hybridisation techniques for cold-starting context-aware recommender systems

65
RecSys - October 2014, Foster City, USA Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems Matthias Braunhofer Free University of Bozen - Bolzano Piazza Domenicani 3, 39100 Bolzano, Italy [email protected]

Upload: matthias-braunhofer

Post on 07-Dec-2014

115 views

Category:

Technology


0 download

DESCRIPTION

Context-Aware Recommender Systems (CARSs) suffer from the cold-start problem, i.e., the inability to provide accurate recommendations for new users, items or contextual situations. In this research, we aim at solving this problem by exploiting various hybridisation techniques, from simple heuristic-based solutions to complex adaptive solutions, in order to take advantage of the strengths of different CARS algorithms while avoiding their weaknesses in a given (cold-start) situation. Our initial research based on offline experiments using various contextually-tagged rating datasets has shown that basic CARS algorithms perform very differently in different recommendation scenarios, and that they can be effectively hybridised to achieve an overall optimal performance. Further research is now required to find the optimal method for hybridisation.

TRANSCRIPT

Page 1: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

Matthias Braunhofer!

Free University of Bozen - BolzanoPiazza Domenicani 3, 39100 Bolzano, Italy

[email protected]

Page 2: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Outline

2

• Context-Aware Recommenders and the Cold-Start Problem

• Related Work

• Context-Aware Rating Prediction Models

• Evaluation and Results

• Conclusions and Open Issues

Page 3: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Outline

2

• Context-Aware Recommenders and the Cold-Start Problem

• Related Work

• Context-Aware Rating Prediction Models

• Evaluation and Results

• Conclusions and Open Issues

Page 4: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Context-Aware Recommender Systems (CARSs) aim to provide better recommendations by exploiting contextual information (e.g., weather)

• Rating prediction function is: R: Users x Items x Context → Ratings

Context-Aware Recommender Systems

3

3 ? 4

2 5 4

? 3 4

1 ? 1

2 5

? 3

3 ? 5

2 5

? 3

5 ? 5

4 5 4

? 3 5

Page 5: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Example: Google Now

• “The right information at just the right time”

4

Nearby photo spots Traffic & transit Nearby attractions

Page 6: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Example: South Tyrol Suggests (STS)

• Our Android app that offers context-aware place of interest (POI) recommendations for the South Tyrol region of ItalyPersonality questionnaire Rating screen Suggestions screen

5

Page 7: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

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 ? 12 5? 3

3 ? 52 5? 3

5 ? 54 5 4? 3 5? ? ?

? ? ?1 ? 1

2 5

? 3

3 ? 5

2 5

? 3

5 ? 5

4 5 4

? 3 5

Page 8: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Our Solution: Hybrid CARS

• Intuition: it is possible to adaptively combine multiple CARS algorithms in order to take advantage of their strengths and alleviate their drawbacks when predicting a user’s rating for an item given a particular cold-start situation

• Example:

7

(user, item, context) tuple

CARS 1

CARS 2

Combination Final score

Score

Score

Hybrid CARS

Page 9: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Outline

8

• Related Work

• Context-Aware Rating Prediction Models

• Evaluation and Results

• Conclusions and Open Issues

• Context-Aware Recommenders and the Cold-Start Problem

Page 10: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Related Work

9

Cold-starting CARSs

… using additional data … better processing known data

Active Learning (Elahi et al., 2013)

Cross-domain recs. (Enrich et al., 2013)

Implicit feedback (Shi et al., 2012)

User / item attributes (Woerndl et al., 2009)

Context similarities (Codina et al., 2013)

Survey data (Baltrunas et al., 2012)

Page 11: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Related Work

9

Cold-starting CARSs

… using additional data … better processing known data

Active Learning (Elahi et al., 2013)

Cross-domain recs. (Enrich et al., 2013)

Implicit feedback (Shi et al., 2012)

User / item attributes (Woerndl et al., 2009)

Context similarities (Codina et al., 2013)

Survey data (Baltrunas et al., 2012)

No unique optimal solution!

