数据到文本生成
数据到文本生成(Data-to-Text Generation)是自然语言处理领域的一个经典问题,旨在将结构化数据转换为流畅且准确的自然语言文本。该任务不仅涉及从输入数据中选择合适的内容进行描述,还需通过表面实现技术生成自然、连贯的表达,以满足不同应用场景的需求,如自动报告生成、天气预报和新闻摘要等。
WebNLG
Control Prefixes (A1, T5-large)
E2E NLG Challenge
S_1^R
WebNLG Full
Cleaned E2E NLG Challenge
DataTuner_FC
RotoWire
HierarchicalEncoder + NR + IR
RotoWire (Relation Generation)
Macro
ToTTo
T5-3B
XAlign
MULTIWOZ 2.1
T5-Base
RotoWire (Content Ordering)
Hierarchical Transformer Encoder + conditional copy
Rotowire (Content Selection)
Hierarchical Transformer Encoder + conditional copy
MLB Dataset (Relation Generation)
Macro
MLB Dataset
Macro
MLB Dataset (Content Ordering)
Macro
Czech Restaurant NLG
MLB Dataset (Content Selection)
DART
T5-B Baseline
E2E
self-mem + new data (random)
SR11Deep
Transition based Deep Input Linearization
ViGGO
DataTuner_FC
WebNLG en
WebNLG ru
AMR3.0
StructAdapt
GenWiki
WikiOFGraph
T5-large
Wikipedia Person and Animal Dataset