Telecommunication potential customer evaluation method based on probabilistic language three-way decision

Abstract

The invention discloses a probabilistic language three-way decision-making-based telecommunication potential customer evaluation method. The method comprises the steps of firstly grouping target information according to decision requirements, converting a real number data set into a probability language decision matrix, extracting probability language term set features and constructing an expectation-reliability score function; respectively obtaining subjective and objective weights of attributes through an analytic hierarchy process and a variable coefficient method, and solving a comprehensive weight by means of a cosine similarity maximization model; then, a relative loss function is constructed based on the score matrix, a comprehensive foreground value of the scheme is calculated, and (non) harmony indexes and net credibility indexes of the scheme are measured and calculated under each attribute in combination with attribute non-compensability and an ELCTRE thought; and finally, making a series of decision rules through the expectation relative loss function. According to the method, the telecommunication company can be assisted to efficiently classify and evaluate the users, scientific guidance is provided for decision making, and the accuracy of potential customer mining and management is improved.

Description

Technical Field

[0001] This invention relates to the field of telecommunications potential customer assessment technology, and specifically to a telecommunications potential customer assessment method based on probabilistic language three-way decision-making. Background Technology

[0002] Telecommunications potential customer assessment is a crucial step for telecommunications companies to identify high-value partners, formulate precise marketing strategies, and optimize resource allocation. Its core lies in comprehensively analyzing various factors related to customers to assess their cooperation potential and value, providing strong support for the company's business decisions. During the assessment process, due to the complexity and dynamism of the market environment, information about potential customers is often incomplete and uncertain. For example, information such as a customer's true needs, financial situation, and willingness to cooperate may be concealed or vaguely stated. Furthermore, the reliability of this information fluctuates due to external environmental factors such as market competition, policy changes, and economic conditions, leading to a degree of ambiguity in the company's perception of potential customers. At the same time, the complex market environment can significantly influence the personal preferences of corporate decision-makers. Different decision-makers, based on their own experience, risk perception, and business philosophy, may have different focuses and judgment criteria when assessing potential customers. This necessitates that the assessment process balance the uncertainty of objective information and the subjective preferences of decision-makers to ensure that the assessment results better reflect the company's actual business needs.

[0003] Currently, various methods exist for assessing potential customers in the telecommunications industry, including the analytic hierarchy process (AHP), fuzzy comprehensive evaluation, machine learning algorithms, and regression analysis. However, these methods have some shortcomings. Many methods primarily analyze static customer information, making it difficult to adapt to dynamic changes in the market environment and customer status, and failing to capture fluctuations in customer potential in a timely manner. To address this issue, some studies have proposed dynamic assessment methods incorporating time series data, analyzing customer information from multiple time points. However, in most cases, these methods only yield a value ranking of potential customers, requiring telecommunications companies to make further judgments based on the ranking. Moreover, decision-makers often tend to make binary decisions, namely, whether to cooperate with the customer. For potential customers in the medium-value range, more information is often needed for in-depth assessment, making the assessment more challenging. Furthermore, the complex external market environment significantly influences decision-makers' personal preferences and risk attitudes. Previous assessments often failed to fully consider the impact of decision-makers' subjective preferences and risk attitudes on the assessment results, leading to a discrepancy between the assessment results and the company's actual decision-making needs.

[0004] Therefore, in order to solve the above problems, this paper proposes a method for evaluating potential telecommunications customers based on three-way decision-making using probabilistic language. Summary of the Invention

[0005] Addressing the core pain points in assessing potential customers for telecommunications companies, this study proposes a novel assessment method based on a three-way decision-making framework using probabilistic language. The study integrates a probabilistic language model into the assessment system to accurately capture decision-makers' preferences, derives specific implementation measures based on the three-way decision-making theory, and sets differentiated reference points according to different target information, clearly presenting decision preferences and risk attitudes under various scenarios. This method effectively solves the problem of traditional assessments neglecting the differences in decision-makers' preferences and solution attributes, providing a scientific basis for telecommunications companies to accurately identify potential customers.

