hybrid event recommendation using linked data and user diversity

22
Hybrid Event Recommendation using Linked Data and User Diversity Houda Khrouf and Raphaël Troncy {khrouf, troncy}@eurecom.fr Eurecom, Sophia Antipolis, France The 7 th ACM Recommender Systems Conference Oct 12-16, 2013 Hong Kong

Upload: houda-khrouf

Post on 24-May-2015

1.247 views

Category:

Health & Medicine


1 download

DESCRIPTION

Paper presented in RecSys ACM conference 2013

TRANSCRIPT

Page 1: Hybrid Event Recommendation using Linked Data and User Diversity

Hybrid Event Recommendation using

Linked Data and User Diversity Houda Khrouf and Raphaël Troncy

{khrouf, troncy}@eurecom.fr

Eurecom, Sophia Antipolis, France

The 7th ACM Recommender Systems Conference

Oct 12-16, 2013 Hong Kong

Page 2: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 2

Outline

Event Recommendation Collaborative Filtering

Content-based

RDF Modeling and Similarity computation

User Interest Detection

Hybrid Approach

Evaluation and Conclusion

10/15/2013

Page 3: Hybrid Event Recommendation using Linked Data and User Diversity

3

Events on the web

10/15/2013

Millions of active users

Thousands of events per day

Highly diverse content

Recommender Systems?

7th ACM Recommender Systems 2013, Hong Kong

Page 4: Hybrid Event Recommendation using Linked Data and User Diversity

What do users think?

10/15/2013 7th ACM Recommender Systems 2013, Hong Kong 4

Seen on Last.Fm

Page 5: Hybrid Event Recommendation using Linked Data and User Diversity

EVENTS

7th ACM Recommender Systems 2013, Hong Kong 5

Is this event interesting?

10/15/2013

Attendees

Places Time

Tags/Topics

Performers

Decision

Decision factors (depends on type) • Where is it? (Location)

• Who’s going? (Participants)

• When is it? (Time)

• What is it? (Content)

• Who is involved? (players)

Page 6: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 6

Collaborative Filtering (CF)

10/15/2013

Predict the event the user will attend

based on the attendance of other like minded users

Best choice to reflect the social dimension, but:

Events are transient items

inducing a very sparse user attendance matrix (sparsity 99%)

Apart from the social information, there is no explicit consideration of the other factors

sim

ilar

Page 7: Hybrid Event Recommendation using Linked Data and User Diversity

Events are entities with attributes and relational attributes (links) to other entities

7th ACM Recommender Systems 2013, Hong Kong 7

Content-based Recommendation (CB)

10/15/2013

Recommend new events that match the user profile based on

their descriptions Event context:

- Location (geo-coordinates, city…)

- Time

- Topics/Tags

- Performers (genres, tags…)

Events similarity depends on the similarity of related entities

Page 8: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 8

User Profile

10/15/2013

The user profile is based on past attended events

Topical Diversity: real-world events range from large festivals to small concerts and social gatherings

A user might be interested in some specific topics/performers during the event

We need to alleviate the profile diversity and detect the user’s interests

Page 9: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 9

Approach and Contributions

10/15/2013

Events similarity

Structured RDF event model

Similarity in Linked Data

Data enrichment with DBpedia

User interests detection using LDA (Latent Dirichlet

Allocation) Hybrid recommendation (CF+CB)

Page 10: Hybrid Event Recommendation using Linked Data and User Diversity

10/15/2013 7th ACM Recommender Systems 2013, Hong Kong 10

LODE Ontology

LODE is a minimal model that encapsulates the factual properties of events: What,

Where, When and Who. URL: http://linkedevents.org/ontology

Page 11: Hybrid Event Recommendation using Linked Data and User Diversity

Linked Data in a Tensor Space

10/15/2013 7th ACM Recommender Systems 2013, Hong Kong 11

For each property p, and for each object op [1] 𝑾𝑾 𝒐𝒐,𝒆𝒆 𝒑𝒑 = 𝒇𝒇 𝒐𝒐,𝒆𝒆

𝒑𝒑 ∗ 𝒍𝒍𝒐𝒐𝒍𝒍|𝑬𝑬|

|𝑬𝑬𝒐𝒐,𝒑𝒑|

subj

ects

objects

[1] T. Di Noia et al. Linked open data to support content-based recommender systems. In 8th International Conference on Semantic Systems, Graz, Austria, 2012.

Page 12: Hybrid Event Recommendation using Linked Data and User Diversity

Events Similarity

10/15/2013 7th ACM Recommender Systems 2013, Hong Kong 12

Similarity between two events:

Similarity according to one property p:

Not adapted for discriminant properties associated with highly sparse adjacency matrix

