【目的/意義】熱點輿情識別對社交媒體監管有重要意義,已有方法大多基于語義分析和社會網絡分析技術,忽略了信息傳播中隱含的動態時序信息。【方法/過程】本文利用卷積神經網絡,提取熱點輿情在社交網絡中的多層次傳播特征;然后與主題分析模型相結合,設計了熱點輿情識別方法。本方法利用了輿情熱度與其傳播過程間的潛在關聯,擺脫了對語義信息和社會網絡信息的過度依賴,適用于歷史數據匱乏或缺失的識別場景。【結果/結論】實驗表明,本方法顯著提升了熱點輿情的識別精確度,具有一定適應性和可擴展性。

【Purpose/significance】The recognition of hot events in social media is significant for public opinion monitoring.Existing methods are mainly based on semantic analysis and social network analysis, but neglect the sequential data of information diffusions.【Method/process】In contrast, we integrate topic model into convolutional neural network to learn the topic features and latent information spreading features from hot event datasets. Our method overcomes the strong dependences on semantic data and social network information, and can be applied to the scenarios without sufficient historical data.【Result/conclusion】Experiment shows that our method improves the recognition accuracy, and has good extendibility and robustness.