Techniques for Context-Aware and Cold-Start Recommendations

PhD Thesis Defense, 28th Ph.D. Cycle, May 2016
Techniques for Context-Aware and
Cold-Start Recommendations
Matthias Braunhofer

Supervisor: Prof. Francesco Ricci 
Free University of Bozen - Bolzano

Piazza Domenicani 3, 39100 Bolzano, Italy

mbraunhofer@unibz.it

Outline
2
• Context-Aware Recommenders and the Cold-Start Problem
• State of the Art
• South Tyrol Suggests Application Scenario
• Hybrid Context-Aware Recommendation Algorithms
• Active Learning for Context-Aware Recommenders
• Conclusions and Future Work

• Recommender Systems (RSs) are information ﬁltering and decision
support tools suggesting interesting items to the user based on feedback

• Explicit feedback (e.g., ratings) vs. implicit feedback (e.g., browsing history)

• Two popular approaches:

• Collaborative Filtering (CF)

• Content-based
Recommender Systems
3

Context is Essential
• Main idea: users can experience the same item diﬀerently depending on the
current contextual situation (e.g., weather, season, mood)

• RSs must take into account this information to deliver more useful (perceived)
recommendations
4

Context-Aware Recommender Systems
• Context-Aware Recommender Systems (CARSs) improve traditional RSs
by adapting their suggestions to the contextual situations of the user and
the recommended items

• Example: Google Now
5

Cold-Start Problem
• CARSs suﬀer from the cold-start problem

• New user problem: How do you recommend to a new user?

• New item problem: How do you recommend a new item with no ratings?

• New context problem: How do you recommend in a new context?
6
1 ? 1 ?
2 5 ?
? 3 ?
3 ? 5 ?
2 5 ?
? 3 ?
5 ? 5 ?
4 5 4 ?
? 3 5 ?
1 ? 1
2 5
? 3
3 ? 5
2 5
? 3
5 ? 5
4 5 4
? 3 5
? ? ?
? ? ?
1 ? 1
2 5
? 3
3 ? 5
2 5
? 3
5 ? 5
4 5 4
? 3 5

Cold-Start Problem
• CARSs suﬀer from the cold-start problem

• New user problem: How do you recommend to a new user?

• New item problem: How do you recommend a new item with no ratings?

• New context problem: How do you recommend in a new context?
6
1 ? 1 ?
2 5 ?
? 3 ?
3 ? 5 ?
2 5 ?
? 3 ?
5 ? 5 ?
4 5 4 ?
? 3 5 ?
1 ? 1
2 5
? 3
3 ? 5
2 5
? 3
5 ? 5
4 5 4
? 3 5
? ? ?
? ? ?
1 ? 1
2 5
? 3
3 ? 5
2 5
? 3
5 ? 5
4 5 4
? 3 5
Focus of this research

Outline
7

Approaches for Cold-Starting CARSs
8
Cold-starting CARSs
8

8
Cold-starting CARSs
… using additional knowledge
sources
… better using existing
knowledge
8

8
Cold-starting CARSs
sources
knowledge
Active learning
(Elahi et al., 2013)
8

8
Cold-starting CARSs
sources
knowledge
Active learning
Cross-domain rec.
(Enrich et al., 2013)
8

8
Cold-starting CARSs
sources
knowledge
Active learning
Cross-domain rec.
Implicit feedback
(Koren, 2008)
8

8
Cold-starting CARSs
sources
knowledge
Active learning
Cross-domain rec.
User / item attributes
(Musto et al., 2013)
Implicit feedback
(Koren, 2008)
8

8
Cold-starting CARSs
sources
knowledge
Active learning
Cross-domain rec.
Selective context acquisition
(Baltrunas et al., 2012)
Implicit feedback
(Koren, 2008)
8

8
Cold-starting CARSs
sources
knowledge
Active learning
Cross-domain rec.
Selective context acquisition
(Baltrunas et al., 2012)
Context hierarchy / similarity
(Codina et al., 2013)
Implicit feedback
(Koren, 2008)
8

Outline
9

Interaction with the System
10
Welcome screen

10
Registration screen

10
Personality questionnaire

10
Questionnaire results

10
Slide-out navigation menu

10
Suggestions screen

10
Active learning

10
Details screen

10
Routing screen

10
Proﬁle page

Research Hypotheses
Tested through User Studies
11

Research Hypotheses
• Personality is useful to elicit more ratings from new users than some state-of-
the-art AL strategies based on heuristics
11

