Semantic Hyperlocal Search for Parlance Mobile Spoken Dialogue System
Current spoken dialogue systems (SDS) for mobile search are mostly domainspecific and make use of static knowledge. Consequently they do not take into account the interests, location and contextual situation of the concrete user. We propose PARLANCE, which is a more dynamic and personalized SDS that incorporates 1) a dynamic knowledge base consisting of modular ontologies that are enriched incrementally with information extracted from the Web; 2) and an evolving user profile. This allows the system to provide answers that are more tailored to the concrete user and to exploit the Web as a source of information, which can improve the quality of experience for the user. The PARLANCE SDS aims to guide the user in his search for information by providing answers that are: (1) Hyperlocal: The current geographical location of the user is taken into account to provide points of interest (POIs) in the neighborhood; (2) Dynamic: New concepts and entities are learned at runtime and included in the appropriate modular ontologies; (3) Personalized: Potential relevant answers adapted to user’s queries are selected and ranked according to user preferences. Complementary, a form of social search is performed by looking at interests of similar user in the neighborhood (i.e. collaborative filtering). The central component in the PARLANCE architecture is the Interaction Manager (IM) which probabilistically decides on the most appropriate next answer to be provided to the user. The IM exploits information from the Semantic Web by interacting with the Knowledge Base (KB), the Web Content Analyzer (WCA) and the Local Search (LS) components, which will be detailed in the next section.