Question answering for data visualizations

摘要

Answering queries for data within data visualisations, such as a bar chart, pie chart, graph and so on, by first identifying a data visualisation (DV) 202, then generating a DV feature map that characterises the DV and includes maintaining a correspondence of spatial relationships of mapped features of the DV within the DV feature map with corresponding features of the DV 204. It then identifies a query for an answer included within the DV 206, and encodes the query as a query feature vector 208. It then generates a prediction of at least one answer location within the DV, based on the DV feature map and the query feature vector 210, and finally determines the answer from the at least one answer location 212. In an alternative embodiment a synthetic dataset is generated to train a convolutional neural network, a long-short term memory encoder and an attention model.