摘要
针对单一网络特征提取不全面,词向量语义表示质量较低等问题,提出了结合ERNIE2.0和多尺度网络的数字图书馆文本分类模型(ERNIE2.0 and Multi-Scale Network)。采用预训练模型ER⁃NIE2.0提取文本动态特征表示,提升词向量语义表示的准确性。多尺度网络捕捉全局序列和结构信息以及不同层次的局部语义特征,软注意力机制赋予模型关注到重点词的能力。在图书馆数据集上进行实验,结果表明,ERNIE2.0-MSN-AT模型分类准确率达到了97.85%,高于实验对比模型,多尺度网络优于单一特征提取网络。
To address the problems of incomplete feature extraction in a single network and low quality of semantic expression of word vectors,a text classification model of digital library combining ERNIE2.0 and Multi⁃Scale Network is proposed.The pre⁃training model ERNIE2.0 is used to extract the dynamic feature representation of the text to improve the accuracy of the semantic representation of word vectors.The multi⁃scale network simultaneously captures the global sequence and structure information as well as the local semantic features at different levels.The soft attention mechanism gives the model the ability to focus on key words.The experiment on the library data set shows that the classification accuracy of ERNIE2.0-MSN-AT model reaches 97.85%,which is higher than the model compared with the exp⁃eriment,and the multi⁃scale network is better than the single feature extraction network.
作者
陈丽春
CHEN Lichun(The Library of Xi’an Conservatory of Music,Xi’an 710061,China)
出处
《电子设计工程》
2023年第19期1-5,共5页
Electronic Design Engineering
基金
国家自然科学基金项目(61502377)。
