaborative filtering recommender systems

need a LOT of motivated regular users.
Issues of Recommender Systems in General

• Cold start and latency problems

• Sparseness of user-item matrix
• Diversity of recommendations
• Scalability
• Privacy and trust
• Robustness
• Utilization of domain knowledge
• Changing user interests (dynamics)
• Evaluation of recommender systems
 Cold Start Problems
• “New user“ and “new item“ problem
• Systems cannot recommend items to new users
with no profile or no interaction history
• Same for new items
– Also “latency problem“: items need some time until they can be
recommended
• Chicken-and-egg problem
– Users will not use system without good recommendations
– No incentive to rate items etc.
– System cannot generate good recommendations
• Possible solutions include explicit user profiling
methods to start interaction
 Data Sparseness
• Common situation
– Lots of users and items
– But only few ratings
 Sparseness of user-item matrix
 Recommender algorithms will not work very well

• In addition, new items are continously added

– Users should also rate these items
– Number of ratings has to keep up with new users and items

• Possible solution include the automatic generation

of ratings
– Implicit user profiling, use of transaction history of users, e.g. click on a video
constitutes a positive rating
 Diversity of Recommendations
• Focus usually on generating recommendations as
„good“ as possible
– But also important: new, unexpected items
– Do not recommend items that are already known
– Do not recommend items that are too similar to already known items
• E.g. user likes „Lord of the Rings 1“  user possibly also likes „Lord of the Rings 2“, but
is this really a useful recommendation?

• Possible solutions
– Use content-based approaches to easier integrate new
items in recommendation process
– Use collaborative filtering to allow „cross-domain“
recommendations
 Scalability
• Algorithms are based on matching users and items
– The more items and users, the higher the computational effort to analyze the data
• Storage/memory and runtime complexity
• Alternatively, the quality of recommendations suffer
– Scalability of recommender systems is an issue in practice

• Problem in particular with memory-based approaches

• Possible solutions include
– Use model-based approach
– Limit the number of items and/or users
• E.g. only consider items that received at least k ratings
– Precompute recommendations for users
• Will reduce runtime complexity to generate recommendations, but may increase strorage
requirements
 Privacy and Trust
• Collecting and interpreting personal data, e.g. ratings
– For example, bought items or visited product Web pages on Amazon
– Control for users?
• Bought product may have been gift for other person
– Privacy problem!
• Tradeoff with recommender quality
– The more information about the user the system is able to collect, the higher
the recommendation quality is in general
• Also trust, how can user trust the quality of a recommended item?
• Possible solutions include
– Consider social relationsships (“social recommender“, “Web of Trust“)
– Let user control their profile information
– Explanations of recommendations
• Why was an item recommended?
 Robustness
• Quality of (collaborative) recommenders depends on quality
of ratings
– Manipulation by users possible
• E.g. by automatic registration of a large number of “users“ and ratings
– Also called “shilling“, profile injection
– Attacks in principle
• “push“: Aim is to push item(s) by inserting a large number of good ratings
• “nuke“: Same with negative ratings
• Possible solutions include
– Make registration for service harder, e.g. request and check
personal information
– Detect attacks and remove corresponding users and ratings
– Adjust algorithms, some algorithms have proven to be more robust
 Utilization of Domain Knowledge
• Systems often regard items in isolation
– No relationships between items
– No domain knowledge
• Example: searching for (books or other products on) “baseball“
– Too many hits  restriction to “baseball technique“, or “baseball
player“, for example
• Based on user model and domain ontology
– Too few hits  broading to “sport“, for example
• Some approaches in current research literature utilize
Semantic Web technologies
– Build and maintain item ontologies
– Also for users
• E.g. “GUMO“ (General User Modellierung Ontology)
Changing User Interests (Dynamics)
• User model is often relatively static
• But dynamic evolution over user interests
– Changes over time, older ratings may not be valid any more
• Also the context of recommendations
– Example: Mobile restaurant guide
• Restaurant may be too far away from current position (location)
• Restaurant may be be closed today (time)
– A good rating for a restaurant after a dinner on a weekend may not be
relevant for recommending a restaurant for a quick lunch on a workday
• Solutions in research literature include
– E.g. explicit distinction between short- and long-term interests
– Context-aware recommender systems
 Evaluation of Recommender Systems
• Goal of personalization is to improve the
interaction of users with system
– May be subjective, hard to evaluate
• General method for recommender systems
– Let users rate recommended items and compare actual user ratings
with predicted rating
– Most important metrics
• “precision“: probability/rate that users did like recommended items
• “recall“: probability/rate that preferred items by users are recommended
– In addition user studies
• User evaluate system in questionnaire etc.
 Types of Recommender Systems
• Collaborative filtering (CF)
• Content-based filtering (CB)
– Individual recommender algorithms
– Also utility- or knowledge-based approaches
• Case-based recommendation
• Hybrid recommender systems
– Combination of several other recommenders
• Additional important variants
– Context-aware and multi-dimensional recommenders
– Decentralized recommender systems
– Recommending for groups
 Types of RS
Three broad types:

