A Hybrid Deep Architecture for Improving Academic Evaluation Capacity in Smart Campus System

A Hybrid Deep Architecture for Improving Academic Evaluation Capacity in Smart Campus System

With the fast growing of education informatics, academic evaluation is important for university study life. It bring a series of research questions, including the research about Elearning behavior which is one of hot issues in it. Meanwhile, the modern education attaches importance to Individualized cultivation. More subjective opinions are collected from the students. Thus, automatic inferring academic behavior is gradually becoming the key of perfecting academic evaluation system (AES). In the paper, we collect the academic evaluation information from the online platform of Hohai University. According to the overall opinions, we give the labels to the detail textual comments. A two-stage network is designed and implemented for subjective text analysis and screening the final answers. The former stage is based on bidirectional long and short term memory (LSTM) networks to output a soft label for each sub-sentence. According to the total number of the questions, we locate their answers, and product the inputs of deep forests by cascading their predicted soft labels. Based on cascade forest structure, multi-level forests are reweighted by the forests’ contributions. A level-wise growing strategy is used to control the cascade level of the entire structure. Experiment results demonstrate our work are competent for a kernel approach of AES. Meanwhile, we capture many problems in the teaching and learning processes, which are easy to be ignored in the conventional questions and answers (QA) step, and offer some constructive suggestions for the future of smart campus systems.