Abstract: 【Purpose/significance】This paper constructs a large-scale acknowledgements function dataset from academic text,and pro?
poses a SciBERT-based acknowledgements function recognition model,in order to provide high-quality data support and effective rec?
ognition method for acknowledgements mining and analysis.【Method/process】The classification rules of acknowledgements function
are extended and improved manually,then the automatic indexing rules template of acknowledgements function is generated,and 1,750, 275 acknowledgement sentences are labeled.Based on this,the SciBERT model is used to carry out vector representation of acknowl? edgement texts,and a Softmax regression model is introduced to do the automatic text classification.Besides,warmup strategies are ad? opted to optimize the model,and we compared the proposed model with the benchmark methods.【Result/conclusion】We obtained a large-scale and high-quality dataset of acknowledgments function,and the precision of the labeling results reaches 93%.The model proposed in this paper performs better than the latest benchmark experiments.On the expanded dataset,the F1 value of the proposed model is higher than 98%,and the prediction results in various categories have also been improved to varying degrees.【Innovation/limi? tation】The proposed model lacks the consideration and fusion of unique features of acknowledgment texts