Oʻzbekiston respublikasi оliy va oʻrta maxsus

Listed below are some of the parameters:

• 
  N is the number of classes (in our scenario, N=2: subjective and objective classes);
  
• 
  M is the number of unique words (terms) in the corpus; R is the number of observed sequences in
  

the training process.

• 
  R is the number of sequences seen during the training procedure;
  

• 
  𝑂^𝑟 = ^{𝑜^𝑟, 𝑜^𝑟, . . . 𝑜^𝑟}are sentences from the training dataset, where Tr is the length of the r-th
  

1 2 𝑇
sentence, 𝑟 = 1,2, … , 𝑅;

• 
  𝜇_𝑖,𝑗 explains the relationship between the i-th term and the j-th class ( 𝑖 = 1, … , 𝑀; 𝑗 = 1,2, … , 𝑁
  
• 
  𝑐_𝑖,𝑗 is the frequency with which the 𝑖-th phrase appeared in class 𝑗;
  
• 
  𝑡_𝑖 = ^∑_𝑗 𝑐_𝑖,𝑗 represents the number of occurrences of the 𝑖-th phrase in the corpus;
  

• 
  The incidence of the I -th phrase in the j -th class.
  

^𝑐𝑖,𝑗

𝑐̅̅_𝑖̅_,̅_𝑦 =
;
𝑡_𝑖

We provide a novel weighting parameter that influences system accuracy by taking the number of classes instead of the number of documents in the well-known IDF (Inverse-Document Frequency) calculation. We call it Pruned ICF, which is similar to IDF (Inverse-Class Frequency)

𝐼𝐶𝐹_𝑖
= log
𝑁

,
² 𝑑𝑁_𝑖

When 𝑖 is a term, 𝑑𝑁 is the number of classes that include the term 𝑖 and 𝑐_𝑖,𝑗 > 𝑞, and where q is a constant

𝑞 =
1



𝛿 ∙ 𝑁

The optimal value of 𝛿 for the corpus examined is 𝛿 =1.4, which is determined experimentally.
The membership degree (𝜇_𝑖,𝑗) of words for relevant classes may be determined by experts or com- puted using analytic formulae. To avoid needing human annotation or lexical expertise, we determined the membership degree of each phrase using the following analytical formula: 𝑖 = 1, … , 𝑀; 𝑗 = 1,2, … , 𝑁:

𝑇𝐹: 𝜇
₌ ^𝑐𝑖^̅ ,𝑗
(1)

∑
𝑖,𝑗
𝑁
𝑣=1
^𝑐𝑖^̅ ,𝑣

𝑇𝐹 ∙ 𝐼𝐶𝐹: ∶ 𝜇
₌ 𝑐_𝑖̅ _,𝑗∙𝐼𝐶𝐹_𝑗
(2)

∑
𝑖,𝑗
𝑁
𝑣=1
𝑐_𝑖̅ _,𝑣∙𝐼𝐶𝐹_𝑣