Contribution to proactivity in mobile context-aware recommender systems

Dissertation document: http://oa.upm.es/21913/

Recommender systems are powerful information filtering tools which offer users personalized suggestions about items whose aim is to satisfy their needs. Traditionally the information used to make recommendations has been based on users’ ratings or data on the item’s consumption history and transactions carried out in the system. However, due to the remarkable growth in mobile devices in our society, new opportunities have arisen to improve these systems by implementing them in ubiquitous environments which provide rich context-awareness information on their location or current activity. Because of this current all-mobile lifestyle, users are socially connected permanently, which allows their context to be enhanced not only with physical information, but also with a social dimension. As a result of these novel contextual data sources, the advent of mobile Context-Aware Recommender Systems (CARS) as a research area has appeared to improve the level of personalization in recommendation. On the other hand, this new scenario in which users have their mobile devices with them all the time offers the possibility of looking into new ways of making recommendations. Evolving the traditional user request-response pattern to a proactive approach is now possible as a result of this rich contextual scenario. Thus, the key idea is that recommendations are made to the user when the current situation is appropriate, attending to the available contextual information without an explicit user request being necessary. This dissertation proposes a set of models, algorithms and methods to incorporate proactivity into mobile CARS, while the impact of proactivity is studied in terms of user experience to extract significant outcomes as to "what", "when" and "how" proactive recommendations have to be notified to users. To this end, the development of this dissertation starts from the proposal of a general architecture for building mobile CARS in scenarios with rich social data along with a new way of managing a recommendation process through a REST interface to make this architecture multi-device and cross-platform compatible. Details as regards its implementation and evaluation in a Spanish banking scenario are provided to validate its usefulness and user acceptance. After that, a novel model is presented for proactivity in mobile CARS which shows the key ideas related to decide when a situation warrants a proactive recommendation by establishing algorithms that represent the relationship between the appropriateness of a situation and the suitability of the candidate items to be recommended. A validation of these ideas in the area of e-learning authoring tools is also presented. Following the previous model, this dissertation presents the design and implementation of new mobile user interfaces for proactive notifications. The results of an evaluation among users testing these novel interfaces is also shown to study the impact of proactivity in the user experience of mobile CARS, while significant factors associated to proactivity are also identified. The last stage of this dissertation merges the previous outcomes to design a new methodology to calculate the appropriateness of a situation so as to incorporate proactivity into mobile CARS. Additionally, this work provides details about its validation in a European e-learning social network in which the whole architecture and proactive recommendation model together with its methods have been implemented. Finally, this dissertation opens up a discussion about the conclusions obtained throughout this research, resulting in useful information from the different design and implementation stages of proactive mobile CARS.

Transcript

1. CONTRIBUTION TO PROACTIVITY IN MOBILE CONTEXT-AWARE RECOMMENDER SYSTEMS Author: Daniel Gallego Vico Director: Gabriel Huecas Fernández-Toribio
2. 2 ? “Information overload occurs when a person is exposed to more information than the brain can process at one time” [Palladino, 2007]
6. 6 Recommender Systems are powerful information filtering tools providing suggestions for items to be of use to a user Recommender System Item Filter available items Request recommendation Personalized recommendation
7. 7 “75% of what people watch is from some sort of recommendation” [Amatriain, 2012] “We use recommendation algorithms to personalize the online store for each customer” [Linden et al., 2003]
8. 8 But items are just the beginning…
9. 9 Social data information can be used to increase the level of personalization Rich user profile Behavior Tastes Consumption trends Social links …
10. 10 Item User Transaction Social relationship (following…) Transaction Action in the system (request…) with a specific context Transaction Related to item (feedback, tag…) with a specific context User Recommender System Profile Features Profile
11. 11 Location “Recommendation techniques can increase the usability of mobile systems providing personalized and more focused content” [Ricci, 2010] Device used Ubiquitous environment Activity
12. 12 Context-aware Recommender Systems (CARS) [Adomavicius & Tuzhulin, 2005] [Verbert et al., 2012] “Context: any information used to characterize the situation of an entity“ “Entity: person, place, or object considered relevant to the interaction between user and application, including the user and application themselves” [Dey, 2001] Mobile CARS
13. Index • Motivation and open challenges • Research methodology • Contributions • Conclusions 13
14. Proactivity Recommendations are made to the user • when the current situation is appropriate • without the need for an explicit request 14
15. Research question 15 How could proactivity be incorporated into current mobile CARS and what are the UX implications?