Research Hypotheses
• Personality can be exploited for eliciting ratings from new users that lead to
an improved system prediction accuracy
11

Research Hypotheses
• Personality can be exploited for eliciting ratings from new users that lead to
an improved system prediction accuracy
• Personality can be helpful to acquire ratings from new users which result in
recommendations better tailored to the user’s context
11

Research Hypotheses
Tested with Ofﬂine Experiments
12

Research Hypotheses
• Hybrid CARS algorithms are beneﬁcial for delivering accurate context-aware
rating predictions in cold-start situations
12

Research Hypotheses
• Hybrid CARS algorithms can achieve a high recommendation ranking quality
in cold-start situations
12

Research Hypotheses
• Hybrid CARS algorithms can achieve a high recommendation ranking quality
in cold-start situations
• Parsimonious and adaptive context acquisition can save time and eﬀort of the
user by eﬀectively identifying what contextual factors to acquire upon rating
an item
12

Outline
13

Hybrid Context-Aware Recommenders
14
• Conjecture: it is possible to adaptively combine multiple CARS algorithms in
order to take advantage of their strengths and alleviate their drawbacks in
diﬀerent cold-start situations

• Example:
(user, item,
context) tuple
CARS 1
CARS 2
Hybridization Final score
Score
Score
Hybrid CARS

• Matrix Factorization (MF) predicts unknown ratings by discovering some
latent features that determine how a user rates an item; items similar to the
target user in this latent space are recommended
Matrix Factorization Methods
15
r11 r12 r13 r14
r21 r22 r23 r24
r31 r32 r33 r34
r41 r42 r43 r44
r51 r52 r53 r54
a b c
x
y
z=
r q p
5 x 4 matrix 5 x 3 matrix 3 x 4 matrix
r42 = (a, b, c) · (x, y, z) = a * x + b * y + c * z
ȓui = qi
Tpu

15
r11 r12 r13 r14
r21 r22 r23 r24
r31 r32 r33 r34
r41 r42 r43 r44
r51 r52 r53 r54
a b c
x
y
z=
r q p
r42 = (a, b, c) · (x, y, z) = a * x + b * y + c * z
ȓui = qi
TpuRating prediction

15
r11 r12 r13 r14
r21 r22 r23 r24
r31 r32 r33 r34
r41 r42 r43 r44
r51 r52 r53 r54
a b c
x
y
z=
r q p
r42 = (a, b, c) · (x, y, z) = a * x + b * y + c * z
ȓui = qi
Tpu
Item preference factor
vector

15
r11 r12 r13 r14
r21 r22 r23 r24
r31 r32 r33 r34
r41 r42 r43 r44
r51 r52 r53 r54
a b c
x
y
z=
r q p
r42 = (a, b, c) · (x, y, z) = a * x + b * y + c * z
ȓui = qi
Tpu User preference factor
vector

• CAMF-CC (Context-Aware Matrix Factorization for item categories) is a
variant of CAMF that extends standard MF by incorporating baseline
parameters for contextual condition-item category pairs

Basic CARS Algorithms
CAMF-CC (Baltrunas et al., 2011)
16
ˆruic1...ck
= qi
T
pu + ri + bu + btcj
j=1
k
∑
t∈T (i)
∑
qi latent factor vector of item i

pu latent factor vector of user u

average rating for item i

bu baseline for user u

T(i) set of categories associated to item i

btcj baseline for item category-contextual condition tcj
ri

SPF (Codina et al., 2013)
17
• SPF (Semantic Pre-Filtering) is a contextual pre-ﬁltering method that, given
a target contextual situation, uses a standard MF model learnt from all the
ratings tagged with contextual situations identical or similar to the target one

• Conjecture: learning the prediction model on a larger number of ratings, even
if not obtained exactly in the target context, will help
• Key step: similarity calculation
1 -0.5 2 1
-2 0.5 -2 -1.5
-2 0.5 -1 -1
Condition-to-item co-occurrence matrix
1 -0.96 -0.84
-0.96 1 0.96
-0.84 0.96 1
Cosine similarity between conditions

Content-based CAMF-CC
18
• It is a novel variant of CAMF-CC that incorporates additional sources of
information about the items, e.g., category or genre information

