SPARA - A recommender system for exploratory browsing
-
Upload
ahmad-assaf -
Category
Technology
-
view
365 -
download
0
description
Transcript of SPARA - A recommender system for exploratory browsing
A Recommender System for Exploratory Browsing
Team: SPARATutor: Tommaso Di Noia
• Exploratory browsing– Multiple domains– Serendipitous
Following links
Content-based Recommender Systems
Recommender System
User profile
…
Top-N Recommendations
Item1, 5Item2, 1Item5, 4Item10, 5….
Items Movie D
Movie K
PropertiesDirector: Steven SpielbergStarring: Justin Bieber…
….
CB-RSs recommend items to a user based on their description and on the profile of the user’s interests
Item RatingMovie A 9/10
Actor J 5/10
Movie C 7/10
Movie A
Slide: Tommaso Di Noia
RecommendationMovie 1
Movie 2
Movie 3
The LOD Aspects
• Subgraph of DBpedia with different types of items (Film, Person, Book) using SPARQL– Rich set of features - DBpedia properties
associated to a type– Links to related items
• Generate user profiles with ratings from IMDB– IMDB to DBpedia URI mapping– Inferring ratings for persons related to movies
• Feed the data to the recommender system
USER INTERFACES
Bertin’s Visual Attributes
Bertin, Semiology of Graphics, 83 Slide: Sheelagh Carpendale
Importance Ordering: Perceptual Properties
Slide: Cecilia Aragon, HCDE, UWMackinlay, APT (A Presentation Tool), 1986
Movies Perspective
The Lost World: Jurassic ParkThe Lost World: Jurassic Park, is a 1997 American science fiction adventure film directed by Steven Spielberg and the second of the Jurassic Park franchise. The film was produced by Gerald R. Molen and Colin Wilson.
Director: Steven SpielbergStarring: Jeff Goldblum, Julianne Moore, Richard Attenborough, Vince Vaughn, Pete Postlethwaite
DEMO
FUTURE WORK
Parameter Tuning
Righteous Kill
starringdirector
subject/broadergenre
Heat
Robe
rt D
e N
iro
John
Avn
etSe
rial k
iller
film
s
Dra
ma
Al P
acin
oBr
ian
Den
nehy
Hei
st fi
lms
Crim
e fil
ms
starring
Robe
rt D
e N
iroAl
Pac
ino
Bria
n D
enne
hy
Righteous KillHeat
… …
Slide: Tommaso Di Noia
Vector Space Model for LOD
+
+
+
… =
𝒔𝒊𝒎𝒔𝒕𝒂𝒓𝒓𝒊𝒏𝒈 ( �⃗�𝒊 , �⃗� 𝒋)=𝒘𝒗𝟏 ,𝒙 𝒊
∗𝒘𝒗𝟏 ,𝒙 𝒋+𝒘𝒗 𝟐 ,𝒙𝒊
∗𝒘𝒗 𝟐 ,𝒙 𝒋+𝒘 𝒗𝟑 ,𝒙 𝒊
∗𝒘 𝒗𝟑 ,𝒙 𝒋
√𝒘 𝒗𝟏 ,𝒙 𝒊
𝟐 +𝒘𝒗𝟐 ,𝒙 𝒊
𝟐 +𝒘𝒗𝟑 ,𝒙𝒊
𝟐 ∗√𝒘𝒗 𝟏 ,𝒙 𝒋
𝟐 +𝒘 𝒗𝟐 ,𝒙 𝒋
𝟐 +𝒘 𝒗𝟑 ,𝒙 𝒋
𝟐
𝜶𝒔𝒕𝒂𝒓𝒓𝒊𝒏𝒈∗𝒔𝒊𝒎𝒔𝒕𝒂𝒓𝒓𝒊𝒏𝒈 ( �⃗� 𝒊 , �⃗� 𝒋)
𝜶𝒅𝒊𝒓𝒆𝒄𝒕𝒐𝒓∗ 𝒔𝒊𝒎𝒅𝒊𝒓𝒆𝒄𝒕𝒐𝒓 ( �⃗� 𝒊 , �⃗� 𝒋)
𝜶𝒔𝒖𝒃𝒋𝒆𝒄𝒕∗ 𝒔𝒊𝒎𝒔𝒖𝒃𝒋𝒆𝒄𝒕 ( �⃗� 𝒊 , �⃗� 𝒋)
𝒔𝒊𝒎𝒔𝒕𝒂𝒓𝒓𝒊𝒏𝒈 ( �⃗�𝒊 , �⃗� 𝒋)Slide: Tommaso Di Noia
Evaluate
• Experimenting with the αp parameters– Learning αp
• Quality of the recommendations– Discounted cumulative gain– Kendall tau rank correlation coefficient
• Precision and Recall• Precision = true positive/ number of predicted positive• Recall = true positive / number of actual positive
• Evaluating the user interfaces• Explanations– Generating explanations– Evaluating explanations
Thank You