Page 12: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Outline

10

• Related Work

• Context-Aware Rating Prediction Models

• Evaluation and Results

• Conclusions and Open Issues

• Context-Aware Recommenders and the Cold-Start Problem

Page 13: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

MF Methods

• Matrix Factorisation (MF) predicts unknown ratings by discovering some latent features that determine how a user rates an item; features associated with the user should match with the features associated with the item

11

r11 r12 r13 r14

r21 r22 r23 r24

r31 r32 r33 r34

r41 r42 r43 r44

r51 r52 r53 r54

a b c

xyz=

r q p5 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 = qiTpu

Page 14: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

MF Methods

• Matrix Factorisation (MF) predicts unknown ratings by discovering some latent features that determine how a user rates an item; features associated with the user should match with the features associated with the item

11

r11 r12 r13 r14

r21 r22 r23 r24

r31 r32 r33 r34

r41 r42 r43 r44

r51 r52 r53 r54

a b c

xyz=

r q p5 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 = qiTpuRating prediction

Page 15: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

MF Methods

• Matrix Factorisation (MF) predicts unknown ratings by discovering some latent features that determine how a user rates an item; features associated with the user should match with the features associated with the item

11

r11 r12 r13 r14

r21 r22 r23 r24

r31 r32 r33 r34

r41 r42 r43 r44

r51 r52 r53 r54

a b c

xyz=

r q p5 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 = qiTpu

Item preference factor vector

Page 16: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

MF Methods

• Matrix Factorisation (MF) predicts unknown ratings by discovering some latent features that determine how a user rates an item; features associated with the user should match with the features associated with the item

11

r11 r12 r13 r14

r21 r22 r23 r24

r31 r32 r33 r34

r41 r42 r43 r44

r51 r52 r53 r54

a b c

xyz=

r q p5 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 = qiTpu User preference factor vector

Page 17: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms CAMF-CC (Baltrunas et al., 2011)

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

12

ruic1,...,ck = qiT pu + µ + bi + bu + btcj

j=1

k

∑t∈T (i )∑

qi latent factor vector of item ipu latent factor vector of user uμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 18: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms CAMF-CC (Baltrunas et al., 2011)

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

12

ruic1,...,ck = qiT pu + µ + bi + bu + btcj

j=1

k

∑t∈T (i )∑

qi latent factor vector of item ipu latent factor vector of user uμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 19: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms CAMF-CC (Baltrunas et al., 2011)

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

12

ruic1,...,ck = qiT pu + µ + bi + bu + btcj

j=1

k

∑t∈T (i )∑

qi latent factor vector of item ipu latent factor vector of user uμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 20: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms CAMF-CC (Baltrunas et al., 2011)

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

12

ruic1,...,ck = qiT pu + µ + bi + bu + btcj

j=1

k

∑t∈T (i )∑

qi latent factor vector of item ipu latent factor vector of user uμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 21: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms CAMF-CC (Baltrunas et al., 2011)

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

12

ruic1,...,ck = qiT pu + µ + bi + bu + btcj

j=1

k

∑t∈T (i )∑

qi latent factor vector of item ipu latent factor vector of user uμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 22: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms CAMF-CC (Baltrunas et al., 2011)

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

12

ruic1,...,ck = qiT pu + µ + bi + bu + btcj

j=1

k

∑t∈T (i )∑

qi latent factor vector of item ipu latent factor vector of user uμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 23: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms SPF (Codina et al., 2013)

• 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: addresses cold-start problems caused by exact pre-filtering

• Key step: similarity calculation

13

1 -0.5 2 1

-2 0.5 -2 -1.5

-2 0.5 -1 -1

1 -0.96 -0.84

-0.96 1 0.96

-0.84 0.96 1

Condition-to-item co-occurrence matrix Cosine similarity between conditions

Page 24: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms Content-based CAMF-CC