[0006] To achieve the above-mentioned technical effects, the present invention is implemented through the following technical solution: a method for evaluating potential telecommunications customers based on probabilistic language three-branch decision-making, characterized by comprising the following steps:

[0007] S1. Preprocess the target information to obtain the probabilistic language decision matrix, and construct the state set and action set of the target in decision theory;

[0008] The state set of a target indicates whether a target belongs to the potential customer set X, and is expressed as follows:

[0009]

[0010] The action set of a target represents the actions of taking action, conducting further research, and not taking action. The probability of each target belonging to set X is calculated based on the decision matrix, as shown in the following expression:

[0011] ac * (*=P,B,N)

[0012] In the formula, ac P ac B and ac N It represents the three actions of acceptance, delay, and rejection;

[0013] S2. Based on the characteristics of the probabilistic language terminology set, construct a probabilistic language scoring function based on expectation-reliability, as shown in the following expression:

[0014] SC(L ij (p))=E(L ij (p)) ω ·RD(L ij (p)) 1-ω ;

[0015] L ij (p) represents scheme A i In attribute C j The probabilistic language terminology evaluation set below; ω is E(L ij (p)) and RD(L ijThe tradeoff coefficient between (p)) ranges from 0 to 1; E(L) ij (p)) represents the expected value of the probabilistic language terminology set, RD(L ij (p)) represents the reliability of the probabilistic language terminology set;

[0016] S3. Calculate the score matrix corresponding to the probabilistic language decision matrix based on the probabilistic language matrix. Its expression is as follows:

[0017] SC = [SC(L ij (p))] n×m

[0018] In the formula, SC(L) ij (p) represents scheme A i (i = 1, 2, ..., n) in attribute C j Score values ​​for (j = 1, 2, ..., m).

[0019] S4. Calculate the subjective and objective weights of the attributes using the analytic hierarchy process (AHP) and the coefficient of variation method. Weight the subjective and objective weights and the weights to be solved onto the score matrix to obtain the weighted score vector of each scheme. Establish an optimization model that maximizes the cosine similarity between the weighted score vector of the scheme corresponding to the attribute weights to be solved and the score vectors corresponding to the subjective and objective weights. Determine the final weights of the attributes by solving the model.

[0020] S5. Construct the relative loss function of the schemes based on the score matrix and attribute weights, and calculate the threshold x for each scheme. i and y i ;

[0021] S6. Under each attribute, calculate the comprehensive prospect value of the scheme based on the score matrix. During the calculation process, the reference point is selected as the maximum and minimum value of the score function under each attribute. Using the ELECTRE idea, calculate the (non)harmony index and net credibility index of the scheme based on the comprehensive prospect value. Thus, a conditional probability calculation method based on prospect value-credibility index is proposed.

[0022] S7. Combine the conditional probability and the threshold x i and y i The comparisons are made; based on this, a series of decision criteria are proposed to classify the objectives; the schemes are ranked within each class according to the net credibility index to obtain the final decision result.

[0023] Furthermore, in S1, the target information is preprocessed to obtain the probabilistic linguistic decision matrix, including the following steps:

[0024] S1.1 Construct the original information matrix, expressed as follows:

[0025] H = [H] ij] n×m

[0026] In the formula, H ij For scheme A i (i = 1, 2, ..., n) in attribute C j Original evaluation information for (j = 1, 2, ..., m);

[0027] Construct the decision matrix in the probabilistic language, as shown in the following expression:

[0028] F = [L] ij (p)] n×m

[0029] In the two formulas above, 1≤i≤n, where n represents the number of schemes; 1≤j≤m, where m represents the number of attributes;

[0030] S1.2 Given a set of language terms, divide the data range corresponding to each attribute into five equal-length intervals s. -2 ,s -1 The expression is as follows: s0, s1, s2;

[0031] S={s -2 ,s -1 ,s0,s1,s2}

[0032] Each interval corresponds to a linguistic term in S, and the ratio of the number of data points in each interval to the total number of data points for the corresponding attribute of the given alternative is considered as the probability of that linguistic term; thus, a probabilistic linguistic decision matrix is ​​formed.

