Interdisciplinary research has become a crucial part of modern scientific progress, but predicting the future impact of studies across multiple research fields remains a significant challenge. Several studies have shown that integrating linguistic information and citation networks can effectively predict the citation counts of research papers. This paper proposes a model that fuses linguistic information and citation networks, which applies even to large-scale data from the academic literature, to classify high-impact documents. Our model applies to large-scale academic datasets, enabling the classification of citation counts in broader academic fields such as interdisciplinary areas. In our experiments, we used a large dataset of scholarly papers in multiple research fields to evaluate the classification accuracy of the citation counts using our model. The experimental results show that our classification accuracy improves by 0.022 points compared to models using only citation networks. This result indicates that our model is valid across multiple research fields. Furthermore, we clustered the embeddings we obtained and demonstrated that highly cited papers are biased at the boundaries of the clusters. This bias indicates that highly cited papers are more interdisciplinary than less cited ones.