لخّصلي

The graph model comprises three primary classes as depicted in Figure 2: the household, its location, and its appliances which are interconnected.Additionally, it includes demographic properties such as GDP, average wage, population density, energy prices, and the carbon intensity of electricity production.The KG is encoded in RDF, stored in a Blazegraph database and can be queried via a SPARQL endpoint.The KG can be explored further on it's LodView deployment 1.

The graph model comprises three primary classes as depicted in Figure 2: the household, its location, and its appliances which are interconnected. Each household node is connected to its respective location node and to all appliance nodes which are contained within the household. The household class describes the household with the start and end date of electricity consumption measurements, occupancy, square footage of the living space, and dwelling type (e.g., house or apartment). The appliance class contains properties detailing appliance electricity consumption such as average daily consumption and average consumption during an on/off event. The location class describes the household location with its coordinates along with the city, country, and continent information. The location class integrates with WikiData and DBpedia to enrich the city and country information. Additionally, it includes demographic properties such as GDP, average wage, population density, energy prices, and the carbon intensity of electricity production. The KG can be explored further on it’s LodView deployment 1. The KG is encoded in RDF, stored in a Blazegraph database and can be queried via a SPARQL endpoint. So we assume that the transformation block from Figure 1 transforms the natural language to the corresponding SPARQL query, and that the LLM receives the natural language question, the SPARQL query and the triples that are retrieved from the KG. We evaluate the RAG with two versions of ChatGPT (4o and 4), Gemini 1.5 and Llama-3-8b-chat-hf.