Deep Learning Scoring Model in the Evaluation of Oral English Teaching

Table 1

RNN optimization features.

Optimization features	Specific optimizations
Dual-directional modelling	The current RNN could only perform calculations under the current information when operating on data information. The dual-directional RNN adopted the ability of bidirectional data infusion and determined the output.
Long-term dependence	When the data sequence was long, the traditional RNN might lead to gradient failure. While the long short-term memory (LSTM) network could solve this issue, with the principle of introducing input gate, forget gate, and output gate.
Optimized computing nodes	Reoptimization was made on the basis of long-term dependence, as remake gate and innovation gate were added. After these two nodes were added, the entire system had a faster computing rate due to reduced parameter scale.