DESIGN AND IMPLEMENTATION OF AN INTELLIGENT MULTIMODAL SEMANTIC RETRIEVAL SYSTEM FOR NEWS MEDIA RESOURCES

Abstract

Aiming at the problems of low retrieval efficiency, insufficient semantic understanding and difficulties in multimodal resource management caused by the rapid growth of news videos, audio materials and text files in the railway integrated media scenario, this paper designs and implements a text news video semantic retrieval system based on DeepSeek multimodal technology. Adopting an intelligent technical route combining multimodal resource text conversion, DeepSeek semantic matching and subtitle timestamp mapping, the system extracts text from multimodal resources, leverages DeepSeek to compute semantic relevance between user queries and resource contents, and finally establishes the mapping relationship between subtitle contents and video timestamps for news video resources, so as to realize fast positioning of video clips from text queries. Different from traditional file level retrieval, the proposed method can refine news video retrieval results to specific time segments. Function modules such as difference comparison and keyword positioning are also integrated into the system. Experimental results demonstrate that the system can improve the retrieval efficiency of news videos and the semantic analysis capability for multimodal resources, and has good practical engineering application value.