Table-to-Text Generation of Energy Advisory Reports : A Comparison of Template and Prompt Based Methods

Sammanfattning

Improving energy efficiency in manufacturing is an important step toward reducing overall energy use and environmental impact. Tailored advisory reports based on a company’s own energy consumption data can be a helpful approach. This study explores whether parts of such reports can be automatically generated from structured data, either through traditional template-based methods or by prompting a large language model. The reports used in the study were produced by the company DAZOQ, which provides energy monitoring and analysis for industrial clients. The two generation methods were tested and evaluated both through automated comparison with the original reports and through human assessment. The results showed that only a small portion of the original content could be generated based on the available data, highlighting the importance of how input is structured. The template-based method achieved higher similarity to the human-written reports, while the prompt-based method produced more varied output. All reports were rated highly by a human evaluator, regardless of method. The study concludes that automatic generation in this domain is possible and may be useful in practice, but further development of both the input data format and the generation approach could enhance the results. Given the scope of the study, no definitive conclusion can be drawn about which method is generally preferable.