构建了基于BERT的双向连接模式BERT-based Bi-directional Association Model(BBAM)以实现在意图识别和槽位填充之间建立双向关系的目标,来实现意图识别与槽位填充的双向关联,融合两个任务的上下文信息,对意图识别与槽位填充两个任务之间的联系进行深度挖掘,从而优化问句理解的整体性能.为了验证模型在旅游领域中的实用性和有效性,通过远程监督和人工校验构建了旅游领域问句数据集TFQD(Tourism Field Question Dataset),BBAM模型在此数据集上的槽填充任务F1值得分为95.21％,意图分类准确率(A)为96.71％,整体识别准确率(Asentence)高达89.62％,显著优于多种基准模型.所提出的模型在ATIS和Snips两个公开数据集上与主流联合模型进行对比实验后,结果表明其具备一定的泛化能力.
Joint model for intention recognition and slot filling in tourism
Intent recognition and slot filling are two core tasks in question-answer comprehension and modeling the two tasks jointly have become a new trend in current research.Based on this,we introduce a BERT-based Bi-directional Association Model (BBAM) to realize the bi-directional association between intention recognition and slot filling,fuse the contextual information of the two tasks,and deeply explore the connection between the two tasks of intention recognition and slot filling,to optimize the overall performance of interrogative sentence understanding.To verify the practicality and effectiveness of the model in the tourism field,TFQD (tourism field question dataset) is constructed in this paper by remote supervision and manual verification,and the F1 score of the BBAM model on this dataset for the in-slot filling task is 95.21％,the accuracy of intention classification (A)is 96.71％,the overall recognition accuracy (Asentence)is as high as 89.62％,which is significantly better than various benchmark models.The results of comparison experiments with the mainstream joint model on two publicly available datasets,ATIS and Snips,also show that the proposed model has a certain generalization ability.
natural language understandingspoken language understandinginterrogative sentence understandingtravel domainintelligent question and answerintention recognitionslot fillingjoint modeling
万方数据
维普
