Leveraging NLP and deep learning models for political opinion mining: sentiment classification and sequential analysis of news media influence on elections

Abstract

Although the impact of digital news media on public opinion has grown significantly, there are currently no computational frameworks that empirically connect news mood to election results. This research suggests an artificial intelligence (AI)-based solution that uses deep learning and natural language processing to analyze how political news sentiment affects election outcomes. A dataset of 17,000 political news stories gathered during the general elections in Pakistan in 2024 was manually classified into three categories: good, negative, and neutral. In order to capture contextual and temporal relationships in political speech, the texts underwent normalization, tokenization, lemmatization, and modeling using a Bidirectional Long Short-Term Memory (BiLSTM) network. To evaluate media influence, party-level mood indices were calculated using model predictions and statistically associated with official vote shares. According to experimental results, the suggested BiLSTM model outperforms conventional machine learning baselines and reaches an accuracy of 89%. Additionally, sentiment indices and electoral results show a statistically significant link, suggesting a quantifiable relationship between media mood and voter behavior. The suggested paradigm offers empirical insights into the function of news sentiment in election dynamics and a repeatable technique for extensive political media study.