Grant: Engage Grants Program

For a tourism agency, writing an hotel description is a complex task, as it must appeal to readers (potentialcustomers) as well provide correct information about all the amenities of the hotel (restaurants, types of rooms,facilities - pools, conference rooms, etc). Frequent updates become problematic since rewriting a descriptioneach time would be extremely costly.Facing such problem, the company Expedia has developed a Content Generation System (CGS) which allowsthe writing of templates. A template is a partial sentence using variables in place of specific hotel amenitiesand descriptions. At generation time, the variable names are replaced by actual variable values to generatesentences. A template can contain ""restaurant_name"" as a variable, and at generation time, it can be replaced,for example, by ""Margo's pub"" or by ""John's cafe"". The restaurant name can change, but the template remainsvalid. The idea behind CGS is very interesting, but such a system is not always easy to use and to maintain.First, it is not easy for human writers to write templates, and to see where they can include variables. This leadsto the first question put forward in this research project. Using text mining approaches, can we automaticallygenerate sentence templates from a corpus of hotel descriptions? Second, the number of templates should beminimized (for easier search and maintenance), but still provide a large generation potential (with variations forsentences). This leads to the second question of our research project. How can a better organization of theamenities, through an ontology instead of a list (as the current system has) help in the automatic extraction of asmaller set of templates which would allow for variations through the related concepts and terms in theontology? Addressing these two questions, and prototyping an automatic template construction system for asmall subset of the amenities will provide insights for the further development of CGS.