8th ACM Conference on Recommender Systems

Foster City, Silicon Valley, California, USA, 6th-10th October 2014

Cross-Domain Recommender Systems

Iván Cantador1, Paolo Cremonesi2
1 Universidad Autónoma de Madrid, Madrid, Spain
[email protected]
2 Politecnico di Milano, Milan, Italy

[email protected]
 2nd ed. of the RSs Handbook
Cross-Domain Recommender Systems

Iván Cantador Ignacio Fernández-Tobıas Shlomo Berkovsky Paolo Cremonesi

2
 Tutorial
Cross-Domain Recommender Systems

Iván Cantador Paolo Cremonesi

3
 Today
Cross-Domain Recommender Systems

Paolo Cremonesi

4
 Recommendations for single domains

• Traditional recommender systems suggest items

belonging to a single domain
• movies in Netflix
• songs in Last.fm
• …
• This is not perceived as a limitation, but as a
focus on a certain market

5
 User profiles in multiple systems
Nowadays, users…
• provide feedback for items of different types
• e.g., in Amazon we can rate books, DVDs, …
• express their opinions on different social media
and different providers
• e.g., Facebook, Twitter, Amazon, Netflix, TripAdvisor

Nowadays providers wish to …

• cross-sell products and services
• provide recommendations to new users
6
Recommendations for multiple domains

Can we leverage all the available personal data

provided in distinct domains to generate better
recommendations?

definition of “domain”
definition of “better recommendations”

7
 Problems related to Cross-Domain RSs
• Machine Learning
• Multi-Task Learning / Transfer Learning
• User Modeling
• aggregation user preferences for cross system
personalization, targeted adv., security
• Context Aware recommender
• different domains as different context
• Hybrid recommender (Ensemble learning)
• AdaBoost → Hybrid
• Bootstrap / Blending → Cross Domain
8
 History of Cross-Domain RSs
• 2002: the term “cross-domain recommenders”
appear for the first time in a patent:
• Triplehop Technologies (now Oracle)
• 2005: some papers suggest “cross-domain” as an
interesting topic
• Mark van Setten, Sean M. McNee, Joseph A. Konstan.
2005
• Shlomo Berkovsky, Tsvi Kuflik, Francesco Ricci. 2005
• 2007: first papers with contributions on “cross-
domain”
• Ronald Chung, David Sundaram, Ananth Srinivasan. 2007
• Shlomo Berkovsky, Tsvi Kuflik, Francesco Ricci. 2007
• Ronald Chung, David Sundaram, Ananth Srinivasan. 2007
9
 History of Cross-Domain RSs
• First papers trying to classify problems and
approaches
• Antonis Loizou. 2009
• Sinno Jialin Pan, and Qiang Yang. 2010
• Bin Li. 2011
• Paolo Cremonesi, Antonio Tripodi, and Roberto
Turrin. 2011
• Fernández-Tobías, Ignacio, Iván Cantador, Marius
Kaminskas, and Francesco Ricci. 2012

10
 Cross-domain recommendations
• Single-Domain: Treat each domain
independently
• Collective-Domain: Merge domains an treat
them as a single domain baseline

• Cross-Domain: Transfer knowledge from source

to target
• assumption: information overlap between users
and/or items across different domains
- overlaps of users, items, attributes, …
11
 Goal of this tutorial

• Taxonomy of problems and techniques

• Literature overview: who is doing what
• Guidelines, based on consolidated as well as
state-of-the-art best practices from the research
community

12
 Contents

1. The cross-domain recommendation problem

2. Cross-domain recommendation techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation

13
Definition of the cross-domain problem

• Domains
• which types of domains exist?
• Goals
• why do we need cross-domain recommenders?
• Tasks
• which parts of the datasets are used?
• Scenarios
• which overlap of information exists between
domains?

14
 Contents

1. The cross-domain recommendation problem

• Definition of domain
• Cross-domain recommendation goals and tasks
• Cross-domain recommendation scenarios
2. Cross-domain recommendation techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation
15
 Definition of domain

• A domain is a particular field of thought, activity,

or interest
• In the literature researchers have considered
distinct notions of domain:
• movies vs. books
• action movies vs. comedy movies
• ...

