Temporal Query Modeling in Evolving News Archives

Abstract

With the advancement of technology, the digital archive of news history is becoming increasingly difficult to navigate due to the prevalence of overlapping and contextually accurate information on the Internet. By taking into account both the user's request and the item's content circulation over time, narrative building aims to enhance information retrieval. By using the constructive time aspects of both inquiries and articles to maximize the search output, this research introduces a new method for developing query modeling in news archives via temporal unsupervised learning. Using the approach to analyse news data that is already timestamped, the method can dynamically identify temporal intent and sensitive time entities and relate documents to the temporal extent of the query. To achieve this, we propose a model that utilizes temporal profiles in conjunction with expansion query technology and temporal re-ranking, enabling the retrieval of relevant documents from specific periods. When compared to baseline models, benchmark news datasets demonstrated superior retrieval performance, with the most notable improvements confirmed for queries with a stronger temporal dependency. Ideal for accessing archives, researching history, and retrieving event-centric material, this technique also consistently captures the drift of a subject with the appearance of details throughout time. This study provides a practical way to improve the temporal search experience in digital journalism archives and highlights the significance of temporal modeling for dynamically capturing global news developments.