| 摘 要: 传统的中文命名实体识别技术在汽车领域的应用效果欠佳,特别是在处理实体之间存在大量的嵌套重叠、不连续和多元关系的这种复杂结构时,现有的模型实体识别性能不足。针对此问题,文章提出了基于多特征融合的汽车用户需求实体识别模型(Bert-Mulfeat-BiGRU-Att-CRF,BMBAC)。该模型的创新点在于融入了BERT(Bidirectional Encoder Representation from Transformers)模型和SKIP-GRAM模型,以及引入了改进的注意力机制算法,实现了多维度的特征提取和深度挖掘句子的语义信息。为验证模型效果,使用自主爬取并构建的数据集进行实验。通过对比实验,BMBAC模型的F1值为82.79%,展现出最优的识别性能,验证了该模型在汽车用户需求实体识别中的有效性,为汽车企业精准捕捉用户需求、推动产品创新提供了技术和方法。 |
| 关键词: 用户需求;命名实体识别;特征融合;BERT;BiGRU |
|
中图分类号: TP391.1
文献标识码: A
|
| 基金项目: 国家自然科学基金资助项目(72201173);国家重点研发计划资助研究成果(2021YFF0900400);上海市教育科学研究项目(C2023292) |
|
| Research on Named Entity Recognition of Automotive User Requirements Based on Multi-Feature Fusion |
|
ZHANG Yifan, ZHAO Jinghua
|
(School of Management, University of Shanghai f or Science and Technology, Shanghai 200093, China)
222421164@st.usst.edu.cn; zhaojinghua@usst.edu.cn
|
| Abstract: The application performance of traditional Chinese named entity recognition techniques in the automotive field is inadequate, particularly when dealing with complex structures characterized by extensive nested overlaps, discontinuities, and multidimensional relationships between entities. Aiming at this issue, this paper proposes an automotive user demand entity recognition model based on multi-feature fusion (Bert-Mulfeat-BiGRU-Att-CRF, BMBAC). The innovation of this model lies in the integration of BERT (Bidirectional Encoder Representations from Transformers) and the SKIP-GRAM model, as well as the introduction of an improved attention mechanism algorithm, achieving multidimensional feature extraction and deeply mining the semantic information of sentences. To verify the model' s effectiveness, experiments are conducted using a dataset that is independently crawled and constructed. Comparative experiments show that the BMBAC model achieves an F1 value of 82. 79% , demonstrating optimal recognition performance and validating the model's effectiveness in recognizing automotive user demand entities. This provides automotive enterprises with techniques and methods to accurately capture user requirements and promote product innovation. |
| Keywords: user requirements; named entity recognition; feature fusion; BERT; BiGRU |