• 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

14

ruic1,...,ck = (qi + xa )a∈A(i )∑ T

pu + µ + bi + bu + btcjj=1

k

∑t∈T (i )∑

qi latent factor vector of item iA(i) set of item attributes xa latent factor vector of item attribute apu latent factor vector of user uμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 25: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms Content-based CAMF-CC

• 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

14

ruic1,...,ck = (qi + xa )a∈A(i )∑ T

pu + µ + bi + bu + btcjj=1

k

∑t∈T (i )∑

qi latent factor vector of item iA(i) set of item attributes xa latent factor vector of item attribute apu latent factor vector of user uμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 26: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms Demographics-based CAMF-CC

• 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

15

ruic1,...,ck = qiT (pu + ya )

a∈A(u )∑ + µ + bi + bu + btcj

j=1

k

∑t∈T (i )∑

qi latent factor vector of item ipu latent factor vector of user uA(u) set of user attributesya latent factor vector of user attribute aμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 27: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Basic CARS Algorithms Demographics-based CAMF-CC

• 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

15

ruic1,...,ck = qiT (pu + ya )

a∈A(u )∑ + µ + bi + bu + btcj

j=1

k

∑t∈T (i )∑

qi latent factor vector of item ipu latent factor vector of user uA(u) set of user attributesya latent factor vector of user attribute aμ overall average ratingbi baseline for item ibu baseline for user uT(i) set of categories associated to item ibtcj baseline for item category-contextual condition tcj

Page 28: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

Page 29: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

Page 30: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

Page 31: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

New user?

Page 32: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

New user?

Y

Page 33: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

New user?

Y

New item?

Page 34: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

New user?

Y

New item?

Y

Page 35: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

New user?

Y

New item?

Y Content-CAMF-CC & Demogr.-CAMF-CC

Page 36: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

New user?

Y

New item?

Y Content-CAMF-CC & Demogr.-CAMF-CC

Score

Page 37: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Heuristic Switching

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

• Conjecture: better tackles all kinds of cold-start situations found in CARSs

16

(user, item, context) tuple

Final score

Y Demogr.-CAMF-CC

Content-CAMF-CC

CAMF-CC

New item?

N

Y

N

New context?

New context?

Y

N

New item?

New user?

Content-CAMF-CC & Demogr.-CAMF-CC

Y

N

Y

N

Final score

Final score

Final score

Score

Score

Score

Score

(new user, new item, known context) tuple

New user?

Y

New item?

Y Content-CAMF-CC & Demogr.-CAMF-CC Final score

Score

Page 38: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Adaptive Weighted adaptively weights each basic CARS algorithm based on its predicted accuracy for the user, item and contextual situation in question

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

• Conjecture: optimises adaptation of differently performing CARS algorithms

Hybrid CARS Algorithms Adaptive Weighted (1/2)

17

(user, item, context) tuple

CAMF-CC

Weighted score Final score

Error model

SPF

Error model

Content-CAMF-CC

Error model

Demogr.-CAMF-

Error model

Score

Error

Score

Error

Score

Error

Score

Error

Page 39: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Adaptive Weighted (2/2)

• 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

18

euic1,...,ck = (qi + xcici∈IC∑ )T (pu + ycu

cu∈UC∑ )+ µ + bi + bu

qi latent factor vector of item ipu latent factor vector of user uIC subset of item-related contextual conditionsxci latent factor vector of contextual condition ciUC subset of user-related contextual conditionsycu latent factor vector of contextual condition cuμ overall average errorbi baseline for item ibu baseline for user u

Page 40: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Adaptive Weighted (2/2)

• 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

18

euic1,...,ck = (qi + xcici∈IC∑ )T (pu + ycu

cu∈UC∑ )+ µ + bi + bu

qi latent factor vector of item ipu latent factor vector of user uIC subset of item-related contextual conditionsxci latent factor vector of contextual condition ciUC subset of user-related contextual conditionsycu latent factor vector of contextual condition cuμ overall average errorbi baseline for item ibu baseline for user u

