Abstract
The primary objective of this work is to develop a Neural Network based on LSTM to predict stock market movements using tweets. Word embeddings, used in the LSTM network, are initialised using Stanford's GloVe embeddings, pretrained specifically on 2 billion tweets. To overcome the limited size of the dataset, an augmentation strategy is proposed to split each input sequence into 150 subsets. To achieve further improvements in the original configuration, hyperparameter optimisation is performed. The effects of variation in hyperparameters such as dropout rate, batch size, and LSTM hidden state output size are assessed individually. Furthermore, an exhaustive set of parameter combinations is examined to determine the optimal model configuration. The best performance on the validation dataset is achieved by hyperparameter combination 0.4,8,100 for the dropout, batch size, and hidden units respectively. The final testing accuracy of the model is 76.14%.
Abstract (translated)
URL
https://arxiv.org/abs/2101.08986
