This repository extends the original paper with a Chinese dataset, primarily using content from the "百家讲坛" as the data source.
In terms of data processing, the methods for acquiring and processing gestures from the original repository are continued. Since Chinese cannot be tokenized like English, this project adopts the gesture that appears most frequently in a sentence as the gesture for that sentence. Based on the original code, minor modifications have been made to processing and dlbilibili to adapt to the processing needs of Chinese data. The processed Chinese data is stored in the data folder.
Furthermore, a summary of Chinese gestures has been conducted, and simple RNN and other models have been attempted for prediction, but the results were not satisfactory. In addition, comparison charts and count analyses with the American gestures from the original repository are provided for reference.
Welcome to communicate and discuss this project!