Page 41: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Adaptive Weighted (2/2)

• 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

18

euic1,...,ck = (qi + xcici∈IC∑ )T (pu + ycu

cu∈UC∑ )+ µ + bi + bu

qi latent factor vector of item ipu latent factor vector of user uIC subset of item-related contextual conditionsxci latent factor vector of contextual condition ciUC subset of user-related contextual conditionsycu latent factor vector of contextual condition cuμ overall average errorbi baseline for item ibu baseline for user u

Page 42: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Adaptive Weighted (2/2)

• 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

18

euic1,...,ck = (qi + xcici∈IC∑ )T (pu + ycu

cu∈UC∑ )+ µ + bi + bu

qi latent factor vector of item ipu latent factor vector of user uIC subset of item-related contextual conditionsxci latent factor vector of contextual condition ciUC subset of user-related contextual conditionsycu latent factor vector of contextual condition cuμ overall average errorbi baseline for item ibu baseline for user u

Page 43: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Adaptive Weighted (2/2)

• 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

18

euic1,...,ck = (qi + xcici∈IC∑ )T (pu + ycu

cu∈UC∑ )+ µ + bi + bu

qi latent factor vector of item ipu latent factor vector of user uIC subset of item-related contextual conditionsxci latent factor vector of contextual condition ciUC subset of user-related contextual conditionsycu latent factor vector of contextual condition cuμ overall average errorbi baseline for item ibu baseline for user u

Page 44: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Adaptive Weighted (2/2)

• 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

18

euic1,...,ck = (qi + xcici∈IC∑ )T (pu + ycu

cu∈UC∑ )+ µ + bi + bu

qi latent factor vector of item ipu latent factor vector of user uIC subset of item-related contextual conditionsxci latent factor vector of contextual condition ciUC subset of user-related contextual conditionsycu latent factor vector of contextual condition cuμ overall average errorbi baseline for item ibu baseline for user u

Page 45: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Adaptive Weighted (2/2)

• 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

18

euic1,...,ck = (qi + xcici∈IC∑ )T (pu + ycu

cu∈UC∑ )+ µ + bi + bu

qi latent factor vector of item ipu latent factor vector of user uIC subset of item-related contextual conditionsxci latent factor vector of contextual condition ciUC subset of user-related contextual conditionsycu latent factor vector of contextual condition cuμ overall average errorbi baseline for item ibu baseline for user u

Page 46: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Hybrid CARS Algorithms Adaptive Weighted (2/2)

• 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

18

euic1,...,ck = (qi + xcici∈IC∑ )T (pu + ycu

cu∈UC∑ )+ µ + bi + bu

qi latent factor vector of item ipu latent factor vector of user uIC subset of item-related contextual conditionsxci latent factor vector of contextual condition ciUC subset of user-related contextual conditionsycu latent factor vector of contextual condition cuμ overall average errorbi baseline for item ibu baseline for user u

Page 47: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Outline

19

• Context-Aware Recommenders and the Cold-Start Problem

• Related Work

• Context-Aware Rating Prediction Models

• Evaluation and Results

• Conclusions and Open Issues

Page 48: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Evaluation Used Datasets

• 3 contextually-tagged rating datasets

20

STS (Braunhofer et al., 2013)

LDOS-CoMoDa (Odić et al., 2013)

Music (Baltrunas et al., 2011)

Domain POIs Movies MusicRating scale 1-5 1-5 1-5Ratings 2,534 2,296 4,012Users 325 121 43Items 249 1,232 139Contextual factors 14 12 8Contextual conditions 57 49 26Contextual situations 931 1,969 26User attributes 7 4 10Item features 1 7 2

Page 49: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Evaluation Evaluation Procedure

• Randomly divide the entities (i.e., users, items or contexts) into ten 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 normalised 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