• Conjecture: alleviates the new item problem of CAMF-CC
ˆruic1...ck
= (qi + xa )
a∈A(i)
∑
T
pu + ri + bu + btcj
j=1
k
∑
t∈T (i)
∑

A(i) set of item attributes

xa latent factor vector of item attribute a


average rating for item i



ri

Demographics-based CAMF-CC
19
• It is a novel variant of CAMF-CC that proﬁles users through known user
attributes (e.g., age group, gender, personality traits)

• Conjecture: alleviates the new user problem of CAMF-CC
ˆruic1...ck
= qi
T
(pu + ya )
a∈A(u)
∑ + ri + bu + btcj
j=1
k
∑
t∈T (i)
∑


A(u) set of user attributes

ya latent factor vector of user attribute a

overall average rating



ri

Hybrid CARS Algorithms
Heuristic Switching
• Heuristic Switching uses a stable heuristic to switch between a set of basic
CARS algorithms depending on the encountered cold-start situation

• Conjecture: better tackles speciﬁc cold-start situations found in CARSs
20
R1: Use content-based CAMF-CC for a new item.
R2: Use demographics-based CAMF-CC for a new user.
R3: Average the predictions of content-based CAMF-CC and
demographics-based CAMF-CC for new contextual
situations or mixtures of cold-start cases.

• Adaptive Weighted adaptively sums the predictions of the basic algorithms
weighted by their estimated accuracies for the user, item and contextual
situation in question

• Extends the two-dimensional adaptive RS presented in (Bjørkøy, 2011)

• Conjecture: optimizes adaptation of differently performing CARS algorithms
Adaptive Weighted (1/2)
21
ˆr
…
∑
…
ˆr1
ˆr2
ˆrm
â1
â2
âm

Adaptive Weighted (2/2)
22
• Builds for each basic CARS algorithm a new user-item-context error tensor
whose entries are the known deviations (errors) of the CARS predictions from
the true ratings

• Uses a separate CARS error prediction model for each of these error tensors
to predict the errors (accuracies) on a particular (user, item, context) tuple
ˆeuic1...ck
= (qi + xci
ci∈IC
∑ )T
(pu + ycu
cu ∈UC
∑ )+ ei + bu


IC subset of item-related contextual conditions

xci latent factor vector of contextual condition ci

UC subset of user-related contextual conditions

ycu latent factor vector of contextual condition cu

average error for item i

ei

Feature Weighted (1/2)
23
• Feature Weighted adaptively sums the
weighted predictions of the basic
algorithms with weights estimated using
meta-features, i.e., the number of user,
item and context ratings

• Is inspired by the Feature-Weighted
Linear Stacking (FWLS) algorithm (Sill et
al., 2009)

• Conjecture: exploits cold-start
conditions under which performance
diﬀerences between the CARS
algorithms can be observed
ˆv1
1
ˆa1
…
ˆr
∑
ˆr1
ˆrm
∑ ∑
…
…
…
…
f1 fn f1 fn
ˆv1
1
ˆvn
1
ˆv1
m
ˆvn
m

Feature Weighted (2/2)
24
• It extends standard linear stacking, which is a method for linearly combining
the predictions of diﬀerent models m ∈ M:

• Feature Weighted models the weight ŵm as a linear function of some meta-
features f ∈ F:

• The rating prediction function is rewritten as:
ˆruic1...ck
= ˆwm
m∈M
∑ ˆruic1...ck
m
ˆwm
= ˆvf
m
f ∈F
∑ f (u,i,c1,...,ck )
ˆruic1...ck
= ( ˆvf
m
f ∈F
∑ f (u,i,c1,...,ck ))
m∈M
∑ ˆruic1...ck
m
ˆruic1...ck
m

Evaluation
Used Datasets
25
• 4 contextually-tagged rating datasets
STS
(Braunhofer et al.,
2013)
CoMoDa
(Odić et al.,
2013)
Music
(Baltrunas et al.,
2011)
TripAdvisor
(www.tripadvisor.
com)
Domain POIs Movies Music POIs
Rating scale 1-5 1-5 1-5 1-5
Ratings 2,534 2,296 4,012 7,154
Users 325 121 43 5,487
Items 249 1,232 139 1,263
Contextual factors 14 12 8 3
Contextual conditions 57 49 26 31
Contextual situations 931 1,969 26 512
User attributes 7 4 10 2
Item features 1 7 2 2

