Many dogs switch owners after initial puppy acquisition, but change of ownership cannot be measured directly as there is no dog registration in England, Scotland or Wales. Knowing the characteristics of dogs that are advertised online can help us understand welfare risks in this u201cpre-ownedu201d population. Source pages from two UK online selling platforms, Gumtree and Pets4Homes, were monitored and new advertisements collected. The resulting rich dataset is huge, and like many other charities and non-profits, Dogs Trust does not have access to robust cloud computing resources for intensive data processing. In the first part of this talk we will detail our Extract, Transform, Load pipeline for this dataset, including a form of locality-sensitive hashing used for deduplication, alongside other programming tips and tricks allowing the pipeline to run in a reasonable timeframe on standard laptop hardware.

In the second part, we will outline how we used regular expressions to clean the deduplicated dataset. Deduplicated advertisements between 21 May 2017 to 16 May 2024 were cleaned to 459,776 advertisements offering dogs u2265 16 weeks old for change of ownership. Key demographic data of sex and breed were identified from free text using regular expressions. Breed could be one of hundreds of dog breeds, designer cross breed terms and their crosses, with many different interpretations and typographical errors entered by advertisers. The dog breeds have between them a huge variety of body shapes and forms, so breed can be used to identify welfare concerns arising from previously identified innate health problems affecting, for example, breathing and mobility. Standard laptop hardware can be used to examine change of ownership through analysing web-scraped advertisements, informing understanding of canine population dynamics and identification of welfare concerns. Re-application of these methods on future free-text data can contribute to longer-term understanding the impacts from welfare-focused interventions.