𝐜𝐜𝐜𝐜𝐜𝐜𝐜𝐜𝐜𝐜𝐩𝐩 𝐞𝐞𝟏𝟏, 𝐞𝐞𝟐𝟐 = ∑ 𝒘𝒘𝒐𝒐,𝐞𝐞𝟏𝟏

𝒑𝒑 ∗ 𝒘𝒘𝒐𝒐,𝐞𝐞𝟐𝟐𝒑𝒑 𝒐𝒐∈𝑶𝑶

∑ 𝒘𝒘𝒐𝒐,𝐞𝐞𝟏𝟏𝒑𝒑 𝟐𝟐

𝒐𝒐∈𝑶𝑶 ∗ ∑ 𝒘𝒘𝒐𝒐,𝐞𝐞𝟐𝟐𝒑𝒑 𝟐𝟐

𝒐𝒐∈𝑶𝑶

𝒔𝒔𝒔𝒔𝒔𝒔 𝒆𝒆𝟏𝟏, 𝒆𝒆𝟐𝟐 = ∑ 𝜶𝜶𝒑𝒑 𝒄𝒄𝒐𝒐𝒔𝒔𝒔𝒔𝒔𝒔𝒑𝒑 𝒆𝒆𝟏𝟏, 𝒆𝒆𝟐𝟐 𝒑𝒑∈𝑷𝑷

|𝑷𝑷|

Page 13: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 13

Events Similarity

Discriminability

Similarity-based Interpolation

10/15/2013

e

o1 p

o2

similar

Interpolation of a fictitious link

p

𝑫𝑫𝒔𝒔𝒔𝒔𝒄𝒄 𝒑𝒑 = 𝒐𝒐 𝒕𝒕𝒕𝒕𝒔𝒔𝒑𝒑𝒍𝒍𝒆𝒆 = < 𝒔𝒔,𝒑𝒑,𝒐𝒐 > ∈ 𝑮𝑮 |

|𝒕𝒕𝒕𝒕𝒔𝒔𝒑𝒑𝒍𝒍𝒆𝒆 = < 𝒔𝒔,𝒑𝒑,𝒐𝒐 > ∈ 𝑮𝑮|

𝑾𝑾 𝒐𝒐𝟐𝟐,𝒆𝒆 𝒑𝒑 = 𝐜𝐜𝐦𝐦𝐦𝐦𝐜𝐜𝐜𝐜𝐜𝐜

𝒐𝒐∈𝑶𝑶𝒑𝒑,𝒆𝒆(𝒐𝒐𝟐𝟐,𝒐𝒐) ∗ 𝒍𝒍𝒐𝒐𝒍𝒍

|𝑬𝑬||𝑬𝑬𝒐𝒐𝟐𝟐,𝒑𝒑|

Page 14: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 14

Interest Detection

10/15/2013

How to detect user interests from diverse event space?

Latent Dirichlet Allocation (LDA) [Blei et al 2003]

Events

Tags

Topic distribution over each event (T=30)

Attended events Eu

Variance of each topic

User Interest Distribution

Diversity score

Mean of the variances

Page 15: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 15

Hybrid Recommendation

10/15/2013

Content-based rank:

Hybrid rank

CF rank: Common events between u and RSVP users

αp = property weight βp = interest weight λ cf = CF weight

𝒕𝒕 𝒖𝒖,𝒆𝒆 = 𝒕𝒕𝒄𝒄𝒄𝒄++ 𝒖𝒖,𝒆𝒆 + 𝝀𝝀𝒄𝒄𝒇𝒇 𝒕𝒕𝒄𝒄𝒇𝒇 𝒖𝒖,𝒆𝒆

𝒕𝒕𝒄𝒄𝒄𝒄++ 𝒖𝒖, 𝒆𝒆 =∑ ∑ 𝜶𝜶𝒑𝒑 𝜷𝜷𝒑𝒑 𝒄𝒄𝒐𝒐𝒔𝒔𝒔𝒔𝒔𝒔𝒑𝒑(𝒆𝒆𝒔𝒔 , 𝒆𝒆)𝒑𝒑∈ 𝑷𝑷𝒆𝒆𝒔𝒔 ∈ 𝑬𝑬𝒖𝒖

𝑷𝑷 ∗ |𝑬𝑬𝒖𝒖|

Page 16: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 16

Experiments

10/15/2013

Open RDF Dataset (EventMedia) Visualization: http://eventmedia.eurecom.fr

SPARQL: http://eventmedia.eurecom.fr/sparql

Learning the rank weights: Linear regression with gradient descent

Genetic Algorithm

Particle Swarm Optimization

Evaluation: training 70% - test 30 %

2.436 events in UK from Last.Fm , 481 active users, 14.748 artists, 897 locations (available on request)

precision/recall of Top-N recommendations

Page 17: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 17

Sparsity Reduction

10/15/2013

location agent subject Without processing 0.9942 0.9174 0.3175

Similarity Interpolation 0.6854 0.7392 -

DBpedia enrichment - - 0.2843

Sparsity rates of adjacency matrices

Page 18: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 18

User Diversity

10/15/2013

Most of users have relatively high interests towards some topics

Score ≈ 1 => strong interest Score ≈ 0 => cold-start users

Page 19: Hybrid Event Recommendation using Linked Data and User Diversity

10/15/2013 7th ACM Recommender Systems 2013, Hong Kong 19

Learning weights evaluation

PSO has better performance

Page 20: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 20

CB+CF evaluation

10/15/2013

Interest Detection 𝛃𝛃𝐜𝐜𝐢𝐢𝐢𝐢𝐞𝐞𝐢𝐢𝐞𝐞𝐜𝐜𝐢𝐢 >

𝟒𝟒 × 𝛃𝛃𝐢𝐢𝐜𝐜−𝐜𝐜𝐢𝐢𝐢𝐢𝐞𝐞𝐢𝐢𝐞𝐞𝐜𝐜𝐢𝐢

High influence of social information in event recommendation

Page 21: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 21

Comparison with other approaches

10/15/2013

Probability based Extended Profile Filtering (UBExtended): T. D. Pessemie et al. Collaborative recommendations with content-based filters for cultural activities via a scalable event distribution platform. Multimedia.Tools Appl., 58(1):167-213, 2012.

Page 22: Hybrid Event Recommendation using Linked Data and User Diversity

7th ACM Recommender Systems 2013, Hong Kong 22

Conclusion

Effectiveness of Semantic Web technologies to steer data retrieval and processing

Importance of the social information and the user interest model in event recommendation

Future work:

Other features: popularity, temporal patterns, weather, etc…

Test the system scalability on large datasets using spatial and/or

temporal indexing of user attendance

10/15/2013