1. Content based RS
2. Collaborative RS
3. Hybrid RS
 Types of RS – Content based RS
Content based RS highlights
– Recommend items similar to those users
preferred in the past
– User profiling is the key
– Items/content usually denoted by keywords
– Matching “user preferences” with “item
characteristics” … works for textual information
– Vector Space Model widely used
 Content-based Recommender Systems

• Basic idea
– Match user profile (interests, ratings, click history, …) with item set
– Often: Systems recommend item which are similar to items that
the active preferred in the past
• Important formalisms
– Representation of items (item model)
• Often as a vector of features
• E.g. Vector Space Model (VSM) for documents
– Representation of users (user model)
• E.g. Rating vector, learned user preferences
– Metrics to match items and users
• Calculate similarity between vectors
Content-based Recommender Systems-Methods
and Variants
• Demograhic filtering, stereotypes
– Grundy example
• Bayesian networks
– Spam example
• Document modeling
– E.g. tf*idf
– Data mining/machine learning methods
– Classification, clustering of items, short example
– Decision trees
• User Customization
– User explicitely specifies interesting categories
• Rule-based systems
– Modeling expert knowledge or learning rules from user behaviour
• Knowledge based systems
 Advantages Content-based Filtering
• No (or less pronounced) „new item“ problem
• Usually good scalability
– Because most approaches are model-based
• Often no explicit profil acquistion needed
– Ratings not needed, transaction history sufficient
• Often no domain knowledge needed
– Item description sufficient
• Often quality of recommendation improves over time
– Better user model
 Disadvantages Content-based Filtering
• Item model limited to analyzed features
– E.g. keywords in document or points-of-interests relevant for mobile
applications
– Features have to be available explicitely
• Overfitting, portfolio effect
– Recommendation based on similarity only
– No real new, unexpected items (diversity often poor)
• Cold start: Often still “ new user“ problem
– But usually less pronounced than with collaborative recommenders
– Rule-system example
• No new user problem, user model is implicitly observed user location
• But rules description have to be provided (trigger)
Disadvantages Content-based Filtering
Content based RS - Limitations
– Not all content is well represented by keywords,
e.g. images
– Items represented by same set of features are
indistinguishable
– Overspecialization: unrated items not shown
– Users with thousands of purchases is a problem
– New user: No history available
– Shouldn’t show items that are too different, or
too similar
 Case-based Recommenders
• A form of content-based recommendation
• Structured information with a well defined set of
features and feature values
– Travel information presented in its price, duration,
accommodation, location, mode of transport, etc.
– Job information presented in the job kinds, salary, business
category of each company, educational level, experience,
location etc.
• Information is represented as cases and the system
recommends the cases that are most similar to the user’s
preferences
Wolfgang Wörndl 47
 Case-Based Reasoning
• Case-based recommendation origins in Case-Based
Reasoning (CBR)
– It is to solve new problems by reusing the solutions to
problems that have been previously solved and stored as
cases in a case-base
– Each case consists of a specification part, which describes
the problem and a solution part, which describes the
solution of the problem
• Solutions to similar prior problems are a useful starting point for
new problem solving
• “The users would like the similar one that they liked
before.”
Wolfgang Wörndl 48
 Simple Example of Case-based
Recommendation

Product #1
HD: 250 GB
I want a laptop with Memory: 2 GB
250GB HD, 1GB Screen Size: 15 inch
Price: $550
memory and 14 inch
screen for $400 Product #2
HD: 150 GB
Memory: 1 GB
Screen Size: 15 inch
Price: $450