16. Open challenges 1. What kind of architecture is suitable for building mobile CARS in scenarios with rich social data? 2. How can proactivity be incorporated into mobile CARS? 3. Which UX factors need to be considered in the implementation of proactive mobile CARS? 16
17. Index • Motivation and open challenges • Research methodology • Contributions • Conclusions 17
18. Design science: guidelines 1. Design as an artifact 2. Problem relevance 3. Design evaluation 4. Research contributions 5. Research rigor 6. Design as a search process 7. Communication of research 18 [Hevner et al. 2004]
19. Research chronology 19 Bankinter RecSys GLOBAL excursion Research stay at TUM 2010 2011 2012 2013 ICCT JITEL ISTP RecSys PEMA ICDS MUSIC FIE WCCIT FIE IJCISIM EPI JSA JET Conferences Journals
20. Index • Motivation and open challenges • Research methodology • Contributions • Conclusions 20
21. 21 Architecture for social mobile CARS Model for proactivity in mobile CARS Proactivity impact in mobile CARS user experience Methods to incorporate proactivity into mobile CARS
22. 22 Architecture for social mobile CARS
23. Objectives • Architecture • Mash-up of several social sources for recommendation • Privacy • Multi-device and Cross-platform • Suitable contextual recommendation model • Social data analysis available at recommendation time • Avoid cold start problem • Validation 23
24. Architecture 24 Mobile app Social Data Source 1 Social Data Source N Social Data Source 3 Social Data Source 2 Data Anonymization Contextual information Display recommendation User feedback User profiles Transactions Items Application Manager REST API Target user profile Request recommendation Personalized recommendation Recommender Server Client
25. Architecture 25 Mobile app Social Data Source 1 Social Data Source N Social Data Source 3 Social Data Source 2 Data Anonymization Contextual information Display recommendation User feedback User profiles Transactions Items Application Manager REST API Target user profile Request recommendation Personalized recommendation Recommender Server Client
26. Recommendation model 26 Recommender Mobile User Interface User Profiles Item, Transactions Target User Profile Mobile Device Context Information User Profile Clustering Items Assignment User’s Cluster Discovery Social Clusters Clusters Trends Map Phase I: Social Context Generation User’s Location Acquisition Location based Filtering Phase II: Location Context Filtering Phase III: User Context Filtering User context Ranking Generation Geo-Located User’s Cluster Trends Map Display Recommendation Personalized Recommendation User Feedback User’s Cluster Trends Map User’s location
27. Validation: banking scenario • “Perdidos en la Gran Ciudad” project • Objective: • Recommend places using banking data • Place • Entities with credit card payments • Restaurants, supermarkets, cinemas, stores… 27
28. Evaluation: demographics • Deployed in Bankinter Labs environment • Banking data sample • 2.5 million credit card transactions • 222,000 places information • 34,000 anonymous customer’s profiles • 57% male, 43% female • Average age: 51 • Average expense per year: 11,719€ 28
29. Evaluation: social clustering results 29 0 10000 20000 30000 40000 50000 60000 70000 1 2 3 4 5 6 7 8 9 10 11 Cluster size (users) Average credit card expense per year (€)
30. Validation: publications • International journals 1. Generating Awareness from Collaborative Working Environment using Social Data Daniel Gallego, Iván Martínez and Joaquín Salvachúa. IJCISIM, 2012 • International conferences 1. An Empirical Case of a Context-aware Mobile Recommender System in a Banking Environment Daniel Gallego and Gabriel Huecas. MUSIC, 2012 2. Generating Context-aware Recommendations using Banking Data in a Mobile Recommender System. Daniel Gallego, Gabriel Huecas and Joaquín Salvachúa. ICDS, 2012 30
31. 31 Model for proactivity in mobile CARS
32. Objectives • Generality • Proactive and request-response recommendations supported • Relationship between appropriateness situation and item suitability • Feedback to learn user behavior 32
33. User Interface Recommendation model 33 Abort (S1 = 0) Phase I: Situation When to make a recommendation? Phase II: Items Which item(s) to recommend? Forced Rec. (S1 = 1) Score S1 S1 > threshold T1? Calculate score for each candidate item • Apply contextual or non- contextual recommender system S2 > threshold T2? Calculate score by weighted combination of context attributes • User Context • Temporal Context • Geographical Context • Social Context Score S1 influences T2 Feedback influences T1&T2 Display Recommendation How to show the recommendation? Score S2 User Interface Recommender 2 1 1
34. Validation: publications • International indexed journal 1. Proactivity and context-awareness: future of recommender systems design Daniel Gallego, Antonio Fumero and Gabriel Huecas. EPI, 2013. • International conferences 1. Enhanced Recommendations for e-Learning Authoring Tools based on a Proactive Context-aware Recommender Daniel Gallego, Enrique Barra, Aldo Gordillo and Gabriel Huecas. FIE, 2013 2. A Model for Proactivity in Mobile, Context-aware Recommender Systems Wolfgang Woerndl, Johannes Huebner, Roland Bader, and Daniel Gallego. RecSys, 2011 34
