A Hybrid Learning Particle Swarm Optimization With Fuzzy Logic for Sentiment Classification Problems

Abstract

Methods based on deep learning have great utility in the current field of sentiment classification. To better optimize the setting of hyper-parameters in deep learning, a hybrid learning particle swarm optimization with fuzzy logic (HLPSO-FL) is proposed in this paper. Hybrid learning strategies are divided into mainstream learning strategies and random learning strategies. The mainstream learning strategy is to define the mainstream particles in the cluster and build a scale-free network through the mainstream particles. The random learning strategy makes full use of historical information and speeds up the convergence of the algorithm. Furthermore, fuzzy logic is used to control algorithm parameters to balance algorithm exploration and exploration performance. HLPSO-FL has completed comparison experiments on benchmark functions and real sentiment classification problems respectively. The experimental results show that HLPSO-FL can effectively complete the hyperparameter optimization of sentiment classification problem in deep learning and has strong convergence.