16
 Definition of domain

• Domains differ because of

• different types of items
- Movies vs. Books
• different types of users
- pay-per-view users vs. users with yearly subscription
• “partition” of users with respect to items
- e.g., users with ratings on
– Books only
– Movies only
– Books and Movies

17
Definition of domain

We focus on two domains

source DS ↔ target DT
(auxiliary)

18
 Domain levels

• Attribute level (Comedy ↔ Thriller)

• same type of items, different values of certain attribute
• Type level (Movies ↔ Books)
• similar types, sharing some attributes
• Item level (Movies ↔ Restaurants)
• distint types, differing in most, if not all attributes
• System level (Netflix ↔ Movielens)
• almost the same items, collected in different ways
and/or from different operators

19
 Domain levels in the literature…

• Attribute level (Comedy ↔ Thriller): 12%

• Movie genres (Movielens, Eachmovie)
• Type level (Movies ↔ Books): 9%
• Amazon
• Item level (Movies ↔ Music): 55%
• Movielens, Last.fm, Delicious, BookCrossing, Facebook
• System level (Netflix ↔ Movielens): 24%
• Last.fm, Delicious

20
 Contents

1. The cross-domain recommendation problem

• Definition of domain
• Cross-domain recommendation goals and tasks
• Cross-domain recommendation scenarios
2. Cross-domain recommendation techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation
21
 Cross-domain recommendation goals
• Addressing the cold-start problem
• recommending to new users
• cross-selling of products
• Improving accuracy
• e.g., by reducing sparsity
• Offering added value to recommendations
• diversity, novelty, serendipity
• Enhancing user models
• discovering new user preferences
• vulnerability in social networks
22
Goals in the literature …

Goal Percentage
Cold start 5%
New user 15%
New item 5%
Accuracy 55%
Diversity 5%
Privacy 5%
User model 10%

23
 Cross-domain recommendation tasks
IS IT IS IT IS IT

S S S
UT
UT

UT
T US
T T
US

US
Multi-domain Linked-domain Cross-domain

= data from source domain

= data from target domain
= target of recommendations 24
 Multi-domain recommendation task
IS IT IS IT IS IT

S S S
UT
UT

UT
T US
T T
US

US
Multi-domain Linked-domain Cross-domain
• Recommend items in both source and target domains
• Goal: cross-selling, improve diversity, novelty,
serendipity
• Approach: sharing knowledge and linking domains
25
 Linked-domain recommendation task
IS IT IS IT IS IT

S S S
UT
UT

UT
T US
T T
US

US
Multi-domain Linked-domain Cross-domain
• Recommend target items to users in the target domains
• Goal: improve accuracy of recommendations in the
target domain (e.g., reduce sparsity)
• Approach: all
26
 Cross-domain recommendation task
IS IT IS IT IS IT

S S S
UT
UT

UT
T US
T T
US

US
Multi-domain Linked-domain Cross-domain
• Recommend items in the target domain to users in the
source domain
• Goal: solve cold-start, new users and new item probl.
• Approach: aggregating knowledge
27
Tasks in the literature …

Task Multi-domain
Multi-domain 20%
Linked-domain 55%
Cross-domain 25%

28
 Contents

1. The cross-domain recommendation problem

• Definition of domain
• Cross-domain recommendation goals and tasks
• Cross-domain recommendation scenarios
2. Cross-domain recommendation techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation
29
 Notation

XU → set of characteristics used to represent users

= FU → side information (features) about users
(e.g., demographics, tags, friends, …)
= I → items rated by users

XI → set of characteristics used to represent items

= FI → side information (features) about items
(e.g., genres, keywords, …)
= U → users rating items
30
 Linking domains
Overlapping attributes of …

… users:
• FU(S) ∩ FU(T) ≠
• e.g., we have demographics of users in both domains

… items:
• FI(S) ∩ FI(T) ≠
• e.g., items share the same set of attributes in both
domains
31
 Linking domains
Mapping attributes of …

… users:
• f : XU(S) → XU(T)
• e.g., friends of …

… items:
• f : XI(S) → XI(T)
• e.g., “vampire” in source and “zombie” in target are
both “horror”
32
 Linking domains
Overlap of …

… items:
• I(S) ∩ I(T) ≠
• e.g., we have same common items between domains

… users:
• U(S) ∩ U(T) ≠
• e.g., we have same common users between domains

33
Cross-domain recommendation scenarios (I)

35
 Contents

1. The cross-domain recommendation problem

2. Cross-domain recommendation techniques
• Categorizations of techniques
• Linking/aggregating knowledge techniques
• Sharing/transferring knowledge techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation
37
Cross-Domain: opportunity or problem?
• The source domain is a potential source of bias
• If the source domain is richer than the target
domain, algorithms learn how to recommend items
in the source domain and consider the target
domain as noise
• The source domain is a potential source of noise
• If the user models in the two domains differ, the
source domain introduce noise in the learning of the
target domain

38
Cross-Domain: opportunity or problem?

is a matter of weights

which is the relative “weight” of the two domains?

how much do we “weight” the information coming

from the source domain?