[0033] For example, in the mentioned telecommunications customer dataset, the minimum and maximum values ​​for attribute C4 (usage frequency) are 0 and 255, respectively. Therefore, the data range for attribute C4 is from 0 to 255. This range is divided into five equal-length intervals: [0,51], (51,102], (102,153], (153,204], and (204,255). These five intervals correspond to the language terms s. -2 ,s -1 ,s0,s1 and s2.

[0034] In this dataset, there are 123 middle school users belonging to alternative plan A1. Among them, the data of 17 middle school users are located in the interval [0, 51] (corresponding to s). -2 The data for 79 middle school users are located in the interval (51, 102] (corresponding to s). -1 The data for 27 middle school users are located in the interval (102, 153] (corresponding to s0). Therefore, the language terms involved in the attribute value of alternative solution A1 on attribute C4 are s -2 s -1 and s0, s-2 The probability is 17 / 123 = 0.138. Similarly, s is calculated. -1 The probabilities of s0 and s0 are 79 / 123 (0.642) and 27 / 123 (0.22), respectively. Therefore, we obtain the evaluation value L of scheme A1 under attribute C4. 14 (p)={s -2 (0.138),s -1 (0.642),s0(0.22)}. Similarly, the evaluation values ​​of other solutions under the corresponding attributes can be obtained.

[0035] Furthermore, in S2, based on the characteristics of the probabilistic language term set, a probabilistic language scoring function based on expectation-reliability is constructed, including the following steps:

[0036] The probabilistic language terminology set is expressed as follows:

[0037]

[0038] In the formula, Let s be the l-th linguistic item in the probabilistic linguistic dataset L(p). t ;p l For the l-th language item s t The corresponding probability; #L(p) is the probability language dataset L(p) containing language terms s t The number of;

[0039] S2.1 Construct a probabilistic language expectation function to calculate the expected value of the i-th target under the j-th attribute. Its expression is as follows:

[0040]

[0041] In the formula, L ij (p) corresponds to scheme A i (i = 1, 2, ..., n) in attribute C j The probabilistic linguistic terminology evaluation value under (j=1,2,…,m); #L ij (p) represents scheme A i (i = 1, 2, ..., n) in attribute C j The number of linguistic terms contained in the probabilistic linguistic term evaluation set under (j=1,2,...,m); Represents the l-th language item s t Membership degree; For scheme A i In attribute C j The probabilistic language evaluation value of the l-th language item s t The corresponding probability;

[0042] in, The expression is as follows:

[0043]

[0044] In the formula, t represents the language term s. t The corresponding subscript; τ is the maximum value of the language term set; γ is the value of s. t The corresponding membership degree;

[0045] S2.2 Construct a probabilistic language reliability function to calculate the reliability of the i-th target under the j-th attribute, where the reliability includes hesitancy hd, clarity dd, and completeness dd;

[0046] The reliability function expression in probabilistic language is as follows:

[0047]

[0048] In the formula, hd, dd, and dd represent the probabilistic language terminology set L, respectively. ij (p) Corresponding hesitation, clarity, and completeness;

[0049] The hesitancy function hd in probabilistic languages ​​is expressed as follows:

[0050]

[0051] In the formula, γ l This represents the membership degree corresponding to the l-th language item; For the probabilistic language terminology set L ij (p) The average membership degree corresponding to the language terms included; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of;

[0052] Among them, the average membership degree The expression is as follows:

[0053]

[0054] In the formula, γ l This represents the membership degree corresponding to the l-th language item; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of;

[0055] The probabilistic language clarity function is expressed as follows:

[0056]

[0057] For scheme Ai (i = 1, 2, ..., n) in attribute C j The probability corresponding to the l-th evaluation term under (j=1,2,...,m); For the probabilistic language terminology set L ij (p) The average probability of the language terms included; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of;

