Abstract: At present, scholars generally adopt external indicators such as citation frequency and hindex to evaluate academic articles, but it is often difficult to reflect the quality of the content of the papers, and may even lead to the tendency of emphasizing "quantity" rather than "quality", thus inhibiting academic innovation and discipline development. In this paper, we propose a method to evaluate the academic articles from the perspective of content differentiation. The research object of this paper is the articles from the discipline of library, information and archival Science. By combing BERT with semantic representation and the ADC (article discriminative capacity), this paper measures the difference degree of articles at the level of research content, and analyzes the article discriminative capacity by the numerical distribution characteristics. Differences in the overall level of ADC are explored separately from the perspective of journals, scholars, and topics. The experiments show that the ADC proposed in this study is reasonable for measuring the content differences of articles. The articles with higher discriminative capacity have more unique research content, otherwise there will be more homogenized content. Archival journals are clearly distinguished from library and information journals due to the uniqueness of subject terminology and research objects. Authors studying information science make up a large percentage of productive authors. Articles on hot topics and frontier applications show higher discriminative capacity due to more indepth and popular topics, while articles introducing traditional theories and literature review show lower discriminative capacity due to more general and mature contents.

Key words: Article discriminative Capacity, Academic articles, Content differentiation, LDA, Academic evaluation