39
 Different approaches (I)
• Two types of cross-domain approaches, based on
how knowledge from the source domain is exploited

knowledge
linkage/transfer
Source Target Source Target
domain domain domain  domain
knowledge
aggregation

+
target domain target domain
recommendations recommendations
Linking/Aggregating knowledge Sharing/Transferring knowledge

40
 Different approaches (I)
• Two types of cross-domain approaches, based on
how knowledge from the source domain is exploited

knowledge
linkage/transfer
Source Target Source Target
domain domain domain  domain
knowledge
aggregation

+
target domain target domain
recommendations recommendations
Linking/Aggregating knowledge Sharing/Transferring knowledge

Fuzzy !!! 41
 Different approaches (II)

• Linking/Aggregating knowledge
• Merging user preferences
• Mediating user modeling data
• Combining recommendations
• Linking domains
• Sharing/Transferring knowledge
• Sharing latent features
• Transferring rating patterns

42
 Contents

1. The cross-domain recommendation problem

2. Cross-domain recommendation techniques
• Categorizations of techniques
• Linking/aggregating knowledge techniques
• Sharing/transferring knowledge techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation
43
 Proposed categorization

• Linking/aggregating knowledge
• Merging user preferences
• Mediating user modeling data
• Combining recommendations
• Linking domains
• Sharing/transferring knowledge
• Sharing latent features
• Transferring rating patterns

44
 Merging user preferences (I)
• Aggregate user preferences
• ratings, tags, transaction logs, click-through data

IS IT

U DS DT

+ user preference
aggregation

active
user recsysST
target domain
recommendations
45
 Merging user preferences (II)

• Pros:
• work well for the new-user problem
• robust (evolution of standard SD techniques)
• facilitate explanation
• Cons:
• need user-overlap between the source and target
domains

46
 Merging user preferences: approaches
• On the aggregated matrix we can apply
“weighted” single-domain techniques
• User-based kNN
- Berkovsky et al. 2007; Shapira et al. 2013; Winoto & Tang
2008;
• Graph-based
- Nakatsuji et al. 2010; Cremonesi et al. 2011; Tiroshi et al.
2013
• Matrix Factorization / Factorization Machine
- Loni et al. 2014

47
 Proposed categorization

• Linking/aggregating knowledge
• Merging user preferences
• Mediating user modeling data
• Combining recommendations
• Linking domains
• Sharing/transferring knowledge
• Sharing latent features
• Transferring rating patterns

48
 Mediating user modeling data (I)
• Aggregate models (CF, CBF, Hybrid) from different
domains
• user similarities, user neighborhoods
IS IT

U DS DT

active
user
recsysS

recsysT
target domain
user neighborhood recommendations
(user similarities) 49
 Mediating user modeling data (II)

• Pros:
• suited to the new-user problem and accuracy goals
• robust (evolution of standard SD techniques)
• Cons:
• need of either user- or item-overlap between the
source and target domains

50
 Mediating modeling data: approaches

• Aggregating collaborative or content similarities

• Berkovsky et al. 2007; Shapira et al. 2013; Shlomo
Berkovsky, Tsvi Kuflik, and Francesco Ricci. 2008.
• Aggregating user neighborhoods
• Berkovsky et al. 2007; Tiroshi & Kuflik 2012; Shapira
et al. 2013
• Aggregating latent features
• Low et al. 2011

51
 Proposed categorization

• Linking/Aggregating knowledge
• Merging user preferences
• Mediating user modeling data
• Combining recommendations
• Linking domains
• Sharing/transferring knowledge
• Sharing latent features
• Transferring rating patterns

52
 Combining recommendations (I)
• Aggregate single-domain recommendations
• ratings, ranking, probability distributions

IS IT

U DS DT

active
user
recsysS

recsysT
target domain
user neighborhood recommendations
(user similarities)
53
 Combining recommendations (II)