Evaluation
Evaluation Procedure
26
• Randomly divide the entities (i.e., users, items or contexts) into 10 cross-
validation folds

• For each fold k = 1, 2, …, 10

• Use all the ratings except those coming from entities in fold k as training
set to build the prediction models

• Calculate the Mean Absolute Error (MAE) and normalized Discounted
Cumulative Gain (nDCG) on the test ratings for the entities in fold k

• Advantage: allows to test the models on really cold entities

• Disadvantage: can’t test for diﬀerent degrees of coldness

Results
Recommendation for New Users
27
MAE
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
STS CoMoDa Music TripAdvisor
CAMF-CC SPF
Content-CAMF-CC Demographics-CAMF-CC
* *
*
* *
*
*
MAEdifftobestbasicalgorithm -0.08
-0.06
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
Average Weighted Heuristic Switching
Adaptive Weighted Feature Weighted
*
Stars denote significant differences w.r.t. CAMF-CC  
(p < 0.05)
Stars denote significant differences w.r.t. best basic CARS algorithm 
(p < 0.05)

Results
Recommendation for New Items
28
(p < 0.05)
(p < 0.05)
MAE
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
CAMF-CC SPF
* *
* *
*
*
*
*
*
*
-0.06
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
* *
*
*
*

Results
Recommendation under New Contexts
29
MAE
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
CAMF-CC SPF
*
*
* *
*
*
*
(p < 0.05)
(p < 0.05)
-0.06
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
* *
*
*
*
* *
*

Summary
30
Algorithm Pros Cons
Average
Weighted
• Simple and fast to train • Sensitive to poorly performing basic
algorithms
• Works only when all basic algorithms are
performing equally well

Summary
30
Algorithm Pros Cons
Average
Weighted
algorithms
Heuristic
Switching
• Simple and fast to train
• Can avoid the impact of poorly performing
basic algorithms
• Depends on the manual choice of the
heuristic

Summary
30
Algorithm Pros Cons
Average
Weighted
algorithms
Heuristic
Switching
basic algorithms
heuristic
Adaptive
Weighted
• Adaptively combines the basic algorithms
based on their strengths and weaknesses
• Complex and slow to train
• Sensitive to the training set used
• Optimized for error minimization
• Sensitive to poorly performing basic
algorithms

Summary
30
Algorithm Pros Cons
Average
Weighted
algorithms
Heuristic
Switching
basic algorithms
heuristic
Adaptive
Weighted
• Sensitive to poorly performing basic
algorithms
Feature
Weighted
• Robust in all cold-start cases

Outline
31

• Conjecture: Active Learning (AL), which identiﬁes the most useful items for
the target user to rate, can be improved for CARSs by leveraging the user’s
personality and by identifying the most useful contextual factors to be entered
upon rating these items
Active Learning for CARSs
32
item ratings
item ratings
request
approximated 
function
supervised
learning
Active Learning
Passive Learning
user

32
item ratings
item ratings
request
approximated 
function
supervised
learning
Active Learning
Passive Learning
personality
(Big-5)
user

32
item ratings
item ratings
request
approximated 
function
supervised
learning
Active Learning
Passive Learning
personality
(Big-5)
user
+ context data
+ context data request

Using Personality in Active Learning
• Main idea: people with similar personality are likely to have similar interests
(Rentfrow & Gosling, 2003), and thus the incorporation of human personality can
help in predicting the items that can be rated by a user
33
Neuroticism
Conscientious-
ness
Openness
ExtraversionAgreeableness
Big Five
Personality
Traits

Personality-Based Binary Prediction
• Input: Target user u. Maximum number of items to be returned N. Binary user-
item rating matrix B. Candidate set of items to be rated Cu

• Output: List of M <= N top-scoring items for which user u is requested to
provide ratings
34


A(u) set of user u’s attributes (i.e., Big-5 scores) 
ya latent factor vector of user attribute a 
average binary rating for item i

xi
ˆxui = qi
T
(pu + ya )
a∈A(u)
∑ + xi + bu

Parsimonious & Adaptive Context Acquisition
• Main idea: for each user-item pair (u, i),
identify the contextual factors that
when acquired with u’s rating for i
improve most the long term
performance of the recommender

