This paper proposes a new method for selecting iconic papers. We use chapter structure recognition technology and combine it with popular deep learning models to identify the chapter structure of the papers. Based on this identification result, we calculate the sum of the number of times each paper is mentioned in all articles, to discover iconic papers with high mention frequency. We focus on the "Method" and "Conclusion" sections to find papers that are frequently mentioned in these two sections and determine iconic papers. With this method, we can more accurately discover and evaluate important research results and provide more valuable references for scholars in related fields.

Taking the authors of extremely highly cited papers as the research object, 18 items of personal information on education, work, awards, and research directions of 287 authors of extremely highly cited papers were obtained through literature information retrieval and network information retrieval. Bibliometric analysis was carried out in terms of cooperation, career length, and education level. The results show that the authors of extremely highly cited papers are more and more cooperating with others; the authors of extremely highly cited papers mostly publish extremely highly cited papers when they are early in their careers; most of them have studied in well-known institutions and received doctoral education; the age of the authors of extremely cited papers will also affect the citation when publishing extremely cited papers.