[0058] Among them, average probability The expression is as follows:

[0059]

[0060] In the formula, p l This represents the probability corresponding to the l-th language item; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of;

[0061] The concentration function in probabilistic languages ​​is expressed as follows:

[0062]

[0063] In the formula, For scheme A i (i = 1, 2, ..., n) in attribute C j The probability corresponding to the l-th evaluation term in (j=1,2,...,m); #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of;

[0064] S2.3. Construct the expected-reliability score function based on the functions in S2.1-S2.2, with the following expression: E(L ij (p)) and RD(L ij (p) represents the expected value function and reliability function of the probabilistic language term set, respectively. The coefficient ω links the two to obtain the expected-reliability-based score function SC(L). ij (p)), this formula can be seen as an extension of the expected value function:

[0065] SC(L ij (p))=E(L ij (p)) ω ·RD(L ij (p)) 1-ω

[0066] In the formula, ω is E(L) ij (p)) and RD(L ij The tradeoff coefficient between (p) and E(L) ranges from 0 to 1; ij (p)) represents the expected value of the probabilistic language terminology set, RD(L ij (p)) represents the reliability of the probabilistic language terminology set;

[0067] Furthermore, in S4, subjective weights are calculated using the analytic hierarchy process (AHP), objective weights are calculated using the coefficient of variation method, and the overall weights are calculated by maximizing the objective function, as detailed below:

[0068] S4.1 Calculate the subjective weights using the Analytic Hierarchy Process (AHP), as shown in the following expression:

[0069] W sj (j=1,2,…,m)

[0070] In the formula, W sj This represents the subjective weight of the j-th attribute; there are a total of m attributes.

[0071] S4.2 Calculate the subjective weighted score vector, as shown in the following expression:

[0072]

[0073] In the formula, W sm SC(L) represents the subjective weight of the m-th attribute; im (p) represents scheme A i In attribute C m The following is based on the expected-reliability score; Scheme A i In attribute C m The score is based on expectation-reliability after subjective weighting.

[0074] S4.3 Calculate the objective weights using the coefficient of variation method, as shown in the following expression:

[0075] First, calculate the mean of the expected-reliability score function for each attribute, as shown in the following expression:

[0076]

[0077] In the formula, SC ij Scheme A i (i = 1, 2, ..., n) in attribute C j The following is based on the expected-reliability score; Indicates attribute C j The mean is based on the expected-reliability score function;

[0078] Then, the standard deviation of the score for each attribute is calculated, as shown in the following expression:

[0079]

[0080] In the formula, SC ij Scheme A i (i = 1, 2, ..., n) in attribute C j The following is based on the expected-reliability score; Indicates attribute C j The mean value is based on the expected-reliability score function; n represents the number of alternatives; St j Indicates attribute C j Based on the standard deviation of the expected-reliability score function;

[0081] Attribute C was obtained j After calculating the mean and standard deviation of the expected-reliability score function, the C of the attribute is then calculated. j The coefficient of variation is expressed as follows:

[0082]

[0083] Finally, assign each attribute C j By standardizing the coefficient of variation of (j=1,2,...,m), we can obtain attribute C. j The objective weight is expressed as follows:

[0084]

[0085] In the formula, Va j Indicates attribute C j coefficient of variation; W oj This represents the objective weight of the j-th attribute, and there are a total of m attributes;

[0086] S4.4 Calculate the objective weighted score vector, as shown in the following expression:

[0087]

[0088] In the formula, W om SC(L) represents the objective weight of the m-th attribute; im (p) represents scheme A i In attribute C m The following is based on the expected-reliability score; Scheme A i In attribute C m The score is based on expectation-reliability after being weighted by objective weights.

