摘要
目前科研热点预测严重依赖于本领域高级专业人员通过大量文献查阅与市场调研的方法确定。随着科技文章及文献数量的迅速增长,单纯人工跟踪和分析科研热点变得困难。简单的统计方式如关键词和文本中词频较高的词来表示热点话题,耗时且准确性欠佳。基于机器学习算法且考虑数据时序变化的关系,提出一种科研热点预测算法。同时,为了缩短算法运行时间、提高执行效率,一种基于阈值的预测加速方法被加入预测框架。实验表明,提出的科研热点预测算法较基准算法查全率平均提高25.75%,查准率平均提高28.25%。
At present,predicting hotspots depends on document consulting and market survey by advanced professionals.With the rapid growth of scientific documents and literatures,it becomes difficult to artificially trace and analyze scientific research hotspots.Simple statistics on keywords or frequencies is time-consuming and their accuracy is low.In the paper we propose a novel prediction approach of scientific research hotspot based on artificial intelligence algorithm,which considers the data changes over time.Moreover,we add an acceleration step into prediction framework using threshold method in order to improve execution efficiency.Our extensive experiments demonstrate that our proposed algorithm has 25.75%better recall ratio and 28.25%higher precision ratio than benchmark algorithm.
作者
马艳
韩英昆
齐达立
刘科
MA Yan;HAN Yingkun;QI Dali;LIU Ke(Shandong Electric Power Research Institute,Jinan 250003,China;State Grid Shandong Electric Power Research Institute,Jinan 250003,China)
出处
《山东电力技术》
2020年第8期5-9,共5页
Shandong Electric Power
基金
国家自然科学基金项目“面向互联网开放域的弱监督关系抽取关键问题研究”(61703234)
山东省高等学校科技计划项目“关注度驱动的证券市场短期炒作热点预测技术研究”(KJ2018BAN046)。
关键词
科技情报
人工智能
时序数据
预测
聚类
神经网络
scientific and technical information
artificial intelligence
time series data
prediction
cluster
neural network