Ilan Platformlari i in Anlamsal Arama Yakla simi

Traditional keyword-based search methods fail to fully capture the search intent behind users' complex queries on classified platforms, often leading to irrelevant results and negatively impacting the search experience. This study examines the effectiveness of semantic search approaches by targeting these problems encountered in traditional search methods. The effectiveness is evaluated by testing on a labeled dataset of used & brand-new classified ads. In this context, the performance of various state-of-the-art multilingual vector embedding models and the representation methods of the advertisements are compared with quantitative measurements. As a result of these measurements, the best model and advertisement vectorization method was determined with an NDCG score of 0.9597.