Product #3
HD: 250 GB
Memory: 1 GB
Screen Size: 14.2 inch
Price: $500
Case-based Recommendation
 Types of RS – Collaborative RS
Collaborative RS highlights
– Use other users recommendations (ratings)
to judge item’s utility
– Key is to find users/user groups whose
interests match with the current user
– Vector Space model widely used (directions
of vectors are user specified ratings)
– More users, more ratings: better results
– Can account for items dissimilar to the ones
seen in the past too
– Example: Movielens.org
 Types of RS – Collaborative RS
Collaborative RS - Limitations
– Different users might use different scales. Possible
solution: weighted ratings, i.e. deviations from
average rating
– Finding similar users/user groups isn’t very easy
– New user: No preferences available
– New item: No ratings available
– Demographic filtering is required
– Multi-criteria ratings is required
 Collaborative Filtering (CF)
• Basic idea: System recommends items which
were preferred by similar users in the past
– Based on ratings
• Express preferences of the active user
• And also other users  Collaborative approach
– Works on user-item matrix
• Memory- or model-based
• No item meta data etc.!
• Assumption: Similar taste in the past implies similar taste
in future
• CF is formalization of “word of mouth“ among
buddies
 General Process
1. Users rate items
2. Find set S of users which have rated similar to
the active user u in the past ( neighborhood)
 Similarity calculation
 Select the k nearest users to the active user

3. Generate candidate items for recommendation

 Items which were rated in neighborhood of u,
 but were not rated by u yet

4. Predict rating of u for candidate items

 Select and display n best items
 Collaborative Filtering

Profile of Profile of
user 1 user 2

Profile of Profile of
current user 3
user
Profile of
user 4
Documents from
like-minded users’
Profile of
profiles
user 5
=> recommended
documents
Example (I)

Source: http://www.dfki.de/~jameson/ijcai03-tutorial/
Example (II)
Example (III)
 Required Metrics
• Metric for user-user similarity
– Mean-squared difference
– Cosine
– Pearson/Spearman correlation

• Select set S of most similar users (to active

user u)
– Similary threshold
– Aggregate neighborhood
– Center-based

• Metric to predict the rating of u for an item i

 Required Metrics
• Metric for user-user similarity
– Mean-squared difference
– Cosine
– Pearson/Spearman correlation

• Select set S of most similar users (to active

user u)
– Similary threshold
– Aggregate neighborhood
– Center-based

• Metric to predict the rating of u for an item i

 User-User Similarity
• Item set I
• Users U,V with u[i] denoting rating of item i by user u
• denotes the rating vector of user u
• denotes the vector norm

• Mean squared difference:

• Cosine similarity:
(Henze, 2006/7)
Example Calculation
 Pearson/Spearman Correlation
• Average rating is taken into account
– is vector of average ratings
• Not suitable for unary ratings
– Unary: Item is marked (or not)
• z.B. “Product was purchased“
– Binary: “good/bad“, “+/-“ etc.
– Scalar: Numerical rating (e.g. 1-5) etc.
– Consider only items which were rated by both users
 Required Metrics
• Metric for user-user similarity
– Mean-squared difference
– Cosine
– Pearson/Spearman correlation

• Select set S of most similar users (to active

user u)
– Similary threshold
– Aggregate neighborhood
– Center-based

• Metric to predict the rating of u for an item i

 Neighborhood of Similar Users
• Goal: Determine set S of users which are most similar
to the active user u
• Center-based
– S contains k most similar users
• Problem: maybe some of the users are not really that similar, if k was chosen too large,
deviators possible

• Similarity threshold
– S contains all users with a similarity bigger than a threshold t
• Problem: maybe too few users in S

• Aggregate neighborhood
– Follow similarity threshold method first
– If S is too small (less than k users)
• Determine „centroid“ of set S and add users which are most similar to centroid ( less
deviators than center-based method)
 Neighborhood of Similar Users
• Goal: Determine set S of users which are most similar
to the active user u
• Center-based
– S contains k most similar users
• Problem: maybe some of the users are not really that similar, if k was chosen too large,
deviators possible

• Similarity threshold
– S contains all users with a similarity bigger than a threshold t
• Problem: maybe too few users in S

• Aggregate neighborhood
– Follow similarity threshold method first
– If S is too small (less than k users)
• Determine „centroid“ of set S and add users which are most similar to centroid ( less
deviators than center-based method)
 Required Metrics
• Metric for user-user similarity
– Mean-squared difference
– Cosine
– Pearson/Spearman correlation