• Pros:
• easy to implement
• independent of the stand alone recommenders
• increase diversity
• independent of context
• Cons:
• need overlap of users
• difficult to tune weights assigned to
recommendations coming from different domains

54
 Combining recommendations:
approaches

• Aggregating estimated values of ratings

(blanding)
• Berkovsky et al. 2007; Givon & Lavrenko 2009
• Combining estimations of rating distribution
• Zhuang et al. 2010

55
 Proposed categorization

• Linking/aggregating knowledge
• Merging user preferences
• Mediating user modeling data
• Combining recommendations
• Linking domains
• Sharing/transferring knowledge
• Sharing latent features
• Transferring rating patterns

56
 Linking domains (I)
• Linking domains by a common knowledge
• item attributes, user attributes, association rules,
semantic networks,
IS IT

U DS DT

recsysT
target domain
DS DT recommendations
multi-domain active
semantic network user

57
 Linking domains (II)

• Pros:
• no need of user or item overlap
• bland with other technique
• Cons:
• difficult to generalize
• designed for particular cross-domain scenarios

58
 Linking domains: approaches
• Overlap of user/item attributes
• Chung et al. 2007
• Overlap of social tags
• Szomszor et al. 2008; Abel et al. 2011; Abel et al. 2013;
Fernández-Tobias et al. 2013
• Overlap of text (BoW)
• Berkovsky et al. 2006
• Semantic networks
• Loizou 2009; Fernández-Tobias et al. 2011; Kaminskas et
al. 2013
• Knowledge-based rules
• Azak et al. 2010; Cantador et al. 2013
59
 Contents

1. The cross-domain recommendation problem

2. Cross-domain recommendation techniques
• Categorizations of techniques
• Linking/aggregating knowledge techniques
• Sharing/transferring knowledge techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation
60
 Proposed categorization

• Linking/aggregating knowledge
• Merging user preferences
• Mediating user modeling data
• Combining recommendations
• Linking domains
• Sharing/transferring knowledge
• Sharing latent features
• Transferring rating patterns

61
 Sharing latent features (I)
• source and target domains are related by means
of shared latent features
IS IT

U DS DT

common
latent features

= x x B = x x B
DS A DT A

target domain
recommendations
active
user recsysT
62
 Sharing latent features (II)

• Pros:
• work well to reduce sparsity and increase accuracy
for both source and target domains
• Cons:
• computationally expensive
• need overlap of users and/or items

63
 Sharing latent features: approaches (I)

• Tri-matrix co-factorization: user and item factors

(U and V) are shared between domains; rating
patterns (B) are different
• Pan et al. 2010 and 2011

RS = U×BS×V’
RT = U×BT×V’

64
Sharing latent features: approaches (II)

• Tensor-based factorization (domain as a context)

• Hu et al. 2013

S T → = × B × C
users

R A
items

65
 Proposed categorization

• Linking/aggregating knowledge
• Merging user preferences
• Mediating user modeling data
• Combining recommendations
• Linking domains
• Sharing/transferring knowledge
• Sharing latent features
• Transferring rating patterns

66
 Transferring rating patterns (I)
• rating patterns are transferred between domains

IS IT

U DS DT

rating
patterns
x RS x RS x RS x
DS A = B DT = A B

target domain
recommendations
active
user recsysT
67
 Transferring rating patterns (II)

• Pros:
• apparently, no need of user or item overlap
• Cons:
• computationally expensive

68
Transferring rating patterns: approaches

• Code-Book-Transfer (CBT): Transferring cluster-

level rating patterns
• Li et al. 2009; Moreno et al. 2012; Gao et al. 2013

RS = US×B×VS’
RT = UT×B×VT’

69
 Transferring rating patterns: CBT

AUX
KNOWLEDGE 1. Extraction of knowledge
TRANSFER 1 EXTRACTION (codebook B) from
auxiliary domain
CODE
BOOK 2. Injection of knowledge in
0

target domain to reduce

TGT INJECTION
2
sparsity
RECOMMENDATION
3. Recommendation in
3
target domain with user-
based kNN
70
 Transferring rating patterns: CBT