• Heuristic: acquire the contextual
factors that have the largest impact
on rating prediction

• Challenge: how to quantify these
impacts?
35

CARS Prediction Model
• We use the new variant of CAMF that we already successfully employed to
estimate the rating prediction accuracy of a CARS algorithm

• Advantage: allows to capture latent correlations and patterns between a
potentially wide range of knowledge sources ⟹ ideal to derive the usefulness
of contextual factors
36
ˆruic1...ck
= (qi + xa
a∈A(i)∪C(i)
∑ )T
⋅(pu + yb
b∈A(u)∪C(u)
∑ )+ ri + bu

A(i) set of conventional item attributes (e.g., genre)

C(i) set of contextual item attributes (e.g., weather)

xa latent factor vector of item attribute a


A(u) set of conventional user attributes (e.g., age)

C(u) set of contextual user attributes (e.g., mood)

yb latent factor vector of user attribute b

ṝi average rating for item i


Largest Deviation
• Given (u, i), it ﬁrst measures the “impact” of each contextual condition cj ∈ Cj
by calculating the absolute deviation between the rating prediction when the
condition holds (i.e., ȓuicj) and the predicted context-free rating (i.e., ȓui):

where fcj is the normalized frequency of cj

• Finally, it computes for each factor the average of these deviation scores, and
selects the contextual factors with the largest average scores
37
ˆwuicj
= fcj
ˆruicj
− ˆrui ,

Experiments 1 and 2
• 2 user studies involving 108 subjects in the 1st and 51 subjects in the 2nd

• Compared personality-based binary prediction with log(popularity) *
entropy and random

• Personality-based binary prediction performed best in terms of:

• Number of acquired ratings
• Rating prediction accuracy
• Quality of context-aware recommendations
38

Experiment 3
Datasets
39
• 3 contextually-tagged rating datasets
CoMoDa
(Odić et al.,
2013)
TripAdvisor
(www.tripadvisor.
com)
STS
(Braunhofer et al.,
2013)
Domain Movies POIs POIs
Rating scale 1-5 1-5 1-5
Ratings 2,098 4,147 2,534
Users 112 3,916 325
Items 1,189 569 249
Contextual factors 12 3 14
Contextual conditions 49 31 57
Avg. # of conditions / rating 12 3 1.49
User attributes 4 2 7
Item features 7 2 1

Experiment 3
40
• Repeated random sub-sampling validation (20 times):

Experiment 3
40
25% 50% 25%
Training set Candidate set Testing set
• Randomly partition the ratings into three subsets

• For each user-item pair (u,i) in the candidate set, compute the N most relevant
contextual factors and transfer the corresponding rating and context information ruic
in the candidate set to the training set as ruic' with c' ⊆ c containing the associated
contextual conditions for these factors, if any
Experiment 3
40
25% 50% 25%

• Measure user-averaged MAE (U-MAE), Precision@10 and Recall@10 on the testing
set, after training the prediction model on the new extended training set
Experiment 3
40
25% 50% 25%

• Measure user-averaged MAE (U-MAE), Precision@10 and Recall@10 on the testing
set, after training the prediction model on the new extended training set
Experiment 3
40
25% 50% 25%
• Repeat

Experiment 3
Evaluation Procedure: Example
41
user-item pair
top two contextual factors
rating transferred to training set
+
+
=
rating in candidate set

Experiment 3
41
(Alice, Skiing)
top two contextual factors
+
+
=

Experiment 3
41
(Alice, Skiing)
Season and Weather
+
+
=

Experiment 3
41
(Alice, Skiing)
Season and Weather
rAlice Skiing Winter Sunny Warm Morning = 5+
+
=

Experiment 3
41
(Alice, Skiing)
Season and Weather
rAlice Skiing Winter Sunny Warm Morning = 5
rAlice Skiing Winter Sunny = 5
+
+
=

Experiment 3
Baseline Methods for Evaluation
42
• Mutual Information: given a user-item pair (u,i), computes the relevance for a
contextual factor Cj as the mutual information between ratings for items
belonging to i’s category (Baltrunas et al., 2012)

• Freeman-Halton Test: calculates the relevance of Cj using the Freeman-
Halton test (Odić et al., 2013)

• Minimum Redundancy Maximum Relevance (mRMR): ranks each Cj
according to its relevance to the rating variable and redundancy to other
contextual factors (Peng et al., 2005)

