摘要：

Abstract:

【Purpose/significance】 Time series pattern classification of entities in science and technology is a premise for effective

knowledge growth prediction. In this paper, a predictability classification scheme of entity time series in science and technology is pro⁃

posed and verified.【Method/process】 First, two kinds of entities of problem and method are extracted from the abstracts of scientific

and technical literature, and the time series samples of fitting, trending and irregularity entities are marked. Second, dynamic time

warping is used to calculate the shape similarity of time series, curve fitting and locally weighted regression are used to extract features

of time series. Finally, the effects of two methods of time series classification based on shape similarity and features are compared.

【Result/conclusion】 Through the experiment in the field of artificial intelligence, it is found that the features extracted by curve fitting

and weighted local regression can effectively carry out entity time series pattern classification with F1 value up to 0.91. Applying clas⁃

sification to the trend prediction of time series can reduce the error of entity time series prediction.【Innovation/limitation】 Applying

time series mining to entity growth prediction provides new solution for science and technology forecasting. In the future, pay more at⁃

tention to the local features of time series, and then think deeply about the process and causes of entity changes.