Chart to text generation

• Main Author: Zhou, Hongyu
• Other Authors: Joty Shafiq Rayhan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/147962
• Institution: Nanyang Technological University

For exploring data and sharing perspectives, the visualizations of information like the bar charts and line charts are always common. For certain individuals, such as those people who are blind or have poor visualization literacy or visually impaired, decoding and making sense of those visualizations would be much difficult. We present a new dataset and a neural model for automatically generating natural language captions for charts in this paper. The captions that are produced provide an overview of the chart and express the key insights found within it. The data-to-text generation task utilize the state-of-the-art model, which uses a transformer based encoder-decoder architecture, was used to construct our neural model. We discovered that our phase1 method outperforms the base model by a large margin on a content selection metric (55.42 percent vs. 8.49 percent) and produces more informative, succinct, and coherent summaries