A Prediction Method of Scientific Achievement based on Attention Characteristics

Abstract

In order to arrange funds and increase profits, scientific management department in enterprise needs to master the situation and development of scientific achievements. Scientific achievements have the characteristic of lag. Prediction in advance is beneficial for managers to improve working efficiency and effectiveness. In the paper we propose a prediction method of scientific achievements based on attention characteristics. The prediction framework consists of four modules. The data of working status are collected by advanced monitoring devices, while the current works ignore this kind of data. The attention features are extracted and processed for the convenience of model training. Our prediction model is time-sensitive, which could find the relationship between the achievements and their working period. The experimental results demonstrate that the proposed method works as well for any kind of scientific achievements. Its estimate accuracy and topic similarity is 48% and 70.6% better than benchmark models separately.