[0089] S4.5 Finally, calculate the overall weight, as shown in the following expression:

[0090] W j (j=1,2,…,m)

[0091] In the formula, W j This represents the weight of the j-th attribute, and there are m attributes in total;

[0092] The comprehensive weighted score vector is calculated as follows:

[0093]

[0094] In the formula, W m SC(L) represents the weight of the m-th attribute; im (p) represents scheme A i In attribute C m The following is based on the expected-reliability score; Scheme A i In attribute C m The score is calculated based on expected reliability after weighting.

[0095] S4.6 Calculate the cosine similarity between the comprehensive weighted score vector, the subjective weighted score vector, and the objective weighted score vector, respectively. The expression is as follows:

[0096]

[0097] In the formula, Scheme A i In attribute C j The expected-reliability score after subjective weighting for (j=1,2,...,m); Scheme A i In attribute C j The expected-reliability score after objective weighting for (j=1,2,...,m); Scheme A i In attribute C j The expected-reliability score after weighting for (j=1,2,...,m);

[0098] The overall weight is calculated by maximizing the objective function, as shown in the following expression:

[0099]

[0100] In the formula, Scheme A i (i = 1, 2, ..., n) with weights W to be solved j (j=1,2,...,m) Weighted expectation-reliability-based scoring function And subject to subjective weighting W oj (j=1,2,...,m) Weighted expectation-reliability-based scoring function Cosine similarity between them; Scheme A i (i = 1, 2, ..., n) with weights W to be solved j (j=1,2,…,m) Weighted score function based on expectation-reliability And by objective weight W sj (j=1,2,…,m) Weighted score function based on expectation-reliability Cosine similarity between

[0101] Furthermore, in S5, the relative loss function of the scheme is constructed based on the score matrix and attribute weights, and the threshold x of each scheme is calculated. i and y i The specific steps are as follows:

[0102] S5.1, Calculation scheme in state X and Take action ac P ,ac B and ac N The aggregate relative loss function is expressed as follows:

[0103]

[0104] In the formula, SC(L) ij (p) represents scheme A i In attribute C j The score below; and Representing attribute C respectively j The minimum and maximum values ​​of the scoring function are given; η is the risk aversion coefficient, ranging from 0 to 0.5. and Representing scheme A respectively i Take action ac in state X P ,ac B and ac N The aggregate relative loss function value; similarly, and Representing scheme A respectively i In state Take action ac P ,ac B and ac N The aggregate relative loss function value. Based on the aggregate relative loss function, the threshold x for each scheme is calculated. i and y i The expression is as follows;

[0105]

[0106] In the formula, and Representing scheme A respectively i Take action ac in state X P ,ac B and ac N The aggregate relative loss function value; similarly, and Representing scheme A respectively i In state Take action ac P ,ac B and ac N The aggregate relative loss function value; η is the risk aversion coefficient, which ranges from 0 to 0.5.

[0107] Furthermore, in S6, based on the score matrix, the comprehensive prospect value of the scheme is calculated. During the calculation process, the maximum and minimum values ​​of the score function under each attribute are selected as reference points. Using the ELECTRE concept, the (non-)harmony index and net confidence index of the scheme are calculated based on the comprehensive prospect value; including the following steps:

[0108] S6.1 Calculate the comprehensive foreground value based on the scoring function matrix, as shown in the following expression:

[0109] Calculate the positive foreground value of the scheme:

[0110]

[0111] Calculate the negative foreground value of the scheme:

[0112]

[0113] Calculate the overall prospect value of the scheme:

[0114] V(Γ(L ij (p)))=δV(Γ + (L ij (p)))+(1-δ)V(Γ - (L ij (p)));

[0115] In the above formula, and Representing attribute C respectively j Minimum and maximum values ​​of the score function under Γ; + (L ij (p)) and Γ - (L ij (p)) represent the loss and gain values ​​relative to the reference point, respectively; V(Γ)+ (L ij (p))) and V(Γ) - (L ij (p))) represent positive and negative prospect values ​​respectively; α and β represent risk attitude coefficients, satisfying 0 < α and β < 1; λ represents loss avoidance factor, satisfying λ > 1; δ represents balance coefficient, satisfying 0 ≤ δ ≤ 1;

