【目的/意義】微博是用戶情感發泄的重要渠道,預警模型將有助于發現異常情緒用戶,以便及時展開干預。【方法/過程】模型首先利用極性詞典和句法序列規則計算微博情感極性程度值,過濾出可疑情緒異常節點;然后利用微博社交網互動關系,計算節點之間的信任值,進一步通過可信反饋對情緒異常節點進行判斷。【結果/結論】實驗表明,基于序列規則+詞典比基于詞典的方法對可疑異常情緒用戶過濾準確性高,而相比這兩種文本挖掘的方法,將可信反饋加入異常情緒判斷進一步提高了識別準確度。

【Purpose/significance】Micro-blog is an important way for blogger to vent emotion. The warning model can helpfind some abnormal emotion users, so that a timely intervening can prevent the occurrence of some extreme behavior.【Method/process】The model calculated micro-blog sentiment degree by polarity lexicon and syntax, and filtered out thesuspicious abnormal emotional node. It calculated trust value among micro-blog users through social network interaction,and further to judge the abnormal emotion users through trusted feedback.【Result/conclusion】The experiment shows thatsyntax rule & dictionary-based has a higher accuracy than dictionary-based when filtering the suspicious abnormalemotion users. Compared to the two ways of text mining, trusted feedback can further improve the recognition accuracy.