AUX
KNOWLEDGE
EXTRACTION
TRANSFER 1

CODE
BOOK
0

TGT 2 INJECTION

3 RECOMMENDATION

71
 Transferring rating patterns: CBT

AUX
KNOWLEDGE
EXTRACTION
TRANSFER 1

CODE Recommendation in
BOOK
0
target domain with
TGT 2 INJECTION user-based kNN

3 RECOMMENDATION

72
Transferring rating patterns: CBT

AUX: MovieLens
TGT: BookCrossing
kNN CBT
MAE 0,5216 0,5064
RMSE 0,4736 0,4492

73
 Transferring rating patterns: CBT

AUX: MovieLens
TGT: BookCrossing
kNN CBT RND
MAE 0,5216 0,5064 0,4963
RMSE 0,4736 0,4492 0,4380

Paolo Cremonesi and Massimo Quadrana. 2014.

Cross-domain recommendations without overlapping data: myth or reality?
74
 Contents

1. The cross-domain recommendation problem

2. Cross-domain recommendation techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation

75
 Data partitioning

Hold-out

Leave-some-users-out

76
Leave-all-users-out
 Goal vs. partitioning technique

Leave-some- Leave-all-users-
Hold-out
users-out out
Cold start X
New user X
New item X
Accuracy X
Diversity X

77
 Contents

1. The cross-domain recommendation problem

2. Cross-domain recommendation techniques
3. Evaluation of cross-domain recommenders
4. Open issues in cross-domain recommendation

78
 Research issues (I)

• Synergy between cross-domain and contextual

recommendations
• different contexts (e.g., location, time, and mood) can
be treated as different domains
• … and vice versa

79
 Research issues (II)
• Cross-domain recommenders to reduce the user
model elicitation effort
• able to build detailed user profiles without the need
to collect explicit user preferences

• New, real life cross-domain datasets

• quite scarce and hard to reach in practice;
• gathered by big industry players, like Amazon, eBay,
and Yelp, but rarely available to the broader research
community
80
8th ACM Conference on Recommender Systems
Foster City, Silicon Valley, California, USA, 6th-10th October 2014

Cross-domain Recommender Systems

Iván Cantador1, Paolo Cremonesi2
1 Universidad Autónoma de Madrid, Madrid, Spain
[email protected]
2 Politecnico di Milano, Milan, Italy

[email protected]
 References (I)

• Surveys
• Dong, G.: Cross Domain Similarity Mining: Research Issues and Potential
Applications Including Supporting Research by Analogy. ACM SIGKDD
Explorations Newsletter 14(1), pp. 43-47 (2012)
• Fernández-Tobías, I., Cantador, I., Kaminskas, M., Ricci, F.: Cross-domain
Recommender Systems: A Survey of the State of the Art. In: Proc. of the
2nd Spanish Conference on Information Retrieval, pp. 187-198 (2012)
• Li, B.: Cross-Domain Collaborative Filtering: A Brief Survey. In: Proc. of the
23rd IEEE International Conference on Tools with Artificial Intelligence, pp.
1085-1086 (2011)
• Pan, S. J., Yang, Q.: A Survey on Transfer Learning. IEEE Transactions on
Knowledge and Data Engineering 22(10), pp. 1345-1359 (2010)

82
 References (II)
• Aggregating knowledge: merging user preferences
• Abel, F., Araújo, S., Gao, Q., Houben, G.-J.: Analyzing Cross-system User
Modeling on the Social Web. In: Proc. of the 11th International
Conference on Web Engineering, pp. 28-43 (2011)
• Cantador, I., Fernández-Tobías, I., Bellogín, A.: Relating Personality Types
with User Preferences in Multiple Entertainment Domains. In: Proc. of
the 1st Workshop on Emotions and Personality in Personalized Services,
CEUR workshop Proceedings, vol. 997 (2013)
• Cremonesi, P., Tripodi, A., Turrin, R.: Cross-domain Recommender
Systems. In: Proc. of the 11th IEEE International Conference on Data
Mining Workshops, pp. 496-503 (2011)
• Fernández-Tobías, I., Cantador, I., Plaza, L.: An Emotion Dimensional
Model Based on Social Tags: Crossing Folksonomies and Enhancing
Recommendations. In: Proc. of the 14th International Conference on E-
Commerce and Web Technologies, pp. 88-100 (2013)
• González, G., López, B., de la Rosa, J. LL.: A Multi-agent Smart User Model
for Cross-domain Recommender Systems. In: Proc. of Beyond
Personalization 2005 - The Next Stage of Recommender Systems Research,
pp. 93-94 (2005)
83
 References (III)