• Select set S of most similar users (to active

user u)
– Similary threshold
– Aggregate neighborhood
– Center-based

• Metric to predict the rating of u for an item i

 CF Recommender (I)
• Given
– Set S with most similar users to u
– s[i] rating of a user (from S) from an item i
• Goal: Predict the rating of u for i
• Easiest option: Arithmetic mean

• Problems
– Similarity of u with members of S is not taken into account
• Solution: Weighting based on similarity
 CF Recommender (II)

– Different users utilize rating scale differently

• Solution: Consider deviation from average rating (for user)

• Note
– Many variations of algorithms in research literature
• For various application domains, with different properties
 Collaborative Filtering
• Amazon and other commercial service use
some form of collaborative filtering
– Exact method usually not published

• Non-commercial example with published

algorithms: http://www.movielens.umn.edu
Advantages- Collaborative Filtering
• Works well in practice
– Quality of recommendations improves with more and more ratings
• Only ratings as input data required, no other data
– In particular, no information (meta data, desciption) about items needed
• CF is able to generate cross-domain („cross genre“)
recommendations  high diversity
– Because item categories etc. are not considered
– Has proven useful in practice
• Quality of recommendation improves with more and more ratings
• Implicit user feedback often adequate
– Unary ratings, e.g. rating = Click on product Web page
 Disadvantages- Collaborative Filtering
• New user and new item problem
– Serious issue in practice
• Often sparseness in user-item matrix
– Algorithms generate worse results with too few ratings
• “Grey sheep“ problem
– Does not work very well for users with “extraordinary“ taste
• Because similar users are not available
– Also “black sheeps“, users that intentionally make incorrect ratings
• CF is prone to manipulation
• Trust and robustness are issues
 Memory- vs. Model-based CF
• Discussion so far
– “ Standard“ CF, operates on user-item matrix
• Thus the raw data of ratings  memory-based approach
– User-user similarity
– Other option: item-item similarity
• I.e. works on columns instead of rows in matrix
• Still collaborative apporach, not content-based, because
it utilizes ratings only
• Model-based CF approaches
– Use user-item matrix to generate a model
– Most common option: Model-based item-item approach
 Model-based Collaborative Filtering
• Process model-based item-item approach
1. Calculate similarity of items
• Result is an item-item matrix with dimension “number of items“
2. Generate recommendation based on ratings of active user and item-
item matrix
• Other options for model include
– Bayesian networks
• Utilize conditional probabilities, “ Email spam“ example
– Covered earlier in course
– Clustering methods, classification of users based on ratings
Discussion Model-Based Approach
• Advantages
– Model and also predictions can be precomputed
• Reduces runtime complexity of recommendation generation significantly
• Scalability important in practice
– Fewer ratings required
• Memory-based CF needs a higher number of ratings to calculate user-
user similarity with accuracy
• Disadvantages
– Storage requirements, item-item matrix grows very big with
many items,
• User-item matrix is usually pretty sparse  efficient storage possible
– Result also often similar to items which the user already knows
• Memory based user-user approach often results in better
recommendations with higher diversity
 Hybrid Recommender Systems
• All recommenders have distinguished pros and cons
• Idea: Combine various methods to avoid disadvantages of
single techniques
– Example: Combine CF with content-based approach to avoid new
user/item problems
• Combine different types
– It is also possible to combine different algorithms of the same type,
e.g. content-based
• Hybrid recommender: Combination of at least two other
recommender algorithms
• Different alternatives for the combination
– Types of hybrid recommenders
 Types of Hybrid Recommender Systems
• Weighted: Score of different recommendation components are combined
(numerically)
• Switching: System chooses among recommendation components and applies
the selected one
• Mixed: Recommendations from different recommenders are presented
together
• Feature Combination (FC): Features derived from different knowledge sources
are combined together and given to a single recommendation algorithm
• Feature Augmentation(FA): One recommendation technique is used to
compute a feature or set of features, which is then part of the input to the
next technique
• Cascade: Recommenders are given strict priority, with the lower priority ones
breaking ties in the scoring of the higher ones
• Meta-level: One recommendation technique is applied and produces some
sort of model, which is then the input used by the next technique
 Other Variations of RS
Cluster Models
– Create clusters or groups
– Put a customer into a category
– Classification simplifies the task of user
matching
– More scalability and performance
– Lesser accuracy than normal collaborative
filtering method
 Other Variations of RS
Algorithm for Amazon’s item to item collaborative
filtering

