.png){kind=icon}
.png){kind=icon}
Return
Automatic statistical chart analysis based on deep learning method
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-023-17420-4.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In order to make full use of the decision support knowledge contained in the structured chart data that matches the needs of the enterprise, we propose a deep learning approach for automatic statistical chart analysis. This method uses LSTM as the basic model for chart analysis, and mainly makes the following improvements: (1) A discriminator layer is added after the embedding layer of the model, so that the model can perform more targeted semantic understanding and text prediction according to the knowledge of chart; (2) In the character sampling process of the model, a random cluster sampling strategy is proposed to improve the quality of chart description; (3) The model is optimized by using the knowledge distillation method, so that it can generate more valuable description text for manufacturing scenarios. Experiments show that this method improves the text quality by 12.4% compared with traditional LSTM. To further evaluate the analysis description quality of ASCAT, we use the same dataset to train RNN, LSTM, AISG, EICT, and AGNLD models. Experiment results show that our ASCAT model obtained better description quality evaluation scores in description quality under the METEOR criteria considering the precision and recall rate, or under the CIDEr criteria with the TF-IDF weight introduced.
Keywords:
Statistical chart analysis
Natural language generation
Quality control platform
Manufacturing enterprises
Knowledge distillation
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W