35. 35 Proactivity impact in mobile CARS user experience
36. Objectives • How to show proactive recommendations? • Design suitable user interfaces to generate proactive recommendations in mobile CARS • Develop a mobile CARS following the previous model • Empirical evaluation among users • Extract valuable outcomes about proactivity impact in mobile CARS user experience 36
37. Proactive notification: Status bar 37 • Based on Android push notifications • Stimulus • Visual  • Acoustic  • Tactile  • User feedback • Ignore • Not now • Expand
38. Proactive notification: Widget 38 • Based on Android app widgets • Always visible in Home screen • Stimulus • Visual  • User feedback • Ignore • Not now • Expand
39. Evaluation: description • Objective • Evaluate UX of proactive mobile interfaces • Scenario: restaurant recommendations • On-line survey to • Compare both mobile proactive approaches • A/B testing methodology • Study the acceptance of the mobile result visualization methods 39
40. Evaluation: demographics • 58 test users • 72% male, 28% female • Average age 29 • 76% owned a smartphone • Split up randomly into • group evaluated first Status bar, then Widget • group evaluated first Widget, then Status bar 40
41. Results: proactivity impact • Scenarios S1 and S2 • Compare users reaction whether a necessity exists or not • Application ignored when recommendation not needed • But users give feedback on their active rejection 41 Status bar Widget
42. Results: proactivity impact • Scenarios S3 and S4 • Compare users reaction whether they are in a hurry or not • Time pressure situations  poor user feedback • Activity influences appropriateness of proactive recommendations 42 Status bar Widget
43. Results: proactivity impact • Scenarios S5 and S6 • Compare users reaction with user-request approach • In situations corresponding to traditional recommendation scenarios proactive ones have also high acceptance 43 Status bar Widget
44. Results: comparison • Widget solution considered better to achieve proactivity by users • When comparing both, Widget always had higher acceptance 44 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% W more easy to use than SB W less annoying than SB W saves more time/clicks than SB W more comfortable than SB Overall, W better than SB 1 (totally disagree) 2 (disagree) 3 (similar) 4 (agree) 5 (totally agree)
45. Validation: publications • International indexed journal 1. Evaluating the Impact of Proactivity in the User Experience of a Context-aware Restaurant Recommender for Android Smartphones. Daniel Gallego, Wolfgang Woerndl and Gabriel Huecas. JSA, 2013 • International conference 1. A Study on Proactive Delivery of Restaurant Recommendations for Android Smartphones. Daniel Gallego, Wolfgang Woerndl and Roland Bader. PeMA at RecSys, 2011. 45
46. 46 Methods to incorporate proactivity into mobile CARS
47. Objectives • Include proactivity in the recommendation model for mobile CARS in social scenarios • Define context-aware methods to calculate the appropriateness of a situation • Validation in real scenario 47
48. Merged recommendation model 48 S1 > T1? Recommender User Interface Phase II: Situation Assessment Display Recommendation Location Context User Context Score S1 Influences T2 Abort (S1 = 0) Forced Rec. (S1 = 1) Score S1 Phase III: Item Assessment Context Information User Feedback Personalized Recommendation User Profiles Items, Transactions Target User Profile User Profile Clustering User’s Cluster Discovery Social Clusters Clusters Trends Map Phase I: Social Context Generation User’s Cluster Trends Map Location Context Rating User Context Rating User’s Cluster Trends Map S2 > T2? Rated Items (Score S2) Located User’s Cluster Trends Map Items Assignment
49. Merged recommendation model 49 S1 > T1? Recommender User Interface Phase II: Situation Assessment Display Recommendation Location Context User Context Score S1 Influences T2 Abort (S1 = 0) Forced Rec. (S1 = 1) Score S1 Phase III: Item Assessment Context Information User Feedback Personalized Recommendation User Profiles Items, Transactions Target User Profile User Profile Clustering User’s Cluster Discovery Social Clusters Clusters Trends Map Phase I: Social Context Generation User’s Cluster Trends Map Location Context Rating User Context Rating User’s Cluster Trends Map S2 > T2? Rated Items (Score S2) Located User’s Cluster Trends Map Items Assignment
50. Determination of appropriateness: definitions • For any contextual model • Each feature has a weight • Each feature value has an appropriateness factor • Feature weight • Appropriateness factor 50 . ∈ 1. . 5 ⊂ ℚ 1  feature definitely not important 5  feature very important … . ∈ 1. . 5 ⊂ ℚ 1  recommendation not at all appropriate 5  recommendation very appropriate … Context Temporal Geographic Activity Device Social Context Location Context User Context
