Despite the thrilling achievements of deep neural networks on a wide range of tasks in computer vision and natural language processing, revolutionary improvements from deep neural IR models are yet to achieve, highlighting the unique challenges in applying deep learning technologies in information retrieval. In addition, comprehensive comparisons between the neural IR models and different state-of-the-art IR models from prior-deep-learning era, like the unsupervised models and the learning to rank methods, are also desired to draw convincing conclusions about the performance of deep neural IR models.
Henceforth, this special issue aims at providing a forum for a better understanding of the challenges in developing more advanced neural IR models, a platform to present novel deep architectures for retrieval tasks, as well as a venue to reproduce the results for current the state-of-the-art neural IR models, and to conduct comprehensive comparisons between them and the traditional IR models, in favor of a better understanding of the neural IR models' performance.
Topics for this issue include the application of deep learning in IR tasks, including but not limited to:
Document retrieval, passage retrieval and question answering
Web search, sponsored search, entity search and search on social network
Encode the established insights and methods from IR community into a neural IR models, e.g., the use of relevance feedback signals and combine the learning to rank features into neural IR architectures
Multi-modal retrieval, e.g, image retrieval on text corpus
Weak supervision to train neural IR models
Domain specific IR, e.g. medical search, patent search
Information routing and filtering
User and task modelling, personalized search, recommendations, search result diversification
Neural query performance prediction
As well as benchmarking different neural IR models and compare them with traditional retrieval paradigms.
Reproduce and compare different state-of-the-art neural IR models
Compare the neural IR models with fine-tuned unsupervised models and learning rank models
Other topics in IR evaluation, especially for the neural IR models