《电子商务 E-business》阅读文献：AVATAR An advanced multi-agent recommender system of personalized TV contents by semantic reasoning

AVATAR: An Advanced Multi-Agent Recommender System of Personalized TV Contents by Semantic Reasoning WISE 2004- November 23rd, 2004 Yolanda Blanco Fernandez yolanda@det vigo.es Department of Telematic Engineering, University of Vigo(Spain)

Slide 1/14 AVATAR: An Advanced Multi-Agent Recommender System of Personalized TV Contents by Semantic Reasoning WISE 2004 - November 23rd, 2004 Yolanda Blanco Fernández yolanda@det.uvigo.es Department of Telematic Engineering, University of Vigo (Spain)

Motivation of TV Recommender Systems a Migration from analogue to digital tv. Recommender Systems ● Related Work Implications: e Main Design Decisions e The Architecture More channels in the same bandwidth ● The Contributions of AVATAR Software applications mixed with audiovisual contents ●TheL| KO language ● Conclusions ● Further Work

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 2/14 Motivation of TV Recommender Systems ■ Migration from analogue to digital TV. ■ Implications: ◆ More channels in the same bandwidth. ◆ Software applications mixed with audiovisual contents. ■ Disoriented users among large amount of irrelevant information. ◆ User cannot use this new type of TV efficiently. ◆ Necessary tools to find interesting TV programs

Motivation of TV Recommender Systems ● Motivation of Tv Migration from analogue to digital Tv Recommender Systems ● Related Work Implications e Main Design Decisions e The Architecture More channels in the same bandwidth ● The Contributions of AVATAR Software applications mixed with audiovisual contents ●TheL| KO language ● Conclusions a Disoriented users among large amount of irrelevant ● Further Work information User cannot use this new type of Tv efficiently. Necessary tools to find interesting TV programs

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 2/14 Motivation of TV Recommender Systems ■ Migration from analogue to digital TV. ■ Implications: ◆ More channels in the same bandwidth. ◆ Software applications mixed with audiovisual contents. ■ Disoriented users among large amount of irrelevant information. ◆ User cannot use this new type of TV efficiently. ◆ Necessary tools to find interesting TV programs

Related Work ● Motivation of Tv a Different approaches in the field of Tv personalization tools Recommender Systems Bayesian techniques e Main Design Decisions Decision trees e The Architecture Content-based methods ● The Contributions of AVATAR Collaborative filtering ●TheL| KO language ● Conclusions ● Further Work

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 3/14 Related Work ■ Different approaches in the field of TV personalization tools: ◆ Bayesian techniques ◆ Decision trees ◆ Content-based methods ◆ Collaborative filtering ◆ ... ■ A common base: limitation in reasoning capabilities. ◆ Mechanisms to represent the knowledge of TV domain are not used in previous proposals. ◆ Reasoning process allows to obtain enhaced recommendations

Related Work ● Motivation of Tv Different approaches in the field of TV personalization tools Recommender Systems ◆ Bayesian techniques e Main Design Decisions Decision trees e The Architecture o Content-based methods ● The Contributions of AVATAR Collaborative filtering ●TheL| KO language ● Conclusions ● Further Work A common base: limitation in reasoning capabilities Mechanisms to represent the knowledge of tv domain are not used in previous proposals Reasoning process allows to obtain enhaced recommendations

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 3/14 Related Work ■ Different approaches in the field of TV personalization tools: ◆ Bayesian techniques ◆ Decision trees ◆ Content-based methods ◆ Collaborative filtering ◆ ... ■ A common base: limitation in reasoning capabilities. ◆ Mechanisms to represent the knowledge of TV domain are not used in previous proposals. ◆ Reasoning process allows to obtain enhaced recommendations

AVATAR: Main Design Decisions ● Motivation of Tv A TV recommender system based on the experience gained Recommender Systems in the field of semantic Web ● Related Work e Main Design Decisions e The Architecture ● The Contributions of AVATAR ●TheL| KO language ● Conclusions ● Further Work Slide 4/14

