June 5 2014 Recommender ommender Systems ms and SPAR ARQL: QL: More e than n a Shotgun tgun Weddin ing? 8 th Alberto Mendelzon International Workshop Cartagena, Colombia (AMW 2014) Victor Anthony Arras ascue Ayala Martin Przyjac aciel-Zab ablocki Thomas as Hornu nung ng Alexan ander Schätzle Georg Lausen University of Freiburg Databases & Information Systems
Overview erview 1. Motivation 2. RecSPARQL 3. Experiments 4. Summary 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 2
Motivat ivatio ion RDF’s flexibility ◦ Recommendation domain Users, Items, Ratings Consumption relationship 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 3
Motivat ivatio ion How to get recommendations from RDF-graphs: SPARQL? ◦ Retrieval of explicit data from RFD-graphs ◦ Graph pattern matching ◦ Flexible ◦ No possible to express fuzzy queries Similarity SPARQL Source: http://courses.ischool.berkeley.edu/i253/f11/ 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 4
Motivat ivatio ion How to get recommendations from RDF-graphs: Recommender Systems ◦ Predicts a degree of preference for a user towards a set of non-consumed items ◦ Based on Information Retrieval techniques Recommender Systems Source: http://courses.ischool.berkeley.edu/i253/f11/ 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 5
Motivat ivatio ion A straightforward approach Classic recommender ◦ Inpu put: users, items, ratings ◦ Outpu put: users, recommended items, predicted rating Recommendations 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 6
Motivat ivatio ion Example: lack of flexibility ◦ Collaborative filtering approach Recommendations from similar users 1 3 5 0 2 1 0 4 1 1 3 2 0 0 3 0 3 4 1 5 1 3 5 0 1 2 1 2 3 2 (A) (B) Extraction / Recommendation Pre-processing System (CF) SPARQL 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 7
Motivat ivatio ion Example: lack of flexibility ◦ Collaborative filtering approach Recommendations from similar users U 5 1 3 5 0 2 U 17 1 0 4 1 1 3 2 0 0 3 U 1 0 3 4 1 5 1 3 5 0 1 U 43 2 1 2 3 2 U 55 (B) Recommendation System (CF) 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 8
Motivat ivatio ion Example: lack of flexibility ◦ Collaborative approach Recommendations from similar users ◦ Customization Neighbors geographically close U 5 Age of neighbors differ by 𝜀 Speak same languages U 17 etc…. U 1 U 43 U 55 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 9
Motivat ivatio ion Example: lack of flexibility ◦ Collaborative approach Recommendations from similar users ◦ Customization Neighbors geographically close U 5 Age of neighbors differ by 𝜀 Speak same languages U 17 etc…. U 1 U 43 Where is the problem? U 55 ◦ Fixed recommender model 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 10
Overview erview 1. Motivation 2. RecSPARQL 3. Experiments 4. Summary 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 11
Our Ap Approach oach RecSPARQL: Recommendations + SPARQL ◦ Extension of SPARQL 1.1 Consistent mechanism to select parts of the graph Flexibility RecSesame ◦ Platform for the evaluation of RecSPARQL queries Based on Sesame’s framework Cache System 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 12
RecSPA cSPARQ RQL in a nutshe shell RECOMMEND [Projected Variables] USING [Recommendation Algorithm] WHERE { [Basic Graph Pattern] } BASED ON { [RecSPARQL Type Pattern] [RecSPARQL Model Building Pattern] } 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 13
RecSPA cSPARQ RQL in a nutshe shell RECOMMEND [Projected Variables] USING [Recommendation Algorithm] ◦ Must contain WHERE { [Basic Graph Pattern] } recommendation BASED ON { [RecSPARQL Type Pattern] entities [RecSPARQL Model Building Pattern] } 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 14
