基于自适应增强（AdaBoost）的径向基（RBF）神经网络改进算法在关键词预测中的应用

Application of AdaBoost-Based RBF Neural Network Algorithm for Keyword Prediction

摘要:

探究学科或领域内研究发展趋势和热点?直以来受到国内外学者们重点关注，??频关键词的频次变化分析是其中重要的研究内容。关键词的变化与时间存在强相关性，但当前仅有少数研究考虑了关键词随时间密切变化的特性。在考虑关键词信息的时间属性基础上，提出?种基于自适应增强（AdaBoost）的径向基（RBF）神经?络预测算法（以下简称“RBF改进算法”），对关键词频次进?分析预测。对中国知网2007－2022年收录的医学图像期刊论文关键词进?处理，其中将2007年至2021年的数据作为实验训练数据，2022年数据作为验证数据，通过算例分析，对比RBF改进算法、反向传播算法和时间序列算法对关键词词频的预测结果。结果发现：通过AdaBoost算法对RBF算法进行改进，能够增强RBF神经?络的泛化能?以及对样本的适应性，同时保留了RBF神经网络较好的非线性映射能力这一优点；RBF改进算法预测结果与实际数据接近，期预测精度优于BP神经网络和时间序列算法，该算法的预测效果更佳。

Abstract:

Exploring the development trends and hotspots of research in disciplines or fields is always an important topic for the domestic and foreign scholars, and analyzing the frequency changes of high-frequency keywords is an important research content. There is a strong correlation between changes in keywords and time, but currently only a few studies have considered the characteristic of keywords changing closely over time. Considering the time attribute of keywords, an Adaboost-based RBF neural network forecasting algorithm (hereinafter referred to as "Improved RBF algorithm") was proposed and applied to the keyword frequency forecasting and analyzing. The keywords of medical image journal papers from 2007 to 2022 collected from CNKI were processed. Taking the data from 2007 to 2021 as the training data and the data from 2022 as the validation data, the prediction results of keyword frequency by Improved RBF algorithm, back propagation (BP) neural network algorithm and auto-regressive moving-average (ARMA) time series analysis algorithm were compared through the analysis of examples. The results showed that improving the RBF algorithm through AdaBoost algorithm can not only enhance the generalization ability and adaptability of the RBF neural network to samples, but also retain good nonlinear mapping ability of the RBF neural network, and the prediction results of the Improved RBF algorithm were close to the actual data, and the prediction accuracy was better than BP neural network algorithm and ARMA time series algorithm. Thus the prediction performance of the proposed algorithm is better.