Improving bibliographic coupling and co-citation: A study evaluating the effects of various full-text citation features

Classical methods for mapping domain knowledge structures, namely bibliographic coupling (BC) and co-citation (CC) analyses, rely on co-reference or CC counts, which may lack precision and reliability. While full-text mining can enhance BC and CC strength, there is limited comparative analysis on the impact of different full-text citation features. This study explores the optimisation effects of four full-text citation features: citation content, sentiment, position and mention frequency. Enhanced strength algorithms for BC and CC relationships were designed based on these features, and a comparative experiment was conducted in the field of oncology. Deep learning techniques were employed to extract various citation features, which were then used in the proposed models and control groups. These full-text citation features were assessed for their effectiveness and characteristics in discovering domain knowledge structure. The study revealed that including full-text citation features improved the traditional methods, aligning more closely with expert knowledge. These features offered distinct insights but also introduced potential drawbacks. The research results hold insights for gaining a deeper understanding regarding the optimisation effects of full-text citation features on traditional bibliometric methods.