基于特征增强的多方位农业问句语义匹配

Multi-Level Semantic Matching of Agricultural Questions Based on Feature Enhancement

农业问句文本数据具有专业名词多、特征稀疏、语句规范性差等特征,难以深入挖掘句间交互关系.为改善农业相似问句的匹配性能,提出一种基于特征增强的多方位农业问句语义匹配模型.模型通过共享参数的双向长短期记忆网络提取上下文向量,分别引入自注意力机制、多维注意力机制增强农业问句文本语义推断特征和文本距离特征,通过多特征增强聚焦语义特征信息,将增强特征嵌入到多方位匹配函数中,从向量值、方向和元素等角度进行句间相似度对比,以捕获句子多样性特征.从农业问答社区导出农业问答文本数据,人工标注相似问句构建试验数据集.试验结果表明:基于特征增强的多方位农业问句语义匹配模型可以增强文本特征之间的交互,获取更多的关系特征信息,在构建的农业问句数据集上正确率及F1值达95.3%和97.3%,与其他5种问句语义匹配模型相比,效果提升明显.

To improve the performance of similarity calculation in agricultural Q&A community,according to the characteristics of agricultural question which are many professional nouns,sparse and poor sentence standardization,a semantic matching model of agricultural question sentences based on features enhancement was proposed.The model extracts context vectors through a bidirectional long-short term memory network that shares parameters.The self-attention mechanism and multi-dimensional attention mechanism are used to enhance the semantic inference features and distance features of agricultural question text data,respectively.Through multi-feature enhancement,the semantic feature information is focused,the enhanced features are embedded in the multi-directional matching function,and the similarity is compared from the perspectives of vector value,direction and element to capture the diversity characteristics of sentences.Agricultural Q&A text data is exported from the agricultural Q&A community,and similar questions are manually labelled to construct experimental datasets.The experimental results showed that the agricultural question semantic matching model based on enhanced multi-feature can enhance the interaction between text features,get more relationship feature information.The accuracy and F1 values of the proposed model were 95.3%and 97.3%.Compared with the other five semantic matching models,the experimental results showed obvious advantages.