[0116] S6.2. Using the ELECTRE concept, the (non)harmony index and net confidence index of the comprehensive prospect value calculation scheme are expressed as follows:

[0117] Harmony index of the calculation scheme:

[0118]

[0119] The non-harmonicity index of the calculation scheme:

[0120]

[0121] Credibility index of the calculation scheme:

[0122]

[0123] Consistency and reliability metrics for the calculation scheme:

[0124]

[0125] Inconsistency reliability index of the calculation scheme:

[0126]

[0127] Net confidence index of the calculation scheme:

[0128] ψ(A i )=ψ + (A i )-ψ - (A i );

[0129] In the above formula, q j p j and v j These are the difference threshold, preference threshold, and rejection threshold, respectively, and 0 < q. j <p j <v j <1; W j Indicates attribute weight; O j (A i A k ) indicates that in attribute C j Option A i For option A kConsistency index; D j (A i A k ) indicates that in attribute C j Option A i For option A k Inconsistency index; O(A) i A k ) represents scheme A i For option A k The overall consistency index; the credibility index Z(A) i A k (This indicates case A) i For option A k Priority, J(A) i A k ) is all D j (A i A k )≤O(A i A k A collection of attributes; ψ + (A i ) represents scheme A i Consistency credibility index; ψ - (A i ) represents scheme A i Inconsistent credibility index; ψ(A i ) represents scheme A i Net credibility index.

[0130] S6.3. Establish a goal programming model and solve for the prospect theoretical parameters α, β, and λ; then the net confidence level of the solution can be obtained; the expression is as follows:

[0131]

[0132] In the formula, ψ(A) i ) represents scheme A i The net credibility index is α and β, which represent the decision-maker's sensitivity to gains and losses, respectively; λ is the loss aversion coefficient.

[0133] S6.4. Solve for the conditional probability based on the idea of ​​relative proximity. The expression is as follows:

[0134]

[0135] In the formula, ψ(A) i ) represents scheme A i Net credibility; maxψ(A) i ) and minψ(A i ) represent the maximum and minimum values ​​of the net harmony index for all schemes, respectively.

[0136] Furthermore, in S7, the conditional probability is related to the threshold x. i and y i The process involves comparing and classifying the objectives, ranking the solutions within each category based on the net credibility index, and obtaining the final decision result; this includes the following steps:

[0137] S7.1. Compare the conditional probability with the threshold to determine the classification result of the scheme, as shown in the following expression:

[0138] (P0) If Pr(X|[A i ])≥x i Then A i ∈POS(X),

[0139] (B0)y i <Pr(X|[A i ])<x i Then A i ∈BND(X),

[0140] (N0) If Pr(X|[A i ])≤y i Then A i ∈NEG(X).

[0141] In the formula, Pr(X|[A i ]) represents scheme A i The conditional probability of belonging to state X; POS(X), BND(X), and NEG(X) represent the acceptance region, boundary region, and rejection region, respectively.

[0142] S7.2 Within each region, sort the proposals according to their net confidence level.

[0143] Furthermore, S7 makes decisions based on the following rule: If a solution A... i The conditional probability is greater than or equal to its corresponding threshold x i Perform the P decision; if the value of the conditional probability is in x i to y i Between these points, execute decision B; if the conditional probability is less than or equal to y. i Then, an N-decision is performed on the proposed solutions. Solutions in the POS(X) domain are ranked higher than those in the BND(X) domain, and similarly, solutions in the BND(X) domain are ranked higher than those in the NEG(X) domain. Within each domain, the solutions are then ranked according to their net confidence level.

[0144] Furthermore, it also includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor; when the computer program is loaded onto the processor, it implements the aforementioned probabilistic language-based three-way decision-making method for evaluating potential telecommunications customers.