Experiment 3
Results: Prediction Accuracy
43
CoMoDa
U-MAE
0.71
0.72
0.73
0.74
0.75
0.76
0.77
0.78
0.79
0.80
0.81
0.82
1 2 3 4
Mutual Information Freeman-Halton mRMR Largest Deviation All factors
STS
0.90
0.91
0.92
0.93
0.94
0.95
0.96
0.97
0.98
0.99
1.00
1 2 3 4
Stars denote signiﬁcant improvements of Largest Deviation over the other considered algorithms  
(p < 0.05)
*
*
* * *
* * *

Experiment 3
Results: Ranking Quality
44
CoMoDa
Precision@10
0.0000
0.0002
0.0004
0.0006
0.0008
0.0010
0.0012
0.0014
0.0016
1 2 3 4
STS
0.005
0.006
0.007
0.008
0.009
0.010
0.011
0.012
0.013
0.014
0.015
0.016
1 2 3 4
*
*
*
*
*
*
*
*
(p < 0.05)

Experiment 3
Results: # of Acquired Conditions
45
STS
Avg#ofacquiredconditions
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1 2 3 4
* * * * *
* * *
* *
*
*
(p < 0.05)

Outline
46

Conclusions
• Novel hybrid recommendation algorithms that, in many cases, eﬀectively
alleviate the cold-start problem of CARS

• New personality-based Active Learning rating acquisition algorithm that can
better estimate what items a (new) user is able to rate

• Novel parsimonious and adaptive context acquisition algorithm that can
identify what contextual factors to acquire from the user upon rating an item,
thus minimizing the user’s rating eﬀort

• Comprehensive evaluation of the proposed solutions in cold-start scenarios
47

Future Work
• Additional experiments and datasets

• Improvement of proposed algorithms

• Proactive Active Learning

• Sequential Active Learning

• Gamiﬁcation approaches
48

Publications
Journal Papers
Fernández-Tobías, I., Braunhofer, M., Elahi, M., Cantador, I., & Ricci, F. (2016). Alleviating the New
User Problem in Collaborative Filtering by Exploiting Personality Information. User Modeling and
User-Adapted Interaction, 1-35. http://dx.doi.org/10.1007/s11257-016-9172-z

Braunhofer, M., Elahi, M., & Ricci, F. (2014). Techniques for cold-starting context-aware mobile
recommender systems for tourism. Intelligenza Artiﬁciale, 8(2), 129-143. http://dx.doi.org/10.3233/
IA-140069

Braunhofer, M., Kaminskas, M., & Ricci, F. (2013). Location-aware music recommendation.
International Journal of Multimedia Information Retrieval, 2(1), 31-44. http://dx.doi.org/10.1007/
s13735-012-0032-2

49

Publications
Conference Papers
Nasery, M., Braunhofer, M., & Ricci, F. (2016). Recommendations with Optimal Combination of
Feature-Based and Item-Based Preferences. To appear in User Modeling, Adaptation, and
Personalization. Halifax, Canada: Springer International Publishing

Braunhofer, M., & Ricci, F. (2016). Contextual Information Elicitation in Travel Recommender
Systems. In Information and Communication Technologies in Tourism 2016 (pp. 579-592). Bilbao,
Spain: Springer International Publishing. http://dx.doi.org/10.1007/978-3-319-28231-2_42 (Second
Best Research Paper Award)

Braunhofer, M., Elahi, M., & Ricci, F. (2015). User Personality and the New User Problem in a
Context-Aware Points of Interest Recommender System. In Information and Communication
Technologies in Tourism 2015 (pp. 537-549). Lugano, Switzerland: Springer International Publishing.
http://dx.doi.org/10.1007/978-3-319-14343-9_39

Braunhofer, M., Elahi, M., & Ricci, F. (2014). Usability Assessment of a Context-Aware and
Personality-Based Mobile Recommender System. In E-Commerce and Web Technologies (pp. 77-88).
Munich, Germany: Springer International Publishing. http://dx.doi.org/10.1007/978-3-319-10491-1_9

Braunhofer, M., Elahi, M., Ge, M., & Ricci, F. (2014). Context Dependent Preference Acquisition with
Personality-Based Active Learning in Mobile Recommender Systems. In Learning and Collaboration
Technologies. Technology-Rich Environments for Learning and Collaboration, Held as Part of HCI
International 2014 (pp. 105-116). Heraklion, Crete, Greece: Springer International Publishing. http://
dx.doi.org/10.1007/978-3-319-07485-6_11
50

