Rationalizing and better-managing energy consumption have become primary objectives in the global effort to prevent climate change. In this context, enhancing our understanding of electricity consumption behavior is crucial. Thus, electricity suppliers have installed millions of smart meters worldwide over the past decade, capturing time-stamped electricity consumption data of the total main power consumed in individual households. Nevertheless, suppliers face significant challenges in extracting detailed information from these aggregated signals, such as identifying which appliances the customer owns and their typical usage. This task is complicated by the reliance on low-frequency smart meter readings, which combine signals from various appliances operating simultaneously. Moreover, the scarcity of annotations and the large amount of long, variable-length consumption series collected further complicate data analysis and interpretation. In this Ph.D. work, we propose a set of new solutions to tackle the appliance detection problem and extract detailed information from smart meter data that overcome the challenges listed above. First, we propose to tackle this task as a binary time series classification (TSC) problem and subsequently describe the Appliance Detection Framework (ADF), designed to enhance classifiers' performance using long and variable consumption series. Moreover, we introduce TransApp, a deep-learning architecture that is first pretrained in a self-supervised way to enhance its performance on appliance detection tasks. Finally, we propose an interactive system based on a combination of TSC and explainable classification that enables the localization of appliance patterns without the use of strong labels.