Search and recommendation are the primary methods for users to access information on the Web and are widely utilized in various scenarios and applications. Recently, there has been rapid development in generative language models, such as ChatGPT, which possess powerful semantic understanding and conversational capabilities, as well as a wealth of prior knowledge. Leveraging these capabilities, generative language models can be applied to various stages of the retrieval and recommendation pipeline, strengthening or replacing traditional solutions. This has led to a paradigm shift in the field of information retrieval, giving rise to generative retrieval and recommendation. These new paradigms aim to use powerful generative language models to retrieve documents or recommend items, and have garnered significant attention.

This Research Topic focuses on addressing the problems of integrating generative language models into search and recommendation systems. While driving search and recommendation into a generative paradigm, the following key issues need to be emphasized. Firstly, how to represent a document or item within a generative language model. Past research has shown that language models struggle to directly generate documents or items themselves, requiring the use of special identifiers as substitutes. Therefore, the selection of identifier types plays a crucial role in the entire generative retrieval and recommendation system. Secondly, how to effectively train generative retrieval and recommendation models. Generative training forms the foundation of generative language models, but when applied to information retrieval, which typically involves discriminative training, pure generative training may seem insufficient. Lastly, how to extend the generative paradigm to cover various scenarios in retrieval and recommendation, such as multimodal documents and products, is also a crucial issue that needs to be explored.

This Research Topic focuses on research that integrates generative models into search, recommendation, and more information delivery systems such as QA and dialogue, including but not limited to the following directions:

1. Development of novel technologies for generative search and recommendation, including but not limited to new models, training paradigms, and inference strategies.

2. Exploration of trustworthiness issues in generative search and recommendation, such as safety, privacy, bias, generalization, and fairness.

3. Development of multimodal generative search and recommendation systems.

4. Exploration of personalized content generation for information delivery, including image, text, video, and audio.

5. Utilization of generative models in various stages of information retrieval systems such as question answering and dialogue systems, question understanding, ranking, content generation, data augmentation, and evaluation.

This Research Topic welcomes various types of submissions, including original research, reviews, surveys, and perspective papers.

Are you interested in contributing a manuscript to this Research Topic?

Generative Search and Recommendation