For each item in product catalog, I1

For each customer C who purchased I1
For each item I2 purchased by customer C
Record that a customer purchased I1
and I2
For each item I2
Compute the similarity between I1 and I2

Similarity between two items depends on number of

customers who bought them both
 Other Variations of RS
Knowledge based RS
– Use knowledge of users and items
– Conversational Interaction used to establish current
user preferences
– i.e. “more like this”, “less like that”, “none of those” …
– No user profiles maintained, preferences drawn
through manual interaction
– Query by example … tweaking the source example to
fetch results
 Some general considerations in RS
Difficult to Set Up
– Lot of development required for setup
– Moving to RS takes time, energy and long-term
commitment
They could be wrong
– RS not just a technical challenge, but also a social
challenge
– Amazon took some heat when it started cross-
promoting its new Clothing site by recommending clean
underwear to people who were shopping for DVD
Maintenance
 Some general considerations in RS

• Context is important in “user X items”

space
• Similarity is a non-uniform concept, is
highly contextual and task-oriented
• Users sometimes need motivation to
rate items
 Possible Improvement in RS
Better understanding of users and items
– Social network (social RS)

1. User level
– Highlighting interests, hobbies, and keywords
people have in common
2. Item level
– link the keywords to eCommerce (by RS algorithms)
 Possible Improvement in RS
System transparency
– Help users understand how the RS works
– Example:
http://www.pandora.com/
Amazon.com

Result:
– Generate trust
– Convince users
 Possible Improvement in RS
Multidimensionality of Recommendations
– Take into consideration the contextual
information
Examples:
Movie- Different context, Different rating
Travel
 Possible Improvement in RS
Randomness and Nonintrusiveness
• Many recommender systems are intrusive in the sense
that they require explicit feedback from the user and
often at a significant level of user involvement
• some recommender systems use nonintrusive rating
determination methods where certain proxies are used
to estimate real ratings
• However, nonintrusive ratings (such as time spent
reading an article) are often inaccurate and cannot fully
replace explicit ratings provided by the user
 Possible Improvement in RS
Other
– Privacy (CF methods)
One-way hash: easily computed one direction,
impossible in the other
– Malicious use (recommendation spam)
Probabilistic techniques to determine the
honesty of a score (unusual pattern)
 Possible Improvement in RS
Common business models adapted:
– Charge recipient of recommendations
– Provide incentives for giving ratings
– Targeted advertisements
– Charge owners of the items
 Possible Improvement in RS
Complicated Problems
– People might change minds afterwards

Study: The variations of an individual’s own

opinion
Conclusion-Relevance to information architecture

• Increase find ability

• Reduce searching efforts
• Improve organizational systems
• Enhance browsing
• Provide more useful “local navigation” options
• “Targeted Advertising” a much better
substitute to common advertisements that are
often irrelevant
 Conclusion
• Data model properties, types and issues of
recommender systems
– Different types of recommender systems have different
advantages and disadvantages
• Content based methods
• Collaborative Filtering
– Basic idea, process in general
• Metrics for user-user similarities
• Determine neighborhood of similar users
• Predict ratings for active user
– Model-based approach, discussion
 References
• G. Adomavicius and A. Tuzhilin: “Toward the Next Generation of
Recommender Systems: A Survey of the State-of-the-Art and Possible
Extensions“, IEEE Transactions on Information Systems, Vol. 17, No. 6,
2005, pp. 734-749
• http://www.dfki.de/~jameson/ijcai03-tutorial/
• S. S. Anand, B. Mobasher (2005): Intelligent techniques for web
personalization. In Intelligent Techniques for Web Personalization, pages
1–36. Springer

• Course „Personalisierung und Benutzermodellierung“, Nicola Henze, SS

2006/SS 2007, Uni Hannover
• J. Ben Schafer, Dan Frankowski, Jon Herlocker and Shilad Sen:
„Collaborative Filtering Recommender Systems”, chapter 9 in [*]
[*] P.Brusilovsky, A.Kobsa, W.Nejdl (eds.): The Adaptive Web, Springer-
Verlag, Berlin/Heidelberg, 2007