Prediction Of The Growing Stock In Stock Market On Analysis Of The Opinions Using Sentiment Lexicon Extraction And Deep Learning Architectures
Main Article Content
Abstract
Extraction of the sentiment in the social networks and stock exchange is a growing research in the area of data mining. Clustering technique considered to be most essential in discovering the significant stock information about the stock market with respect to various opinion of the traders in the online social network. Conventional clustering technique considered to be inadequate as multiple opinions of the traders is non actionable and sparse in nature. The extracting the sentiment is alone not feasible in the opinion mining as numerous information is related to stock forecasting and stock prediction analysis on the social network data extraction. Subspace clustering technique will be a suitable solution to challenges exhibited in the clustering on inclusion of domain knowledge and parameter sensitive information's. Further thresholding mechanism is capable of predicting the data Sensitiveness. In spite of numerous benefits of the subspace clustering approach, extracting the correct dimension found to be inconsistent and challenging issue. In order to manage those issues, hybrid deep learning architecture named as BERT-CNN model with incorporation of sentiment and domain knowledge has been proposed in this article. Initially the opinion of lexicon is extracted using BERT which is a natural language processing model and extracted feature is represented as word embedding vector. Those embedding vector processed using convolution layers with inbuilt kernel and filter constraints to eliminate the inconsistent dimension. Resultant feature is down sampled in the max pooling layers to obtain high level features as feature map. ReLU Activation Function is applied to fully connected layer to explore the growing stock features on processing the weight of the feature on the particular domain knowledge. Extracted actionable feature representing the growing stocks is clustered and refined periodically in its subspaces on utilizing the domain knowledge. An experimental result on the twitter benchmark dataset containing stock news proves that proposed architecture outperforms other conventional technique on performance metrics such as accuracy and efficiency.