In general, hadith consists of isnad and matan (content). Matan can be separated into several components for example a story, main content, and some additional information. Other texts besides main content, such as isnad and story can interfere the retrieval process of relevant documents because most users typically use simple queries. Thus, in this paper, we proposed a Named Entity Recognition (NER) component weighting model in improving the Indonesian hadith retrieval system. We did 3 test scenarios, the first scenario (S1) did not separate the hadith into several components, the second scenario (S2) separated the hadith into 2 components, isnad and matan, and the third scenario separated the hadith into 4 components, isnad, background story, content, and additional information. From the experimental results, it is found that the TF-IDF with rocchio algorithm in query expansion outperforms DocVec. Also, separation and weighting of the hadith components affect the retrieval performance because isnad can be considered as noise in a query. Separation of 2 separate components had the best overall results in general although 4 separate components showed better results in some cases with precision up to 100% and 70% recall.