• Aggregating knowledge: merging user preferences

• Loni, B, Shi, Y, Larson, M. A., Hanjalic, A.: Cross-Domain Collaborative
Filtering with Factorization Machines. In: Proc. of the 36th European Conf.
on Information Retrieval (2014)
• Nakatsuji, M., Fujiwara, Y., Tanaka, A., Uchiyama, T., Ishida, T.:
Recommendations Over Domain Specific User Graphs. In: Proc. of the
19th European Conference on Artificial Intelligence, pp. 607-612 (2010)
• Szomszor, M. N., Alani, H., Cantador, I., O'Hara, K., Shadbolt, N.: Semantic
Modelling of User Interests Based on Cross-Folksonomy Analysis. In:
Proc. of the 7th Intl. Semantic Web Conference, pp. 632-648 (2008)
• Tiroshi, A., Berkovsky, S., Kaafar, M. A., Chen, T., Kuflik, T.: Cross Social
Networks Interests Predictions Based on Graph Features. In: Proc. of the
7th ACM Conference on Recommender Systems, pp. 319-322 (2013)
• Winoto, P., Tang, T.: If You Like the Devil Wears Prada the Book, Will You
also Enjoy the Devil Wears Prada the Movie? A Study of Cross-Domain
Recommendations. New Generation Computing 26, pp. 209-225 (2008)

84
 References (IV)
• Aggregating knowledge: mediating user modeling data
• Berkovsky, S., Kuflik, T., Ricci, F.: Cross-Domain Mediation in Collaborative
Filtering. In: Proc. of the 11th International Conference on User Modeling,
pp. 355-359 (2007)
• Low, Y., Agarwal, D., Smola, A. J.: Multiple Domain User Personalization.
In: Proc. of the 17th ACM SIGKDD Conf. on Knowledge Discovery and Data
Mining, pp. 123-131 (2011)
• Pan, W., Xiang, E. W., Yang, Q.: Transfer Learning in Collaborative Filtering
with Uncertain Ratings. In: Proc. of the 26th AAAI Conference on Artificial
Intelligence, pp. 662-668 (2012)
• Shapira, B., Rokach, L., Freilikhman, S.: Facebook Single and Cross Domain
Data for Recommendation Systems. UMUAI 23(2-3), pp. 211-247 (2013)
• Stewart, A., Diaz-Aviles, E., Nejdl, W., Marinho, L. B., Nanopoulos, A.,
Schmidt-Thieme, L.: Cross-tagging for Personalized Open Social
Networking. In: Proc. of the 20th ACM Conference on Hypertext and
Hypermedia, pp. 271-278 (2009)
• Tiroshi, A., Kuflik, T.: Domain Ranking for Cross Domain Collaborative
Filtering. In: Proc. of the 20th International Conf. on User Modeling,
Adaptation, and Personalization, pp. 328-333 (2012) 85
 References (V)

• Aggregating knowledge: combining recommendations

• Berkovsky, S., Kuflik, T., Ricci, F.: Distributed Collaborative Filtering with
Domain Specialization. In: Proc. of the 1st ACM Conference on
Recommender Systems, pp. 33-40 (2007)
• Givon, S., Lavrenko, V.: Predicting Social-tags for Cold Start Book
Recommendations. In: Proc. of the 3rd ACM Conference on Recommender
Systems, pp. 333-336 (2009)
• Zhuang, F., Luo, P., Xiong, H., Xiong, Y., He, Q., Shi, Z.: Cross-domain
Learning from Multiple Sources: A Consensus Regularization Perspective.
IEEE Transactions on Knowledge and Data Engineering 22(12), pp. 1664-
1678 (2010)

86
 References (VI)
• Linking/transferring knowledge: linking domains
• Azak, M.: Crossing: A Framework to Develop Knowledge-based
Recommenders in Cross Domains. MSc thesis, Middle East Technical
University (2010)
• Berkovsky, S., Goldwasser, D., Kuflik, T., Ricci, F.: Identifying Inter-domain
Similarities through Content-based Analysis of Hierarchical Web-
Directories. In: Proc. of the 17th European Conference on Artificial
Intelligence, pp. 789-790 (2006)
• Cao, B., Liu, N. N., Yang, Q.: Transfer Learning for Collective Link
Prediction in Multiple Heterogeneous Domains. Proc. of 27th
International Conf. on Machine Learning, pp. 159-166 (2010)
• Chung, R., Sundaram, D., Srinivasan, A.: 2007. Integrated Personal
Recommender Systems. In: Proc. of the 9th International Conference on
Electronic Commerce, pp. 65-74 (2007)

