The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9643

Title:Tpps: Dynamic Natural Language Generation in Technical Domains
Date: 1996
Abstract:Using hypertext for advertising and selling products has become very popular lately, because it supplies user-friendliness and provides an extended information domain that the user can browse through. Although this seems very beneficial it has two major disadvantages: a) The users can get 'lost', they may have to go through a lot of nodes which are not interesting to them to finally reach what they are looking for. As hypertext systems are usually static (once a hypertext node has been designed it cannot be changed), they cannot be tailored to the user's interests and needs. b) A great amount of canned text is required for each node, even if these texts are similar. All the texts used in a static hypertext system have to be predefined, they cannot be generated on line. We have tried to give some answers to these problems through the prototype TPPS (Technical Product Presentation System) we have built. The TPPS simulates a shopping consultant for printers. It generates dynamically descriptions for printers as well as follow up links according to a user model which indicates the users' preferences, needs and interests. In this thesis two main subjects are addressed: a) How to generate Natural Language descriptions of products, how to produce different expressive and interesting texts, b) How to consider the users' preferences and to introduce to them the most interesting links in the most effective way.

[Search These Pages] [DAI Home Page] [Comment]