Presentation is loading. Please wait.

Presentation is loading. Please wait.

An Ontology Based Recommendation System An Ontology Based Recommendation System for Elderly and Disabled Persons Ingo Zinnikus, Anton Bogdanovich, Ralph.

Similar presentations


Presentation on theme: "An Ontology Based Recommendation System An Ontology Based Recommendation System for Elderly and Disabled Persons Ingo Zinnikus, Anton Bogdanovich, Ralph."— Presentation transcript:

1 An Ontology Based Recommendation System An Ontology Based Recommendation System for Elderly and Disabled Persons Ingo Zinnikus, Anton Bogdanovich, Ralph Schäfer

2 An Ontology Based Recommendation System Structure of the talk SAID: Project Overview –Objectives –General System Description Information Service –Agent based personalised access to the web –Basic features of the recommendation system: concept representation, probability calculation

3 An Ontology Based Recommendation System Introduction: SAID Project Data Contract Number: IST-2000-25024 Started: 1st January 2001. Duration: 30 months. Participants: –EPTRON SA. Spain. (Project Coordinator) –VIA DIGITAL. Spain. –Ayuntamiento de Madrid. Spain. –VESTIA Housing. Netherlands. –CASEMA. Netherlands. –City of The Hague. Netherlands. –DFKI. Germany. –University of Edinburgh. UK.

4 An Ontology Based Recommendation System Introduction: Objectives. Objectives: Improve the Quality of life of the Disabled & Elderly by contributing to their independent living. Integrate the Disabled & Elderly in the IST. Improve and optimise processes and methods for Service providers. Advance the state of the art in the field of Digital TV Interactive services: interfaces, communications, MHP, etc. Advance the state of the art in the field of personalised assistants based on Intelligent Agents. As a consequence: Develop an Integrated Platform able to provide remote services for the Disabled & Elderly. Develop innovative Tools for the Service Providers.

5 An Ontology Based Recommendation System Introduction: Innovation. Digital TV: MHP Active system: Autonomous Agents. Simplified Interfaces. Mobile phones. Complete integrated Service. Massive Audience. Viable technologies in an emerging Sector. Proposed Advantages: –Reduce Costs. –Extend the number and scope of users. –Extend the Catalogue of Services. –Offer Added Value Services

6 An Ontology Based Recommendation System Introduction: Environments Which is the Environment of the End-Users? Domestic Users at Home. Mobility Restrictions. Health Problems. Accessibility Problems. Isolation. Loneliness. No computer skills. Which is the Environment of the Service Providers ? Personnel shortage. Continuous increase of Demand. Long Waiting Lists. High costs. 24h attention. Not possible. Lack of technical resources. Lack of technical skills. Only basic demands are covered.

7 An Ontology Based Recommendation System Introduction: Project Description Which is our Proposal ? Increase the quality of life by promoting independent living. Increase processes automation and cost reduction. Extend the catalogue of services. Offer educational possibilities. Increase opportunities for personal communication. Use already existing technical infrastructure. Strong emphasis on personalisation. How can we do it ? Personalised attention through remote services. The TV set as the only interface to the user: DTV. User Interfaces specifically designed for the Disabled & Elderly. Active System: Intelligent Software Agents maintain separate personal user profiles for each user. Tools to automate and ease the job of social assitants: medical histories, planning, user profiles, agenda, WAP. Central Facility able to provide 24h Attention: Videoconferencing, alarms, supervision, advisers, etc.

8 An Ontology Based Recommendation System Introduction: System diagram

9 An Ontology Based Recommendation System User environment OpenTV technology. 2nd generation STB. Oversimplified user interface. Modem link with server subsystem. Information: –Active mode. –Interactive mode. Reminders: –Five reminder types. Alarms: –Confirmation. –Reception & Attention feedback. Other: –Services menus. –Entertainment. Data line Set-Top Box Remote control Digital TV input TV set

10 An Ontology Based Recommendation System Client Platform. Services Menu Options: Information Videoconference Education Shopping Household Back (to TV) Entertainment Characteristics: Clean Design. Few Concepts. Two keys Few Key Strokes. Few Levels. “Intelligent.” Graphics & Sound.

11 An Ontology Based Recommendation System SAID Information Service: Two modes Interactive Mode: Information explicitly requested by the user. Simple user interface. Personalised list of topics. Search trees automatically pruned by Intelligent Agents based on user profile. Active Mode: Information automatically searched by the intelligent agents based on user profile. Use while watching TV. Non Disturbing: Flashing Icon -> Summary -> Complete information

12 An Ontology Based Recommendation System Sources for Information Services World Wide Web (WWW) –Direct access of Web pages –Search Bots Databases –Internal: SAID database –External: Service providers

13 An Ontology Based Recommendation System Motivation: Why not using a browser? Elderly and disabled people often are reluctant to use the internet because of technophobia. The advantage of a browser is a disadvantage for elderly people! The user receives not only information s/he looks for but also advertising, “spam” etc. Hyperlinks are confusing and lead to information overflow.