[0145] The beneficial effects of this invention are:

[0146] This invention considers the uncertainty of information when assessing potential telecommunications customers, and makes three-way decisions for the target, making the classification results more reasonable and helping enterprises make quick decisions. At the same time, prospect theory and ELECTRE are introduced into the decision-making process. Different customer states receive different target information, thus obtaining different reference points, which can reflect different decision preferences and risk attitudes when facing different customers, making the decision results more in line with real-world scenarios. Attached Figure Description

[0147] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0148] Figure 1 This is a flowchart of the telecommunications potential customer evaluation method based on probabilistic language three-way decision-making as described in this invention. Detailed Implementation

[0149] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0150] Example 1

[0151] A method for evaluating potential telecommunications customers based on a probabilistic three-way decision-making system includes the following steps:

[0152] Step 1: Preprocess the target information to obtain the probabilistic language decision matrix, and construct the state set and action set in decision theory. The ac represents a set of states, indicating whether a target belongs to the set of potential customers X; * (*=P,B,N) represents the action set of a target, which respectively represent the actions of taking action, conducting further research, and not taking action for a target;

[0153] When handling business, telecommunications companies collect customer information, and the backend records a large amount of complex customer data. Processing this large volume of information is time-consuming and often meaningless. Therefore, in the potential customer assessment process, five indicators are selected for evaluation: customer age, subscription duration, payment amount, number of calls, and usage frequency. This example aims to distinguish which age groups are potential customers; therefore, A1, A2, A3, A4, and A5 are respectively assigned to middle school students, university students, young users, middle-aged users, and elderly users.

[0154] The original information matrix of the target is represented as H = [H ij ] n×m The decision matrix in probabilistic language is represented as F = [L]. ij (p)] n×m 1≤i≤n, where n represents the number of solutions, 1≤j≤m, and m represents the number of attributes; the decision matrix of this probabilistic language is shown in Table 1:

[0155] First, calculate the difference between the maximum and minimum values ​​of the data for each attribute. Divide this difference into several intervals, each interval corresponding to a different language term. Then, count the number of data points contained in each interval under each scheme. Divide the number of data points contained in each interval by the total number of data points under that scheme to obtain the probability corresponding to that interval. In the mentioned telecommunications customer dataset, the minimum and maximum values ​​of the frequency of use of attribute C4 are 0 and 255, respectively. Therefore, the data range for attribute C4 is from 0 to 255; this range is divided into five equal-length intervals: [0,51], (51,102], (102,153], (153,204], and (204,255). These five intervals correspond to the language terms s. -2 s -1 s0, s1, and s2. In this dataset, there are 123 middle school users belonging to alternative scheme A1. Among them, the data of 17 middle school users are located in the interval [0, 51] (corresponding to s0, s1, and s2). -2 The data for 79 middle school users are located in the interval (51, 102] (corresponding to s). -1 The data for 27 middle school users are located in the interval (102, 153] (corresponding to s0). Therefore, the language terms involved in the attribute value of alternative solution A1 on attribute C4 are s -2 s -1 and s0, s -2 The probability is 17 / 123 = 0.138. Similarly, s is calculated. -1 The probabilities of s0 and s0 are 79 / 123 (0.642) and 27 / 123 (0.22), respectively. Therefore, we obtain the evaluation value L of scheme A1 under attribute C4. 14 (p)={s -2 (0.138),s -1(0.642),s0(0.22)}. Similarly, the evaluation values ​​of other solutions under the corresponding attributes can be obtained.

[0156] Table 1 Probabilistic Language Decision Matrix

[0157]

[0158]

[0159] Step 2: Calculate the expectation-reliability score based on the probability language matrix:

[0160] The steps for calculating the expectation-reliability score based on the probability language matrix are as follows:

[0161] (a) Calculate the expected value of the i-th target under the j-th attribute:

[0162]

[0163] (b) Calculate the reliability of the i-th target under the j-th attribute:

[0164]

[0165] The formulas for calculating the hesitancy (hd), clarity (dd), and completeness (cd) included in reliability are as follows:

[0166]

[0167] (c) Calculate the score function based on expectation-reliability:

