基于BERT的电机领域中文命名实体识别方法

针对电机领域实体识别精度较低的问题,提出一种融合BERT预训练语言模型的中文命名实体识别方法。利用BERT预训练语言模型增强字的语义表示并按照上下文特征动态生成字向量,将字向量序列输入双向长短期记忆神经网络进行双向编码,同时通过条件随机场算法标注出实体识别结果。根据电机文本特点对自建数据集进行标注,并将电机领域实体划分为实物、特性描述、问题/故障、方法/技术等4个类别。实验结果表明,与基于Bi LSTM-CRF、Bi LSTM-CNN和Bi GRU的实体识别方法相比,该方法具有更高的准确率、召回率和F1值,并且有效解决了电机领域命名实体识别任务中标注数据不足及实体边界模糊的问题。

在线教学视觉隐私保护学生专注度监测研究

新冠疫情下,在线教学已成为学校教学的重要形式,这就引发了教学活动方式转变,特别是学生专注度监测等问题面临隐私保护和数据冗余等挑战。为此,文章引入视觉隐私保护机制,提出面向在线教学过程的新型智能化学生专注度监测方案,其系统源于多层压缩感知、Cascade RCNN、改进LBP以及类字典学习等技术创新融合。对各环节,文章分别选择配套数据集进行实验并分析论证。结果表明:相较于传统思路,所提方案在视觉隐私保护、人脸提取和专注度监测等方面具有明显优势。本研究有利于改善在线教学过程的学习质量监管,同时也为疫情下教学技术发展提供了探索和积累。

News recommendation based on time factor and word embedding

Existing algorithms of news recommendations lack in depth analysis of news texts and timeliness. To address these issues, an algorithm for news recommendations based on time factor and word embedding(TFWE) was proposed to improve the interpretability and precision of news recommendations. First, TFWE used term frequency-inverse document frequency(TF-IDF) to extract news feature words and used the bidirectional encoder representations from transformers(BERT) pre-training model to convert the feature words into vector representations. By calculating the distance between the vectors, TFWE analyzed the semantic similarity to construct a user interest model. Second, considering the timeliness of news, a method of calculating news popularity by integrating time factors into the similarity calculation was proposed. Finally, TFWE combined the similarity of news content with the similarity of collaborative filtering(CF) and recommended some news with higher rankings to users. In addition, results of the experiments on real dataset showed that TFWE significantly improved precision, recall, and F1 score compared to the classic hybrid recommendation algorithm.