Research on the Algorithm for Researchers Profile Construction Based on Word Fusion Embedding Bert Model

This study aims to propose a Bert model based on word fusion embedding. By introducing word fusion embedding vectors to extract key information of scientific researchers and construct research’s profile. By integrating, analyzing, and mining the intellectual property and project data of scientific researchers, a keyword extraction model is proposed. This model mines representative words by learning the contextual information of the text. This study proposes a character and word fusion model, the model avoids the effect of keyword extraction caused by word segmentation errors. By introducing the attention mechanism and the fusion of character and word vectors, a unique profile of scientific researchers is constructed, which helps to comprehensively and intuitively understand scientific researchers. It is important in improving the efficiency of scientific research management and optimizing the evaluation mechanism of researchers. The method proposed in this study has achieved remarkable results in improving the accuracy of keyword extraction, outperforms other algorithms. The top5 accuracy is $73.7 \%$.