【目的/意義】微博情感分析對公共安全事件管控有著重要意義。現有研究將單條微博作為整體進行分析,情感分析最小單元局限于字或詞,而對微博從詞到句子,從句子到單條微博這種多層粒度文本結構產生的影響關注不足,基于此本文提出一種融合雙層注意力的Bi-LSTM模型提升情感分析性能。【方法/過程】以紅黃藍幼兒園涉嫌虐童事件為例,通過Bi-LSTM提取微博詞級和句子級特征,結合雙層注意力機制學習各級特征權重分布,以遞進順序綜合局部情感得到整條微博的情感分類。【結果/結論】實驗結果表明,本研究提出的微博情感分析模型F1值、準確率分別達到97.39%、97.62%,相比于SVM、RF、XGBOOST和LSTM,該模型能夠在公共安全事件微博情感分析方面取得較好效果。

【Purpose/significance】The micro-blog sentiment analysis is of great significance for the management and control of public safety events. Previous studies often analyze a single micro-blog as a whole and the minimum unit of sentiment analysis is limited to words, but insufficient attention is paid to the influence of the structure of the micro-blog text that from words to sentences and from sentence to single micro-blog. In order to solve this problem and improve analytical performance, a bidirectional long short term memory model with dual-layer attention is proposed.【Method/process】Taking the incident of child abuse in the Red, Yellow and Blue kindergarten as an example, the micro-blog word level and sentence level features are extracted through Bi-LSTM, and the dual-layer attention mechanism is used to learn the weight distribution of characteristics of each level,then integrate local sentiments in a progression order to get the sentimental classification results of the whole micro-blog.【Result/conclusion】The experimental results show that the F1-score and accuracy of the micro-blog sentiment analysis model proposed in this study reaches 97.39% and 97.62%.Compared with SVM, RF,XGBOOST and LSTM, better results can be achieved in this micro-blog sentiment analysis model of public security events.