隐树模型几个关键指标的辨证意义

數據挖掘的各種算法模型能否真正量化地解釋證候,關鍵是該模型所產生的有關指標能否合理地量化解釋證和一組癥狀的關系。文章以抑郁癥臨床流行病學調查數據構筑的證候隱樹模型為例子,分析了該模型所產生的互信息、累積互信息、信息覆蓋度、條件概率等幾個指標的辨證意義。認為:互信息可作為確定一組與某證有密切關聯的癥狀的依據;累積互信息和互信息數值之間的比較可以把握癥狀提供給證的信息,從而判斷癥狀的診斷價值;信息覆蓋度可以考察與某證相關聯的一組癥狀中究竟有多少癥狀、或有哪些癥狀就足可以把握該證的基本特征;而條件概率則可以通過該證關聯的一組癥狀所表現出的變化來定量地刻畫這個特征。Whether a kind of algorithm model can quantitatively interpret syndrome of traditional medicine, the key is whether some of the indexes from the model can reasonably and quantitatively the relationship between the syndrome and symptoms.And this paper take the latent tree model of sy which be constructed by the data from the depression, as an example, to analyze the differe syndrome meaning of the indexes, such as mutual information (MI) , cumulate mutual information information rate (IR) , conditional probability.Subsequently, the MI may be as a important basis for co the association between a syndrome and a group of symptoms;and meanwhile, according to the com of the MI and CMI, we could be obtained the syndrome information given by symptoms and estim diagnostic value of the symptoms.And then the IR can be used to estimate how many symptoms an symptoms, which are close relevance to one of syndrome, can be took as a confirming of the character of the syndrome, and finally, conditional probability could contribute to a better revea character of a syndrome quantitatively by the changing that a groups of symptoms relevance to a sy had happened on different condition.