51. Determination of appropriateness: methods • Situation model recommendation score • Context influence factor • Situation decision score S1 51 ∑ . ∗ . . ∈ 0,1 ⊂ ℚ → 1 1 ∗ ∗ + ∗
52. Validation: ViSH scenario • Virtual Science Hub • e-learning social network • Allows collaboration among teachers/scientists to • Share and create enhanced educational content • Improve science curriculum of pupils in the 14-18 age range • Proactive mobile CARS recommends • Learning objects • People with similar interests 52
53. ViSH context features for proactivity 53 Feature Values Social context Social clusters Generated, not generated Location context Geographical User is in or out his city/working area Temporal Morning, Afternoon, Evening, Night User Context Device Desktop, Tablet, Smartphone Activity Away Idle Browsing the platform After filling in the profile While creating new content While editing content While looking for content After finishing the creation of new content While viewing content created by others After viewing content created by others
54. Evaluation: description • Objective • Obtain e-learning user model with values of • Appropriateness of feature values • Feature weights • On-line survey to • Measure the impact of proactive recommendations in educators daily work • 5-point Likert scale questions methodology 54
55. Evaluation: demographics • 104 test users • 64% European teachers • 36% European scientists • 52 men and 52 women • Average age 40 • Usage frequency of recommender systems • 31.73% never • 29.81% hardly • 29.92% regularly • 11.54% frequently 55
56. Results: feature weights 56 0% 20% 40% 60% 80% 100% Geographical Temporal Device Activity Definitely not important Not important Neutral Important Very important 1 2 3 4 Most important contextual feature Less important contextual feature
57. 0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100% Away Idle Browsing After filling in the profile While creating new content While editing content While looking for content After creating new content While viewing others' content After viewing others' content Not at all appropriate Not appropriate Neutral Appropriate Very appropriate Results: appropriateness of feature values 57
58. Validation: publications • International journal 1. Methods to Incorporate Proactivity into Context-Aware Recommender Systems for E-Learning Daniel Gallego, Enrique Barra, Pedro Rodríguez and Gabriel Huecas. JET, 2013 • International conferences 1. Incorporating Proactivity to Context-Aware Recommender Systems for E-Learning Daniel Gallego, Enrique Barra, Pedro Rodríguez and Gabriel Huecas. WCCIT, 2013 2. A Model for Generating Proactive Context-Aware Recommendations in e-Learning Systems Daniel Gallego, Enrique Barra, Sandra Aguirre and Gabriel Huecas. FIE, 2012 58
59. Index • Motivation and open challenges • Research methodology • Contributions • Conclusions 59
60. Contributions 60 • General architecture for mobile CARS with rich social data • Implementation in real ubiquitous and social scenario • Novel model for proactivity in mobile CARS • Mobile user interfaces for proactive recommendations • Outcomes on proactivity impact regarding UX • Methods for incorporating proactivity into mobile CARS • Implementation of proactive mobile CARS in a social network
61. Validation: research projects 61 Perdidos en la gran ciudad
62. Validation: research projects 62
63. Validation: dissemination of results • 2 international indexed journals • 2 international journals • 9 international conferences • 1 national conference 63
64. Open challenges addressed • What kind of architecture is suitable for building mobile CARS in scenarios with rich social data? • Social data analysis available at recommendation time • Social contextual pre-filtering  slow changes • Location and User contextual post-filtering  rapid changes • Social data sources as separate modules • Anonymization  privacy • API REST for managing recommendations • Multi-device and cross-platform 64
65. Open challenges addressed • How can proactivity be incorporated into mobile CARS? • Recommendation model • Contextual situation assessment, then item assessment • Relation between situation appropriateness and item suitability • Feedback to learn from user’s behavior • Methods • Weight of contextual features • Appropriateness of contextual feature values • Situation decision for proactivity as a combination of both 65
66. Open challenges addressed • Which UX factors need to be considered in the implementation of proactive mobile CARS? • Current user activity is the most influential • Other factors are more scenario-dependent • Different proactive notifications offered to the user • E.g. Widget less annoying than Status bar notification 66
67. Future work • Long-term experience with proactive recommendations • Different recommendation profiles for the same user • Novel user interfaces for proactive mobile CARS • Improve current methods to incorporate proactivity • Application to different scenarios 67
68. Thank you Questions? 68