[0168] SC(L ij (p))=E(L ij (p)) ω ·RD(L ij (p)) 1-ω ;

[0169] in, τ is the maximum value of the language term set; γ l To represent the membership degree corresponding to the l-th language item, p l Let l be the probability corresponding to the l-th language item. #L ij (p) Scheme A i In attribute C j The linguistic terms s contained in the lower probability linguistic term evaluation set t The number of; L ij (p) represents scheme A i In attribute C j The probabilistic language terminology evaluation set below; ω is E(Lij (p)) and RD(L ij The tradeoff coefficient between (p)) ranges from 0 to 1; E(L) ij (p)) represents the expected value of the probabilistic language terminology set, RD(L ij (p) represents the reliability of the probabilistic language terminology set.

[0170] Step 3: Construct the corresponding score matrix based on the score values:

[0171] Table 2. Score based on expectation-reliability

[0172]

[0173] Step 4: Calculate subjective weights using the analytic hierarchy process (AHP) and objective weights using the coefficient of variation method. Weight the subjective and objective weights and the weights to be solved onto the scoring function matrix to obtain the weighted scoring function vector for each scheme. Maximize the cosine similarity of the weighted scoring function vectors to solve for the attribute weights.

[0174] First, calculate the subjective weights using the Analytic Hierarchy Process (AHP):

[0175] W sj (j = 1, 2, ..., m);

[0176] Calculate the subjective weighted score vector:

[0177]

[0178] Calculate the objective weights using the coefficient of variation method:

[0179] First, calculate the mean of the expected-reliability score function for each attribute, as shown in the following expression:

[0180]

[0181] Then, the standard deviation of the score for each attribute is calculated, as shown in the following expression:

[0182]

[0183] Attribute C was obtained j After calculating the mean and standard deviation of the expected-reliability score function, the C of the attribute is then calculated. j The coefficient of variation is expressed as follows:

[0184]

[0185] Finally, assign each attribute C j By standardizing the coefficient of variation of (j=1,2,...,m), we can obtain attribute C. jThe objective weight is expressed as follows:

[0186]

[0187] Calculate the objective weighted score vector:

[0188]

[0189] Finally, calculate the overall weight:

[0190] W j (j = 1, 2, ..., m);

[0191] Calculate the overall weighted score vector:

[0192]

[0193] Calculate the similarity between the comprehensive weighted score vector, the subjective weighted score vector, and the objective weighted score vector, respectively:

[0194]

[0195] Calculate the overall weights by maximizing the objective function:

[0196]

[0197] Among them, SC(L ij (p)) represents the score of the i-th scheme under the j-th attribute in the expected-reliability scoring function matrix.

[0198] Table 3 Attribute Weights

[0199]

[0200] Step 5: Calculate the aggregated relative loss function based on the score function matrix and the comprehensive weights, and then calculate the threshold for each scheme:

[0201] First, calculate the solution in state X and... Take action ac P ,ac B and ac N The aggregate relative loss function is shown in the table below:

[0202] Table 4. Aggregate Relative Loss Function

[0203]

[0204] Secondly, based on the aggregate relative loss function, the threshold x for each scheme is calculated. i and y i :

[0205]

[0206] in, and Representing attribute C respectively j The minimum and maximum values ​​of the score function under the given conditions; and Representing scheme A respectively i Take action ac in state X P ,ac B and ac N The aggregate relative loss function value; similarly, and Representing scheme A respectively i In state Take action ac P ,ac B and ac N The aggregate relative loss function value; η represents the risk aversion coefficient, which ranges from 0 to 0.5.

[0207] Table 5 x, y thresholds

[0208]

[0209]

[0210] Step 6: Calculate the comprehensive prospect value based on the score function matrix, where the reference point is selected as the maximum and minimum score function values ​​for each attribute; substitute the comprehensive prospect value into the ELECTRE method to calculate the net confidence of the scheme; calculate the conditional probability based on the idea of ​​relative proximity.

[0211] