Top-k Web Service Composition in the Context of User Preferences - PowerPoint PPT Presentation

Top-k Web Service Composition in the Context of User Preferences Karim Benouaret 1 , Djamal Benslimane 1 , Allel Hadjali 2 1 LIRIS, University of Lyon {karim.benouaret, djamal.benslimane}@liris.cnrs.fr 2 IRISA, Rennes1 University allel.hadjali@enssat.fr 1 er septembre 2011

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Outline Introduction 1 Service composition based preference queries 2 Top-k service composition 3 Experimental evaluation 4 Conclusion 5 Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 2 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Outline Introduction 1 Service composition based preference queries 2 Top-k service composition 3 Experimental evaluation 4 Conclusion 5 Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 3 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Problem description Data Web services • network accessible software entities • returning some information to the user (e.g., a weather forecast service or a news service) Data Web service composition • a combination of primitive Data Web services • answering user’s complex queries User preferences • important to customize the composition process • rank-order the Data Web service compositions • flexible manner : linguistic terms (e.g., “rather cheap" or "‘not expensive") • modeled using fuzzy sets Objective : find the top-k Data Web service compositions according to user preferences Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 4 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Example Service Functionality Constraints Returns the automa- kers y in a given - S 11 ($ x, ? y ) country x S 21 ($ x, ? y, ? z, ? t ) Returns the cars y z is cheap , t is short S 22 ($ x, ? y, ? z, ? t ) z is accessible, t is [12 , 24] along with their prices S 23 ($ x, ? y, ? z, ? t ) z and warranties t for z is expensive, t is long a given automaker x z is [9000 , 14000] , S 24 ($ x, ? y, ? z, ? t ) t is [6 , 24] S 31 ($ x, ? y, ? z ) y is weak , z is small Returns the power y y is ordinary, z is S 32 ($ x, ? y, ? z ) and the consumption z approximately 4 for a given car x S 33 ($ x, ? y, ? z ) y is powerful, z is high S 34 ($ x, ? y, ? z ) y is [60 , 110] , z is [3 . 5 , 5 . 5] Q 1 :“return the French cars, preferably at an affordable price with a warranty around 18 months and having a normal power with a medium consumption" Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 5 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Overview of our approach Challenges • how to retain the most relevant services • how to generate the top-k compositions Contribution • compute matching degrees between user preferences and services’ constraints • propose a ranking criteria based on a fuzzufication of Pareto dominance to select the most relevant services/compositions • to avoid returning similar compositions, we also propose a diversified top-k compositions Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 6 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Outline Introduction 1 Service composition based preference queries 2 Top-k service composition 3 Experimental evaluation 4 Conclusion 5 Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 7 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Terminology Q :- ( q 1 , ..., q n ) : a preference query S = { S 1 , ..., S n } : a set of service classes S i = { S i 1 , ..., S in i } : a set functionally similar services S i ⊑ q i : services of S i can be used to answer q i M = { M 1 , ..., M m } a set of matching methods Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 8 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Matching degrees between services and query components S ij q i CBM G-IBM L-IBM K-IBM S 11 q 1 - - - - (1 , 0 . 57) (1 , 0) (1 , 0) (0 . 80 , 0) S 21 (0 . 89 , 1) (0 , 1) (0 . 90 , 1) (0 . 50 , 1) S 22 q 2 (0 . 20 , 0 . 16) (0 , 0) (0 , 0) (0 , 0) S 23 S 24 (0 . 83 , 0 . 88) (0 . 60 , 0 . 50) (0 . 60 , 0 . 50) (0 . 60 , 0 . 50) (0 . 50 , 0 . 36) (0 , 0) (0 , 0) (0 , 0) S 31 (0 . 79 , 0 . 75) (0 , 0 . 25) (0 . 60 , 0 . 50) (0 . 40 , 0 . 50) S 32 q 3 (0 . 21 , 0 . 64) (0 , 0) (0 , 0) (0 , 0) S 33 (0 . 83 , 0 . 85) (0 . 50 , 0 . 50) (0 . 50 , 0 . 50) (0 . 50 , 0 . 50) S 34 Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 9 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Outline Introduction 1 Service composition based preference queries 2 Top-k service composition 3 Experimental evaluation 4 Conclusion 5 Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 10 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Current approaches Scoring function • computes a score for each service as an aggregate of the individual matching degrees • requires users to assign weights to individual matching degrees • users lose the flexibility to select their desired services • one matching method Skyline • compromises the services which are not nominated • privileges services with a large variance • one matching method Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 11 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Pareto dominance vs fuzzy dominance Pareto dominance : u ≻ v ⇐ ⇒ ∀ i ∈ [1 , d ] , u i ≥ v i ∧ ∃ k ∈ [1 , d ] , u k > v k � d i =1 µ ≫ ( u i ,v i ) Fuzzy dominance : deg ( u ≻ v ) = , where d  0 ifx − y ≤ ε    µ ≫ ( x, y ) = 1 ifx − y ≥ λ + ε x − y − ε otherwise   λ Comparison ( u = (1 , 0) , v = (0 . 90 , 1) ) • neither u ≻ v nor v ≻ u • deg ( u ≻ v ) = 0 . 25 and deg ( v ≻ u ) = 0 . 50 ( ε = 0 , λ = 0 . 2 ) Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 12 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Associating score with a Service/Composition Service’s score : S ij ∈ S i , indicates the average extent to which S ij dominates the whole services of its class S i 1 � m � m ij ≻ S   =1 deg ( S ı FDS ( S ij ) = � ik ) ( |S i |− 1) m 2 ı =1 S ik ∈S i k � = i Composition’s score : C = { S 1 j 1 , ..., S nj n } FDS ( C ) = 1 � n i =1 d i · FDS ( S ij i ) d Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 13 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion An efficient generation of top-k compositions straightforward method : • generate all possible compositions • compute their scores • return the top-k ones • high computational cost Optimization technique ( theorem 1 ) : C = { S 1 j 1 , ..., S nj n } ∃ S ij i ∈ C ; S ij i / ∈ top - k. S i = ⇒ C / ∈ top - k. C . Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 14 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion An efficient generation of top-k compositions (our example) Services Class Score Top-k S 11 S 1 - S 11 S 21 ✟ 0.487 ✟ S 22 0.653 S 22 S 2 S 23 ✟ 0.035 S 24 ✟ S 24 0.538 ✟ 0.094 S 31 ✟ S 32 0.593 S 32 S 3 S 33 ✟ 0.130 S 34 ✟ S 34 0.743 Compositions Score Top-k C 1 = { S 11 , S 22 , S 32 } 0.623 C 2 = { S 11 , S 22 , S 34 } 0.698 C 2 C 3 = { S 11 , S 24 , S 32 } 0.566 C 4 C 4 = { S 11 , S 24 , S 34 } 0.640 Straightforward method : 16 compositions ( � n i i =1 |S i | ) Our method : 4 compositions( ≤ k n i ) Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 15 / 21

Introduction Service composition based preference queries Top-k service composition Experimental evaluation Conclusion Diversity-aware Top-k Compositions Different similar services could exist in each class S i leading to similar compositions Diversification is then needed to improve user satisfaction Quality ( S ij ) = FDS ( S ij ) × RelDiv ( S ij , dtopk. S i ) RelDiv ( S ij , dtopk. S i ) = � 1 dtopk. S i = ∅ � Sik ∈ dtopk. S i Dist ( S ij ,S ik ) otherwise | dtopk. S i | Karim Benouaret/Top-k Web Service Composition in the Context of User Preferences 16 / 21