87
 References (VII)
• Linking/transferring knowledge: linking domains
• Fernández-Tobías, I., Cantador, I., Kaminskas, M., Ricci, F.: 2011. A Generic
Semantic-based Framework for Cross-domain Recommendation. In: Proc.
of the 2nd International Workshop on Information Heterogeneity and
Fusion in Recommender Systems, pp. 25-32 (2011)
• Loizou, A.: How to Recommend Music to Film Buffs: Enabling the
Provision of Recommendations from Multiple Domains. PhD thesis,
University of Southampton (2009)
• Shi, Y., Larson, M., Hanjalic, A.: Tags as Bridges between Domains:
Improving Recommendation with Tag-induced Cross-domain
Collaborative Filtering. In: Proc. of the 19th International Conference on
User Modeling, Adaption, and Personalization, pp. 305-316 (2011)
• Zhang, Y., Cao, B., Yeung, D.-Y.: Multi-Domain Collaborative Filtering. In:
Proc. of the 26th Conference on Uncertainty in Artificial Intelligence, pp.
725-732 (2010)
88
 References (VIII)
• Linking/transferring knowledge: sharing latent features
• Enrich, M., Braunhofer, M., Ricci, F.: Cold-Start Management with Cross-
Domain Collaborative Filtering and Tags. In: Proc. of the 14th Intl.
Conference on E-Commerce and Web Technologies, pp. 101-112 (2013)
• Fernández-Tobías, I., Cantador, I.: Exploiting Social Tags in Matrix
Factorization Models for Cross-domain Collaborative Filtering. In: Proc. of
the 1st Intl. Workshop on New Trends in Content-based Recommender
Systems (2013)
• Hu, L., Cao, J., Xu, G., Cao, L., Gu, Z., Zhu, C.: Personalized
Recommendation via Cross-domain Triadic Factorization. In: Proc. of the
22nd International Conference on World Wide Web, pp. 595-606 (2013)
• Pan, W., Liu, N. N., Xiang, E. W., Yang, Q.: Transfer Learning to Predict
Missing Ratings via Heterogeneous User Feedbacks. In: Proc. of the 22nd
Intl. Joint Conference on Artificial Intelligence, pp. 2318-2323 (2011)
• Pan, W., Xiang, E. W., Liu, N. N., Yang, Q.: Transfer Learning in
Collaborative Filtering for Sparsity Reduction. Proc. of the 24th AAAI Conf.
on Artificial Intelligence, pp. 210-235 (2010)
89
 References (IX)

• Linking/transferring knowledge: transferring rating patterns

• Cremonesi, P., Quadrana, M.: Cross-domain Recommendations without
Overlapping Data: Myth or Reality? In: Proc. of the 8th ACM Conference
on Recommender Systems (2014)
• Gao, S., Luo, H., Chen, D., Li, S., Gallinari, P., Guo, J.: Cross-Domain
Recommendation via Cluster-Level Latent Factor Model. In: Proc. of the
17th and 24th European Conference on Machine Learning and Knowledge
Discovery in Databases, pp. 161-176 (2013)
• Lee, C. H., Kim, Y. H., Rhee, P. K.: Web Personalization Expert with
Combining Collaborative Filtering and Association Rule Mining
Technique. Expert Systems with Applications 21(3), pp. 131-137 (2001)

90
 References (X)

• Linking/transferring knowledge: transferring rating patterns

• Li, B., Yang, Q., Xue, X.: Can Movies and Books Collaborate? Cross-domain
Collaborative Filtering for Sparsity Reduction. In: Proc. of the 21st
International Joint conference on Artificial Intelligence, pp. 2052-2057
(2009)
• Li, B., Yang, Q., Xue, X.: Transfer Learning for Collaborative Filtering via a
Rating-matrix Generative Model. In: Proc. of the 26th International Conf.
on Machine Learning, pp. 617-624 (2009)
• Moreno, O. Shapira, B. Rokach, L. Shani, G.: TALMUD: Transfer Learning
for Multiple Domains. In: Proc. of the 21st ACM Conf. on Information and
Knowledge Management, pp. 425-434 (2012)

91