Public opinion prediction on social media by using machine learning methods

Nowadays, the willingness of the public to express their opinions on social media has extremely increased, being facilitated by the online social network. As a result, public opinion events pose challenges for decision makers in public opinion prediction technologies. However, the shortcomings of existing models include low accuracy of the clustering method, leading opinion detection, and scale prediction of public opinion. Emerging from this objective, this paper introduces a Public Opinion Prediction (POP) model whose predictive accuracy and computational efficiency are transformative by employing machine learning methods, which can well predict not only the scale and trend, but also can accurately predict the opinions of the public on social media. The POP model consists of three parts: (1) the Preference-based online social Network Clustering(NPC) method to decrease the dimensions, (2) the improved Whale Optimization Algorithm based on the Leading Opinion Detection(WOA-LOD) algorithm to detect the leading opinions in online social networks, and (3) the Susceptible Individuals Removed model with Death and Birth rate(SIRDB) to predict and simulate the development tendency and scales of the public opinions. By implementing the POP model in real data which includes two datasets with 359 and 898 users respectively in Weibo social media and comparing it with other existing methods. As a result, NPC and WOA-LOD achieve a 60%–70% improvement in accuracy for cluster method and leading opinions detection; SIRDB achieves a greater than 95% improvement when comparing other traditional methods on the accuracy of scale prediction. All experiment results show the POP model exhibits state-of-the-art performance in not only detecting the leading opinions but also prediting the scale and tendency, which performs perfectly in practical management.