Large-Scale Data Processing for Information Retrieval Applications

Developing Information Retrieval (IR) applications such as search engines and recommendation systems require training of models that are growing in complexity and size with immense collections of data that contain multiple dimensions (documents/items text, user profiles, and interactions). Much of the research in IR concentrates on improving the performance of ranking models; however, given the high training time and high computational resources required to improve the performance by designing new models, it is crucial to address efficiency aspects of the design and deployment of IR applications at large-scale. In my thesis, I aim to improve the training efficiency of IR applications and speed up the development phase of new models, by applying dataset distillation approaches to reduce the dataset size while preserving the ranking quality and employing efficient High-Performance Computing (HPC) solutions to increase the processing speed.