可信度模型
1. 阈值可信度模型
1️⃣表示不确定性:
IF E THEN H(CF(H,E),λ)
- C F ( E ) ∈ [ 0 , 1 ] CF(E)\in{[0,1]} CF(E)∈[0,1]:证据 E E E的可信度, C F ( E ) = 1 / 0 CF(E)=1/0 CF(E)=1/0表示证据绝对存在/不存在
- C F ( H , E ) ∈ [ 0 , 1 ] CF(H,E)\in{}[0,1] CF(H,E)∈[0,1]:证据 E E E为真时 H H H的可信度, C F ( H , E ) = 0 , 1 CF(H,E)=0,1 CF(H,E)=0,1对应 P ( H ∣ E ) = 0 , 1 P(H|E)=0,1 P(H∣E)=0,1
- λ ∈ [ 0 , 1 ] \lambda{}\in{}[0,1] λ∈[0,1]:阈值,只有证据 E E E可信度 C F ( E ) ≥ λ CF(E)\geq\lambda{} CF(E)≥λ时知识 E E E才能被应用
2️⃣组合证据的不确定性
- C F ( E 1 ∧ E 2 ∧ . . . ∧ E n ) = m i n { C F ( E 1 ) , C F ( E 2 ) , . . . , C F ( E n ) } CF(E_1\land{}E_2\land{}...\land{}E_n)=min\{CF(E_1),CF(E_2),...,CF(E_n)\} CF(E1∧E2∧...∧En)=min{CF(E1),CF(E2),...,CF(En)}
- C F ( E 1 ∨ E 2 ∨ . . . ∨ E n ) = m a x { C F ( E 1 ) , C F ( E 2 ) , . . . , C F ( E n ) } CF(E_1\lor{}E_2\lor{}...\lor{}E_n)=max\{CF(E_1),CF(E_2),...,CF(E_n)\} CF(E1∨E2∨...∨En)=max{CF(E1),CF(E2),...,CF(En)}
3️⃣不确定性的传递算法: C F ( E ) ≥ λ CF(E)\geq\lambda{} CF(E)≥λ时 C F ( H ) = C F ( H , E ) × C F ( E ) CF(H)=CF(H,E) × CF(E) CF(H)=CF(H,E)×CF(E)
4️⃣结论不确定性合成算法:
IF Ei THEN H(CF(H,Ei),λi) i=1,2,...,n
前提: ∀ i ∈ { 1 , 2 , . . . , n } , C F ( E i ) ≥ λ i \forall{i}\in\{1,2,...,n\},\,CF(E_i)\geq\lambda{}_i ∀i∈{1,2,...,n},CF(Ei)≥λi
结论 H H H的可信度:
方法 C F ( H ) CF(H) CF(H)等于 极大值法 max { C F i ( H ) } i = 1 , 2 , . . . , n \max\{CF_i(H)\}\,\,i=1,2,...,n max{CFi(H)}i=1,2,...,n 加权和法 ∑ i = 1 n C F ( H , E i ) × C F ( E i ) ∑ i = 1 n C F ( H , E i ) \cfrac{\displaystyle{}\sum_{i=1}^{n}CF(H, E_i)\times CF(E_i)}{\displaystyle{}\sum\limits_{i=1}^{n}CF(H, E_i)} i=1∑nCF(H,Ei)i=1∑nCF(H,Ei)×CF(Ei) 有限和法 m i n { ∑ i = 1 n C F i ( H ) , 1 } min\{\sum\limits_{i=1}^{n}CF_i(H),\,1\} min{i=1∑nCFi(H),1} 递推法 C 0 = 0 C_0=0 C0=0, C k = C k − 1 + ( 1 − C k − 1 ) × C F ( H , E k ) × C F ( E k ) C_k = C_{k-1} + (1 - C_{k-1}) \times CF(H, E_k) \times CF(E_k) Ck=Ck−1+(1−Ck−1)×CF(H,Ek)×CF(Ek)
2. 加权可信度模型
1️⃣知识的不确定性:
- 表示:
IF E1(w1) AND E2(w2) AND ... AND En(wn) THEN H (CF(H,E),λ)
- 阈值 λ \lambda{} λ:由专家给出
- 加权因子 ω i \omega_i ωi:由专家给出,满足 ∑ i = 1 n ω i = 1 \sum\limits_{i=1}^{n}\omega_i=1 i=1∑nωi=1
2️⃣组合证据的不确定性: C F ( E ) = ∑ i = 1 n [ ω i ∗ C F ( E i ) ] ∑ i = 1 n ω i CF(E)=\cfrac{\sum\limits_{i=1}^{n}[\omega_i*CF(E_i)]}{\sum\limits_{i=1}^{n}\omega_i} CF(E)=i=1∑nωii=1∑n[ωi∗CF(Ei)]
3️⃣不确定性传递算法:当 C F ( E ) ≥ λ CF(E)\geq\lambda{} CF(E)≥λ时 C F ( H ) = C F ( H , E ) × C F ( E ) CF(H)=CF(H,E)\times{}CF(E) CF(H)=CF(H,E)×CF(E)
3. 前件带不确定性的可信度模型
1️⃣表示:
- 表示1: c f i cf_i cfi是子条件 E i E_i Ei的可信度由专家给出,子证据 E i E_i Ei的可信度 c f i ′ cf_i' cfi′
IF E1(cf1) AND E2(cf2) AND ... AND En(cfn) THEN H (CF(H,E),λ)
- 表示2:加上权值
IF E1(cf1,w1) AND E2(cf2,w2) AND ... AND En(cfn,wn) THEN H (CF(H,E),λ)
3️⃣不确定性匹配
- 无加权因子: ∑ i = 1 n max ( 0 , c f i − c f i ′ ) ≤ λ \sum\limits_{i=1}^{n} \max(0, c_{f_i} - c'_{f_i}) \leq \lambda i=1∑nmax(0,cfi−cfi′)≤λ
- 有加权因子: ∑ i = 1 n ω i × max ( 0 , c f i − c f i ′ ) ≤ λ \sum\limits_{i=1}^{n} \omega_i \times \max(0, c_{f_i} - c'_{f_i}) \leq \lambda i=1∑nωi×max(0,cfi−cfi′)≤λ
4️⃣不确定性的传递算法
- 无加权因子: C F ( H ) = [ ∏ i = 1 n ( 1 − max { 0 , c f i − c f i ′ } ) ] × C F ( H , E ) CF(H) = \left[ \prod\limits_{i=1}^{n} (1 - \max\{0, c_{f_i} - c'_{f_i}\}) \right] \times CF(H,E) CF(H)=[i=1∏n(1−max{0,cfi−cfi′})]×CF(H,E)
- 有加权因子: C F ( H ) = [ ∏ i = 1 n ( 1 − ω i max { 0 , c f i − c f i ′ } ) ] × C F ( H , E ) CF(H) = \left[ \prod\limits_{i=1}^{n} (1 - \omega_i\max\{0, c_{f_i} - c'_{f_i}\}) \right] \times CF(H,E) CF(H)=[i=1∏n(1−ωimax{0,cfi−cfi′})]×CF(H,E)