14 An Ontology Based Recommendation System Solution: Pre-Selected web pages Only a limited set of web pages is selected in order to correspond the needs of elderly and disabled people. Web pages are analysed and categorised in advance. Each web page we use has an associated node in an “ontology tree”. There is no need to spend time at searching a web page for interesting information

15 An Ontology Based Recommendation System Ontology TV Gardenin g Wellnes s Movi e Spor t Plant s Tip s Senior s Health Disease s Al l Simple Ontology The ontology in SAID serves not only as a knowledge storage but also has a functional constituent. Each of the tree nodes serves as a container for additional data that points to a specific web page and helps to extract only desirable information from it using predefined rules

16 An Ontology Based Recommendation System Hierarchical Structure Browsing by iterated selection of concepts All TV Wellness Gardening Movies Sport Soap Operas Comedy Action Science Fiction Ontology tree Action

17 An Ontology Based Recommendation System Dynamical Information Parsing the web page Action URL RULES Snatch Gladiator

18 An Ontology Based Recommendation System Ontology: The “Configurator” tool. Consists of a tree of predefined topics Additional fields for web sites and rules to extract information which can be changed and modified e.g. by a social worker Gives the ability to add/rename/delete the concepts of ontology. Allows to keep in touch with rapidly changing internet environment and quickly react to it’s changes without any modification of the program’s code.

19 An Ontology Based Recommendation System From an Ontology to User Preferences The ontology tree is more a general representation of interesting topics for a type of user than a representation of a specific user’s preferences. In order to generate user preferences, we annotate every concept with a supplementary index which represents the user’s interest in this topic. Question: how to adapt this index to the user’s preferences?

20 An Ontology Based Recommendation System User Preferences are part of user profile name surname id... preferences... Preferences User Profile

21 An Ontology Based Recommendation System Steps towards a recommendation: Adapting to the real preferences according to the decisions of the user –Individual estimates of the user’s interests –The more interactions, the better the estimates Predicting possible user interest for a specific object (web site, text, image, etc.) on the basis of previous decisions

22 An Ontology Based Recommendation System Unobtrusive monitoring of the user’s actions Observation: An object is presented to the user. The user accepts or rejects the object. Meaning: If the user accepts (rejects) the object, her overall evaluation of the object is probably very high (low).

23 An Ontology Based Recommendation System Counting acceptances and occurrences of topics Sport Football Racing Basketball 6/30 2/30 12/30 4/30 6/31 2/31 12/31 4/31 6/31 2/31 13/31 5/31 Items presented Racing

24 An Ontology Based Recommendation System Bayesian Reasoning for tree-like User Preferences For conditional probabilities we have This means for the example Since in the case of our ontology we have We can therefore conclude which gives us in our example

25 An Ontology Based Recommendation System Shortcoming of this solution Calculation only with expected value, but no probability distribution and variance Variance would allow a better estimation of the reliability of a suggestion use of full Bayesian networks to model more differentiated behaviour !?

26 An Ontology Based Recommendation System Bayesian Networks Probabilistic inference mechanism Off-the-shelf tools available for reasoning Technical properties –Nodes correspond to random variables: uncertainty is represented in form of probability distribution. –Edges represent uncertain relations, represented as conditional probability tables. –Standard algorithms to evaluate networks.

27 An Ontology Based Recommendation System Combining Hierarchy with Bayesian nets A subtree represents a Bayesian net only if the concepts within this subtree are independent of each other. (E.g. ‚TV‘ is a media and therefore not independent of e.g. ‚Wellness‘) All Wellness TV Movie Sport Politics Action Preferences Bayesian net [0.0,0.1] [0.1,0.2] [0.2,0.3] [0.3,0.4] [0.4,1.0] [0.0,0.1] [0.1,0.2] [0.2,0.3] [0.3,0.4] [0.4,1.0] (3 times)

28 An Ontology Based Recommendation System Crossing link between branches in the hierarchy Identical or synonymous concepts can be linked together and... All TV Wellness Gardening Emission Sport Soap Operas Travelling Gardening Politics Preferences

29 An Ontology Based Recommendation System Concept is incorporated into the Bayesian net... are integrated into a larger Bayesian net Gardening Emission Travelling Gardening Politics [0.0,0.1] [0.1,0.2] [0.2,0.3] [0.3,0.4] [0.4,1.0] (3 times) [0,1] [1,2] [2,3] [3,4] [4,10]

30 An Ontology Based Recommendation System The Recommendation functionality On each level an additional node ‘Recommendation’ is offered to the user After choosing ‘Recommendation’ concepts in the current subtree with a probability higher than a specific threshold are presented Browsing the ontology tree is facilitated

31 An Ontology Based Recommendation System Conclusion We presented the SAID system which provides support for elderly and disabled persons Information service replaces traditional web Browser by browsing through an ontology tree Tree-like ontology allows Bayesian calculation of conditional probabilities as a basis for a recommendation system


Download ppt "An Ontology Based Recommendation System An Ontology Based Recommendation System for Elderly and Disabled Persons Ingo Zinnikus, Anton Bogdanovich, Ralph."

Similar presentations


Ads by Google