search and recom- mendation are foundational infrastructures to satisfying users’ in- formation needs. As the two sides of the same coin, both revolve around the same core research problem, matching queries with documents or users with items (View Highlight)
we provide a comprehensive survey of the emerging paradigm in information systems and summarize the developments in generative search and recommendation from a unified perspective. Rather than simply categorizing existing works, we abstract a unified framework for the generative paradigm and break down the existing works into different stages within this framework to highlight the strengths and weaknesses (View Highlight)
Nowadays, two information access modes — search and recommendation — serve as foundational infrastructures for information delivery. The objective of search is to retrieve a list of documents (e.g., Web documents, Twitter posts, and answers) given the user’s explicit query [57]. By contrast, recommendation systems aim to recommend the items (e.g., E-commerce products, micro-videos, and news) by implic- itly inferring user interest from the user’s profile and historical interactions [1] (View Highlight)
Search and recommendation are the two sides ofthe same coin [6]. Search is users’ active information retrieval with explicit queries while recommendation is passive information filtering for users. (View Highlight)
. Search can be formulated as a matching between queries and documents, and recommendation can be considered a match- ing between users and items. (View Highlight)
Machine learning-based search and recommendation. (View Highlight)
learn a match- ing function using machine learning techniques (e.g., learning to ranks [65, 89] and Matrix Factorization [61]) to estimate the relevance scores on query-document pairs or user-item pairs. (View Highlight)
Deep learning-based search and recommendation (View Highlight)
m. This paradigm leverages the powerful represen- tation ability of deep learning-based methods to encode inputs (i.e., queries, documents, users, and items) into dense vectors in a latent space [55] and learn the non-linear matching functions. (View Highlight)
he series of “learning to rank” algorithms [65, 89] were also proposed to develop effective rank losses for machine learning based search methods: the pointwise ap- proaches [19, 102, 119] transform ranking into regression or classifi- cation on single documents; the pairwise approaches [10, 107, 135] regard the ranking into classification on the pairs of documents; the listwise approaches [12, 30, 106, 146, 155] aim to model the ranking problem in a straightforward fashion and overcome the drawbacks of the aforementioned two approaches by tackling the ranking problem directly. (View Highlight)
