Neural data-to-text generation with dynamic content planning

Neural data-to-text generation models have achieved significant advancement in recent years. However, these two shortcomings: the generated texts tend to miss some vital information, and they often generate descriptions that are not consistent with structured input data. To alleviate problems, we propose a model Dynamic content Planning, named NDP 2for abbreviation. The can utilize previously text dynamically select appropriate entry from given We further design reconstruction mechanism novel objective function reconstruct whole of used data sequentially hidden states decoder, which aids accuracy text. Empirical results show achieves superior performance over state-of-the-art on ROTOWIRE NBAZHN datasets, terms relation (RG), selection (CS), ordering (CO) BLEU metrics. human evaluation result shows by proposed better than corresponding ones NCP most time. And using mechanism, fidelity be improved significantly.