基于训练模型改进的语音问句信息抽取方法

为进一步提高客户语音问句实体信息抽取的准确性,增强智能问答系统知识图谱中信息抽取技术的整体效果,首先对语义标注进行优化,随后在BiLSTM-CRF(bidirectional long short-term memory conditional random filed)基础上加入BERT(bidi-rectional encoder representation from transformers)模型对句子进行实体抽取学习.在具体实验中,以语音问句事件文本为数据来源,对其进行语义标注和实体抽取实验.结果 表明,在语义标注优化的基础上同时加入BERT改进模型,信息抽取结果均高于BiLSTM-CRF方法,且改进模型的调和平均值达到91.53％,即可为增强事件实体信息抽取提供实践意义.

Improved Information Extraction Method of Speech Question Based on Training Model

In order to further improve the accuracy of the entity information extraction of customer voice consultation questions and enhance the overall effect of the information extraction technology in the knowledge graph of the intelligent question answering system, the authors optimized semantic annotation and applied the BiLSTM-CRF (Bidirectional Long Short-Term Memory Conditional Random Filed) with BERT (Bidirectional Encoder Representation from Transformers) model to learn entity extraction of sentences. Entity extraction and semantic annotation were validated in specific experiments with the voice text of customer voice consultation question events as the data source. The results show that, based on semantic annotation optimization, the BERT model works better than the single BiLSTM-CRF method for semantic annotation extraction. The proposed model harmonic mean reaches 91.53% which is of practical significance for enhancing the event entity information extraction.