Contemporary canine health and welfare research is increasingly breed-centric due to substantial evidence highlighting how intensive phenotypic selection and inbreeding to achieve breed standards and consumer demand impact canine health and welfare. Popular breeds now face high risks of inherited and conformational disorders affecting cardiovascular, respiratory, auricular, dermatological, ophthalmic, and musculoskeletal health, resulting in acute and chronic suffering. Furthermore, correlations exist between specific breeds and public safety and health concerns, such as dog bites, baiting and fighting, which have thus far been ineffectively addressed by legislation.

A critical methodological challenge exists in analysing relevant, large-scale datasets as they do not contain standardised breed language, spelling, or structure. This compromises accurate identification, data linkage, and introduces misclassification bias. To address this, this project aims to identify breed and phenotypic descriptors from classified adverts using a BERT language model, fine-tuned for a Named Entity Recognition (NER) task. Initially, this will support the interrogation of breed demand using large-scale classified advertisement data, focusing on unstructured free-text listings where breed names are frequently misspelled, abbreviated, or creatively described (e.g., designer crossbreeds). Both pre-trained and domain-adapted models will be tested to assess their performance in identifying tokens containing breed information. The best-performing model will then be used to test various methods for cleaning extracted breeds, crossbreeds and phenotypes using regular expressions and generative language models. These approaches will be evaluated based on their performance in breed correction, explainability, and energy efficiency.

The resulting normalised dataset and accompanying pipeline will enable direct comparisons between advertised breed popularity and external health and welfare datasets. This approach provides a scalable method for linking canine health and welfare-related datasets, supporting the development of more collaborative and evidence-based interventions and policy reforms to improve canine health and welfare.