Oʻzbekiston respublikasi оliy va oʻrta maxsus

Fuzzy Control System for Detecting Subjectivity. In the first step, we estimate the membership function using a statistical technique rather than expert knowledge. Then, using the back-propagation technique, we perform fuzzy operations and alter parameters. We now present our algorithm ( 𝑟 = 1, 2, … , 𝑅).

1. 
   
   𝑖,𝑗
   Formulas (1)-(2) are used to compute the membership degree of words ( μ^𝑟 ) in the 𝑟 -𝑡ℎ sen-
   

tence.



¹ https://huggingface.co/Sanatbek/uzbek-restaurant-domain-reviews/tree/main

term.

2. 
   Maximum membership level is determined with regard to courses for each term of the 𝑟 − 𝑡ℎ
   

𝜇̅^𝑟 = 𝜇^𝑟 ,

𝑖,𝑗 𝑖,𝑗

𝑖,𝑣
𝑗 = 𝑎𝑟𝑔 max_{1≤𝑣≤𝑁} 𝜇^𝑟 , (3)
𝑖 = 1, … , 𝑀.

3. 
   Maxima means are computed for all classes:
   

^∑𝑘∈𝑍 ^𝜇̅𝑟

𝜇̅ =
𝑟
𝑖,𝑗
𝑇_𝑟
𝑖,𝑗_,

𝑍^𝑟 = {𝑖: 𝜇̅^𝑟 = max_{1≤𝑣≤𝑁} 𝜇̅^𝑟 }, (4)
𝑗 𝑖,𝑗 𝑖,𝑗
𝑗 = 1, … , 𝑁.
For the defuzzification procedure, we use the Center of Gravity Defuzzification (CoGD) approach. A fuzzy control model is used to train objective and subjective statements chosen according to classes. This is how the objective function is defined [2]:
_{1 ∑}^𝑁 ̅_𝜇̅̅̅^𝑟̅_𝑦 ²

𝐸⁽𝑦⁾ =
_{∑𝑁 (} 𝑗=1 ^{𝑦 𝑗} _{− 𝑑 )}

→ min
𝑁,
(5)

𝑦

∑
₂ 𝑗=1
𝑁
𝑗=1
̅_𝜇̅̅̅^𝑟̅
𝑦∈𝑅

𝑦 = (𝑦₁, 𝑦₂, , … , 𝑦_𝑁) desired output.This function’s partial derivatives are computed using the following formula:

𝑗
𝑅
𝜕𝐸(𝑦)
= ∑

𝜇̅^𝑟
𝑁

∑

𝑦

∑
₍ 𝑗=1
𝑚̅𝑦_𝑗

𝑦
− 𝑑

) , 𝑡 = 1,2, … , 𝑁.

𝜕𝑦_𝑡

𝑟=1
𝑁

∑
𝑗=1
_𝜇̅̅_̅^𝑟̅
𝑁
𝑗=1
̅_𝜇̅_̅^𝑟̅ ^𝑟

Using the optimum values of 𝑦^∗, the conjugate gradient approach minimizes function (5). The index of the classes gained in the result is shown by rounding 𝑦^̅:
_∑^𝑁 _𝜇̅_̅̅̅^𝑟̅_𝑦^∗
𝑦^̅ = ^{𝑗=1 𝑦 𝑗}. (6)

𝑦

∑
𝑁
𝑗=1
Acceptance technique (𝑠):
̅_𝜇̅̅̅^𝑟̅
𝑠 =


𝑖_𝑠𝜖𝐼, 𝑖𝑓 𝑦^̅𝜖(𝑖_𝑠 − ∆₁, 𝑖_𝑠 + ∆_1,)
^{ 𝑟𝑒𝑓𝑒𝑐𝑡, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

where 𝑖_𝑠 is the proper class index and 𝐼 = 1,2, . . . , 𝑁 Here, ∆₁∈ [0; 0.5] is the major quantity in- fluencing dependability of system? It is simple to determine which feature vector produces the greatest results for 𝐹𝐶𝑆. Table 1 displays the average accuracy of 𝐹𝐶𝑆 based on (1)-(2) features in the unrestricted
scenario over tenfold cross validation. It should be noted that these findings are dependent on the classification technique and may change for various classifiers.

Features	Accuracy (%)
𝑇𝐹	89.87
𝑇𝐹 ∙ 𝐼𝐶𝐹	91.3

Table 1: 𝐹𝐶𝑆 results based on 𝑇𝐹 and 𝑇𝐹 ∙ 𝐼𝐶𝐹 characteristics.
We also examined FCS based on Delta TFIDF characteristics [3]. As the DeltaIDF weighting coef- ficients for all classes are identical, the use of DeltaIDF weighting has no effect on the FCS's precision. As seen in Table 1., the accuracy of the approach rises when pruned ICF weighting is used. Table 2 dis- plays the results of subjectivity detection by FCS with various ∆₁values based on 𝑇𝐹 ∙ 𝐼𝐶𝐹. It can be ob- served that the rejection rate for ∆₁= 0.5 is 0.01 percent. During testing, 0.01% of the sentences include these phrases, which after ICF pruning becomes 0 and the system rejects these sentences.

	Correct (%)	Rejection (%)	Error(%)
∆1= 0.3	76.41	20.86	2.73
∆1= 0.4	85.11	10.14	4.75
∆1= 0.5	91.3	0.01	8.69

Table 2. Average results of 10 folds cross validation accuracy of FCS using 𝑇𝐹 ∙ 𝐼𝐶𝐹 feature with varying alpha values.

Figure 1. The structure of MANN in ANFIS.


Detecting subjectivity using an Adaptive Neuro Fuz [4]zy Inference System. Fig. 1 depicts the basic framework of the Adaptive Neuro Fuzzy Inference System. As a result of linguistic assertions, the fuzzy interface block offers a vector input to a Multilayer Artificial Neural Network (MANN) [4]. In the first step, we employed statistical estimate of term membership degree using (2) rather than linguistic claims. We then used fuzzy operations (3) and (4).
MANN was applied to the fuzzyfication operation's output. The neural network input vector is derived from the fuzzyfication operation output vector (fig. 2). The outputs of MANN are used as indices of sentence-appropriate classifications. The backpropagation method is used to train MANN.
We provide two stipulations for the acceptance decision:
1. 𝑦̅_𝑘 ≥ ∆₂,
2. 𝑦̅_𝑘 − 𝑦̃_𝑝 ≥ ∆₃,