RecSPA cSPARQ RQL in a nutshe shell RECOMMEND [Projected Variables] ◦ Algorithm used USING [Recommendation Algorithm] WHERE { [Basic Graph Pattern] } BASED ON { [RecSPARQL Type Pattern] [RecSPARQL Model Building Pattern] } Algorihtms ◦ Content-based (CB) ◦ Collaborative Filtering (CF) ◦ Hybrid (H) 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 15
RecSPA cSPARQ RQL in a nutshe shell RECOMMEND [Projected Variables] USING [Recommendation Algorithm] WHERE { [Basic Graph Pattern] } BASED ON { [RecSPARQL Type Pattern] ◦ Similarity [RecSPARQL Model Building Pattern] } criteria Content-based Collaborative filtering ◦ Genre ◦ Watched movies and ratings ◦ Cast ◦ Age ◦ Director ◦ Geographical location 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 16
RecSPA cSPARQ RQL in a nutshe shell RecSPARQL in a nutshell ◦ Similar to RECOMMEND [Projected Variables] projection USING [Recommendation Algorithm] ◦ Makes it possible WHERE { [Basic Graph Pattern] } to project BASED ON { [RecSPARQL Type Pattern] recommendations [RecSPARQL Model Building Pattern] } ◦ ?movi vie.REC 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 17
RecSPARQL’s flexi xibi bility lity Filters ◦ Recommend only action movies FILTER ( ?genre.REC = “action”) ?genre.REC EC ?genre.REC EC 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 18
RecSPARQL’s flexi xibi bility lity Filters ◦ Recommend only from users whose age differ by at most 5 years FILTER ( abs(xsd:integer(?age) - xsd:integer(?age.REC)) <= 5 ) ?age ?age.REC EC 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 19
RecSPARQL’s flexi xibi bility lity Much more… ◦ Recommend movies watched under a certain context FILTER ( ?watchTime.REC = “weekend” && ?company.REC = “partner”) . ◦ Recommend movies whose directors have the same citizenship of the user for which we want the recommendations ◦ Recommend …. 2. Appr proach Recommender Systems and SPARQL: More than a Shotgun Wedding? 20
Experi riment ments 1. Motivation 2. RecSPARQL 3. Experiments 4. Summary 4. Experime ments Recommender Systems and SPARQL: More than a Shotgun Wedding? 21
Experi riment ments Example: lack of flexibility ◦ Collaborative approach Recommendations from similar users U 5 U 17 Gradual restriction of U 1 Neighborhood: U 43 ◦ Neighbors geographically close ◦ Age of neighbors differ by 𝜀 U 55 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 22
Experi riment ments Example: lack of flexibility ◦ Collaborative approach Recommendations from similar users U 5 U 17 FILTER ( U 1 abs(xsd:integer(?age) - xsd:integer(?age.REC)) <= %K% ) . U 43 FILTER ( abs(xsd:integer(?zip) - U 55 xsd:integer(?zip.REC)) <= %L% ) 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 23
Experi riment ments Benefitial ◦ Backed by our experiments ◦ Customization ◦ Similarity among users in the neighborhood Neighbors geographically close Increases Age of neighbors differ by 𝜀 U5 U17 U1 U43 U55 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 24
Experi riment ments Benefitial ◦ Backed by our experiments ◦ Customization ◦ Similarity among users in the neighborhood Neighbors geographically close Increases Age of neighbors differ by 𝜀 U5 U17 1 3 5 0 2 U1 1 0 4 1 1 3 2 0 0 3 U43 0 3 4 1 5 1 3 5 0 1 2 1 2 3 2 U55 1. Motivation Recommender Systems and SPARQL: More than a Shotgun Wedding? 25
Summary mary Tight integration of recommender systems with SPARQL Customizable recommendations on arbitrary RDF graphs Futu ture e Work ◦ Enhance the integration of both paradigms ◦ Support more recommendations techniques ◦ Increase the expressiveness of RecSPARQL Binding variables Sub-queries Property paths V. A. Arrascue Ayala, M. Przyj yjaciel-Zablocki, T. Hornung ung, A. Schä hätzle, G. Laus usen, n, Extend ending ng SPARQL QL for Recommend endations ns. In SWIM M (ACM M SIGMOD MOD), pages 1-8, 2014. 5. Summa mmary Recommender Systems and SPARQL: More than a Shotgun Wedding? 26
Recommend
More recommend