An Intelligent TV assistant
Participants
- Tobias Röhlig
- Simon Schmid
- Silke Terhorst
- Sara Winter
Supervisors
- Christoph Broschinski
- Christina Unger
- Ramin Yaghoubzadeh
Motivation
- People are often faced with a very large television program with theoretically more than 100 stations and thousands of broadcasts
- An intelligent room assistant could help people to deal with this overload of information by filtering the television program by the user's preferences
- Such an intelligent agent could manage the whole program, present it to the user and make suggestions for the user to show him new and maybe interesting broadcasts
Application Scenario
The user comes home and would like to watch TV. Unfortunately he doesn't know the TV program by heart or even have a TV guide so he asks the artificial room assistant Billie.For example: "At what time will the news be broadcasted this evening?" In this case Billie looks up all news and tells the user where and when the news will be broadcasted.
Another case: "What does station XY broadcasts this evening?" Now Billie tells the user every title he has found for this station.
The user can also ask: "Is there something interesting on TV this evening?" In this case Billie searches for all genres the user likes.
If the user is interested in a program Billie can inform him about the broadcasting time, the station and the genre of the chosen program. He can also give a detailed description.
Objectives
The project goals are- Download and parsing of the TV program
- Setup a RDF-Database from the TV program and establish a database connection
- Setup a dialog model
- Integration of the speech recognition
- Setup a user profile that can be used to influence the selection of the TV-program that is presented to the user
Description
The Intelligent TV assistant:- understands spoken language (German in this case)
- uses spoken language
- can search for genre
- can search for station
- can search for title
- can search for broadcasting time
- knows preferred user-genre
- knows preferred user-time
Results
The TV-scenario has been implemented into the existing Billie-system (Socialble Agens Group: http://www.techfak.uni-bielefeld.de/ags/soa/):- The TV-Program can be downloaded and parsed from an external source
- A RDF-database has been set up and can be filled with the TV-Program
- Billie has been enabled to parse questions in natural language about this program and to present the desired answers according to the entries in the database
- A complex dialog model has been designed and implemented in order to allow a dialogue in natural language
- A user model has been set up so that user-preferences influence the way results are presented by the virtual agent
Discussion and Conclusion
- The planned results have been achieved and a prototype of an intelligent TV-assistant has been implemented.
- It is very relevant to design a complex dialog model in order to be able to parse and understand natural language statements in the current context because similar statements may need a different interpretation according to the state of the dialogue.
- The main benefit of an intelligent TV-assistant is a well designed user model so that suggestions for programs fit to the user's interests. Such an implementation means an enhancement to general TV-program information systems.
- Speech recognition and production in a multi-language system proved to be difficult to handle as our system is designed for German users but a lot of movies have an English title.
Outlook
- Improvement of the synchronization between the agent's gestures and it's audio response.
- Enhance the user profile to provide a more detailed model of the user and his preferences.
- The agent should understand more complex questions from the user.
- Connection to an existing calendar view to make it possible for the user to plan his TV-program for the current week.