Abstract: 【Purpose/significance】With the continuous deepening of interdisciplinary and fusion of disciplines,scientific research work increasingly requires the cooperation of multiple scholars.Identifying potential cooperation relationships and recommending suitable partners for scholars can effectively improve the efficiency of scientific research【. Method/process】This study is based on the dynamic network representation learning model to study the cooperation relationship prediction of scholars.First,a dynamic network representa⁃tion learning model DynNE_Atten is proposed.Secondly,use the data in the field of library and information to construct a dynamic sci⁃entific research cooperation network and a dynamic keyword co-occurrence network.And then obtain the author vector representation
and keyword vector representation through the DynNE_Atten model.At the same time,we obtain affiliation units of authors.Finally,it in⁃tegrates the characteristics of the author's cooperation,theme and unit characteristics to predict the possible future cooperation【. Result/conclusion】The experimental results show that the dynamic network representation learning model proposed in this paper only needs fewer input samples in the time series link prediction task,and can achieve higher accuracy; compared with the scholars without fusion features.The fusion model shows obvious advantages in the prediction of cooperative relations.[Innovation/limitations] This study pro⁃poses a new dynamic network representation learning model,and then combines topic features and features of author's affiliation to pre⁃dictions scientific research cooperation,which has achieved good results.However,the method only considers the heterogeneity of data in the feature fusion method and doesn't consider the heterogeneity of the network.