Real-Time Detection of Public Relations Crisis Signals Using Hybrid LSTM and Isolation Forest Model

Abstract

With the widespread use of social media, public relations crises have become increasingly rapid and unpredictable. Traditional manual monitoring and public opinion analysis methods are difficult to cope with the rapidly developing information environment. To address this challenge, this study proposes a new real-time PR crisis signal detection model that effectively combines long short-term memory networks and isolation forest algorithms. This model aims to achieve early warning of potential crises by efficiently processing large amounts of social media and news data. The experiment was conducted on a real-world public relations crisis event dataset, which includes text, user interaction, and sentiment analysis data. The performance of the proposed fusion model was compared with independent LSTM and isolated forest models as baselines. The experimental results show that the fusion model performs robustly in various types of public relations crisis signals. It achieved significant high accuracy in detecting social media popularity and public opinion fluctuations, improving by 3.4% and 6.3% respectively compared to single LSTM and Isolation Forest models. In addition, the fusion model exhibits significantly enhanced robustness in high noise environments. At a pollution rate of 60%, it maintained an accuracy of 54.1% and an F1 score of 0.823, demonstrating its powerful anomaly detection capability. The proposed fusion model provides an accurate, real-time, and robust solution for PR crisis signal detection by utilizing the complementary advantages of LSTM to capture time patterns and isolation forest to identify anomalies, which has significant practical application value for enterprises and organizations.