融合时序分类的科技领域实体增长预测研究

Research on Entity Growth Prediction in Science and Technology Field Fusing Time Series Classification

[目的/意义]科技领域实体增长机制是进行预测型科技情报分析的核心,要有效地进行科技发展态势感知和预测,必须深入了解科技领域细粒度知识增长机制。文章提出并验证了融合时序分类的科技领域实体增长预测方案。[方法/过程]首先,根据实体词频时间序列的增长模式,将实体分为可拟合、有趋势和无规律3种类别;其次,利用曲线拟合、局部加权回归方法抽取特征,构建特征向量,再用MLP模型实现了高精度的实体分类;最后,构建融合时序分类的集成模型与基线模型进行对比验证效果。[结果/结论]根据实验结果,验证显示集成预测模型相对于基线模型,误差减少了13%以上,进一步证实了预测结果在科学性和准确性方面的优势。综合考虑,所提出的融合时间序列分类的实体增长预测方案在可行性和应用价值方面具备潜力。

[Purpose/significance]The growth mechanism of entities in the field of science and technology is integral to the predictive analysis of scientific and technological information.To effectively anticipate advancements in science and technology,it is vital to deeply explore the nuanced knowledge growth mechanisms within the field.This paper proposes and validates a scheme for predicting the growth of entities within the science and technology domain using time-series classification.[Method/process]Firstly,based on the growth pattern of entity word frequency time series,entities are categorized into three classes:fitable,trend,and irregular.Then,curve fitting and local weighted regression methods are employed to extract features and construct feature vectors,while a Multilayer Perceptron(MLP)model is used to achieve high-precision entity classification.Subsequently,an integrated model fusing time series classification is constructed for comparison with the baseline model to verify its effectiveness.[Result/conclusion]According to our experimental results,the integrated prediction model reduces error by over 13%compared to the baseline model,further attesting to its advantages in terms of scientific robustness and accuracy.In conclusion,the proposed scheme of predicting entity growth in the science and technology field,which incorporates time series classification,holds significant potential for feasibility and application value.