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 4/14 AVATAR: Main Design Decisions (I) ■ A TV recommender system based on the experience gained in the field of Semantic Web. ■ Semantic reasoning about user preferences and TV contents. Key elements are: ◆ TV-Anytime specification for: ■ Generic descriptions of TV programs: title, genre, set of keywords, etc. ■ User preferences ■ User viewing history ◆ Knowledge about TV domain provided by an OWL ontology

AVATAR: Main Design Decisions ● Motivation of Tv A TV recommender system based on the experience gained Recommender Systems in the field of semantic Web ● Related Work e Main Design Decisions e The Architecture Semantic reasoning about user preferences and TV ● The Contributions of contents. Key elements are AVATAR ●TheL| KO language ● Conclusions ● Further Work Slide 4/14

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 4/14 AVATAR: Main Design Decisions (I) ■ A TV recommender system based on the experience gained in the field of Semantic Web. ■ Semantic reasoning about user preferences and TV contents. Key elements are: ◆ TV-Anytime specification for: ■ Generic descriptions of TV programs: title, genre, set of keywords, etc. ■ User preferences ■ User viewing history ◆ Knowledge about TV domain provided by an OWL ontology

AVATAR: Main Design Decisions ● Motivation of Tv A TV recommender system based on the experience gained Recommender Systems in the field of semantic Web ● Related Work e Main Design Decisions e The Architecture Semantic reasoning about user preferences and T ● The Contributions of contents. Key elements are AVATAR ●TheL| KO language TV-Anytime specification for ● Conclusions ● Further Work a Generic descriptions of TV programs: title, genre, set of keywords, etc User preferences User viewing history Knowledge about Tv domain provided by an OWL ontology Slide 4/14

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 4/14 AVATAR: Main Design Decisions (I) ■ A TV recommender system based on the experience gained in the field of Semantic Web. ■ Semantic reasoning about user preferences and TV contents. Key elements are: ◆ TV-Anytime specification for: ■ Generic descriptions of TV programs: title, genre, set of keywords, etc. ■ User preferences ■ User viewing history ◆ Knowledge about TV domain provided by an OWL ontology

AVATAR: Main Design Decisions(I) ● Motivation of Tv a The system must be updated when user preferences Recommender Systems ● Related Work lange e Main Design Decisions Goal: personalized and higher quality recommendations e The Architecture ● The Contributions of AVATAR ●TheL| KO language ● Conclusions ● Further Work

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 5/14 AVATAR: Main Design Decisions (II) ■ The system must be updated when user preferences change. ◆ Goal: personalized and higher quality recommendations. ■ AVATAR must be flexible enough to favor updating process. ◆ MHP applications tuned in user’s receiver (Set-Top Box or STB). ■ MHP applications run in the context of a service or event. ◆ Problem: All user actions must be recorded all the time: local agent to watch the viewer behaviour. ◆ Local agent stores feedback information. ◆ Normalized access by TV-Anytime MHP API

AVATAR: Main Design Decisions(I) ● Motivation of Tv The system must be updated when user preferences Recommender Systems change ● Related Work e Main Design Decisions Goal: personalized and higher quality recommendations e The Architecture ● The Contributions of AVATAR AVATAR must be flexible enough to favor updating process ●TheL| KO language ● Conclusions MHP applications tuned in user's receiver(Set-Top Box or ● Further Work STB)

● Motivation of TV Recommender Systems ● Related Work ● Main Design Decisions ● The Architecture ● The Contributions of AVATAR ● The LIKO language ● Conclusions ● Further Work Slide 5/14 AVATAR: Main Design Decisions (II) ■ The system must be updated when user preferences change. ◆ Goal: personalized and higher quality recommendations. ■ AVATAR must be flexible enough to favor updating process. ◆ MHP applications tuned in user’s receiver (Set-Top Box or STB). ■ MHP applications run in the context of a service or event. ◆ Problem: All user actions must be recorded all the time: local agent to watch the viewer behaviour. ◆ Local agent stores feedback information. ◆ Normalized access by TV-Anytime MHP API