Implementation of Bidirectional Long Short-Term Memory for Stock Entity Identification

Abstract

One of the financial products in the capital market that is in great demand is stock. Shares are proof of ownership of a company that fluctuates and tends to have a high level of risk and nonlinear price changes. To make the right investment decision, investors are required to be able to analyze the abundant stock information carefully and quickly. In facing this challenge, Named Entity Recognition (NER) can be a potential solution in analyzing stock information by recognizing stock entities and grouping them into certain labels. In this research, NER is developed with the Bidirectional Long Short-Term Memory algorithm, which is used to identify five stock entities: company name, stock code, stock index, industry sector, and sub-sector. With an accuracy of 99.81% on the test data, the Bi-LSTM algorithm can identify the entities well and group each token into the five entities.