SELLMA: Semantic Location through On-Device LLMs and WiFi Sensing

Understanding a user's semantic location is of critical importance in numerous areas of mobile computing, such as mobile healthcare, mobile advertising, and mobile personal assistance. Nevertheless, inferring semantic location remains challenging and often relies on translating raw geographical coordinates via third-party online services. In this paper we introduce SELLMA, an approach for semantic location inference that harnesses Wi-Fi SSID sensing and on-device querying of a specially crafted LLM. We implement SELLMA in Android and show that it can uncover a number of environmental and geographical descriptors of a users location in a privacy-preserving manner, without the need for GPS querying, and without reliance on Web-based services.