Publications
Conference Papers (contd.)
Braunhofer, M., Codina, V., & Ricci, F. (2014). Switching hybrid for cold-starting context-aware
recommender systems. In Proceedings of the 8th ACM Conference on Recommender systems
(pp. 349-352). Foster City, Silicon Valley, California, USA: ACM. http://dx.doi.org/
10.1145/2645710.2645757

Braunhofer, M., Elahi, M., Ricci, F., & Schievenin, T. (2013). Context-aware points of interest
suggestion with dynamic weather data management. In Information and Communication
Technologies in Tourism 2014 (pp. 87-100). Dublin, Ireland: Springer International Publishing.
http://dx.doi.org/10.1007/978-3-319-03973-2_7

Elahi, M., Braunhofer, M., Ricci, F., & Tkalcic, M. (2013). Personality-based active learning for
collaborative ﬁltering recommender systems. In AI*IA 2013: Advances in Artiﬁcial Intelligence (pp.
360-371). Turin, Italy: Springer International Publishing. http://dx.doi.org/
10.1007/978-3-319-03524-6_31

Enrich, M., Braunhofer, M., & Ricci, F. (2013). Cold-Start Management with Cross-Domain
Collaborative Filtering and Tags. In E-Commerce and Web Technologies (pp. 101-112). Prague,
Czech Republic: Springer Berlin Heidelberg. http://dx.doi.org/10.1007/978-3-642-39878-0_10
51

Publications
Workshop, Demo & Doctoral Consortium Papers
Braunhofer, M., Fernández-Tobías, I., & Ricci, F. (2015). Parsimonious and Adaptive Contextual
Information Acquisition in Recommender Systems. In Proceedings of the Joint Workshop on
Interfaces and Human Decision Making for Recommender Systems, IntRS 2015, co-located with
ACM Conference on Recommender Systems (RecSys 2015). Vienna, Austria: ACM.

Braunhofer, M., Ricci, F., Lamche, B., & Wörndl, W. (2015). A Context-Aware Model for Proactive
Recommender Systems in the Tourism Domain. In Proceedings of the 17th International
Conference on Human-Computer Interaction with Mobile Devices and Services Adjunct (pp.
1070-1075). Copenhagen, Denmark: ACM. http://dx.doi.org/10.1145/2786567.2794332

Braunhofer, M. (2014). Hybridisation techniques for cold-starting context-aware recommender
systems. In Proceedings of the 8th ACM Conference on Recommender systems, Doctoral
Symposium (pp. 405-408). Foster City, Silicon Valley, California, USA: ACM. http://dx.doi.org/
10.1145/2645710.2653360

Braunhofer, M. (2014). Hybrid solution of the cold-start problem in context-aware recommender
systems. In User Modeling, Adaptation, and Personalization, Doctoral Consortium (pp. 484-489).
Aalborg, Denmark: Springer International Publishing. http://dx.doi.org/
10.1007/978-3-319-08786-3_44
52

Publications
Workshop, Demo & Doctoral Consortium Papers (contd.)
Braunhofer, M., Elahi, M., & Ricci, F. (2014). STS: A Context-Aware Mobile Recommender System
for Places of Interest. In Extended Proceedings of User Modeling, Adaptation, and Personalization
(pp. 75-80). Aalborg, Denmark.

Braunhofer, M., Elahi, M., Ge, M., Ricci, F., & Schievenin, T. (2013). STS: Design of Weather-Aware
Mobile Recommender Systems in Tourism. In Proceedings of the First International Workshop on
Intelligent User Interfaces: Artiﬁcial Intelligence meets Human Computer Interaction (AI*HCI 2013).
A workshop of the XIII International Conference of the Italian Association for Artiﬁcial Intelligence
(AI*IA 2013). Turin, Italy.

53

Questions?
Thank you.

Techniques for Context-Aware and Cold-Start Recommendations

More Related Content

What's hot (20)

Similar to Techniques for Context-Aware and Cold-Start Recommendations (20)

More from Matthias Braunhofer (7)

Recently uploaded (20)

Techniques for Context-Aware and Cold-Start Recommendations