【目的/意義】隨著網絡信息傳遞的快速增長,對網絡輿情的實時監控日益重要,由此帶來了不斷攀升的管理成本。因此利用無監督的輿情預警來替代人工監控具有普遍的實踐意義。【方法/過程】基于SIS的傳染病模型,具體分析了輿情傳播的時間演化特征,定義了輿情爆發的臨界時刻。通過對輿情擴散系數以及遺忘率的分析,實現了以無監督預警時間為目標的優化策略。【結果/結論】優化結果顯示:預設的擴散系數并非越小越好,在無監督預警時間達到最大時,存在一個最優的擴散系數。同時,最優擴散系數可以隨著輿情傳播環境的變化實現動態的自適應調整,始終保持無監督預警效用的最大化。

【Purpose/significance】With the rapid growth of message transfer on Internet, the real-time monitoring of message propagation is getting more and more important. However, it results in increased cost of management. Therefore, it is of general practical meaning to replace the manual monitoring by unsupervised propagation alert.【Method/process】Based on the SIS epidemic model, the characteristics of time evolution of message propagation on Internet are analyzed, as well as the definition of the critical time of burst. By using the coefficient of message diffusion and the rate of forgetting, we build up an optimized strategy for extending the span of unsupervised alert.【Result/conclusion】The optimized results show that the span of unsupervised alert is not always increasing as the diffusion coefficient decreased. There exists an optimal diffusion coefficient maximizing the span. Moreover, the optimized diffusion coefficient is self-adapted when the message propagation circumstance changed. Therefore, the effect of the unsupervised alert strategy can be maximized all the time.