21

Page 50: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Results Recommendation for New Users

22

MAE

0.00.20.40.60.81.01.21.41.61.82.02.22.4

STS CoMoDa Music

CAMF-CC SPF Content-based CAMF-CCDemographics-based CAMF-CC Average Weighted Heuristic SwitchingAdaptive Weighted

1-nD

CG

@1

0.00.10.20.30.40.50.60.70.80.91.0

STS CoMoDa Music

Page 51: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Results Recommendation for New Items

23

MAE

0.00.10.20.30.40.50.60.80.91.01.11.21.31.4

STS CoMoDa Music

CAMF-CC SPF Content-based CAMF-CCDemographics-based CAMF-CC Average Weighted Heuristic SwitchingAdaptive Weighted

1-nD

CG

@1

0.00.10.20.30.40.50.60.70.80.91.0

STS CoMoDa Music

Page 52: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Results Recommendation under New Contexts

24

MAE

0.00.10.20.30.40.50.70.80.91.01.11.2

STS CoMoDa Music

CAMF-CC SPF Content-based CAMF-CCDemographics-based CAMF-CC Average Weighted Heuristic SwitchingAdaptive Weighted

1-nD

CG

@1

0.00.10.20.30.40.50.60.70.80.91.0

STS CoMoDa Music

Page 53: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Outline

25

• Context-Aware Recommenders and the Cold-Start Problem

• Related Work

• Context-Aware Rating Prediction Models

• Evaluation and Results

• Conclusions and Open Issues

Page 54: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

likes

FREERIDING

ALPING SKIING

likesMUSEUM

MUSEUM

likes

Page 55: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

FREERIDING

ALPING SKIING

likesMUSEUM

MUSEUM

likes

Page 56: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

similar

FREERIDING

ALPING SKIING

likesMUSEUM

MUSEUM

likes

Page 57: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

similar

likely likes

FREERIDING

ALPING SKIING

likesMUSEUM

MUSEUM

likes

Page 58: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

similar

likely likes

FREERIDING

Skiing

ALPING SKIING

likes

Skiing

MUSEUM

MUSEUM

likes

Page 59: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

similar

likely likes

FREERIDING

Skiing

ALPING SKIING

likes

similarSkiing

MUSEUM

MUSEUM

likes

Page 60: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

similar

likely likes

FREERIDING

Skiing

ALPING SKIING

likes

similarlikely likesSkiing

MUSEUM

MUSEUM

likes

Page 61: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

similar

likely likes

FREERIDING

Skiing

ALPING SKIING

likes

similarlikely likesSkiing

MUSEUM

MUSEUM

likes Wet weather

Wet weather

Page 62: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

similar

likely likes

FREERIDING

Skiing

ALPING SKIING

likes

similarlikely likesSkiing

MUSEUM

MUSEUM

likes

similar

Wet weather

Wet weather

Page 63: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

• Various cold-start situations require different CARS solutions

• Hybridisation of several CARS techniques, each of which has its own strengths and weaknesses, allows to achieve best (cold-start) performance

• First developed and tested hybrid CARS algorithms are able to outperform the state-of-the-art CARS algorithms (i.e., CAMF-CC and SPF)

Conclusions

26

SKIING

18-25 Male

18-25 Male

likes

similar

likely likes

FREERIDING

Skiing

ALPING SKIING

likes

similarlikely likesSkiing

MUSEUM

MUSEUM

likes

similarlikely likes

Wet weather

Wet weather

Page 64: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Open Issues

• Review additional knowledge sources which may be used to incorporate additional information about users, items and contextual situations

• Check the availability of large-scale, contextually-tagged datasets with item and user attributes

• Revise the used evaluation procedure and evaluation metrics

• Identify the best-performing hybridisation method for cold-start situations

• Design and execute a live user study

27

Page 65: Hybridisation Techniques for Cold-Starting Context-Aware Recommender Systems

RecSys - October 2014, Foster City, USA

Questions?

Thank you.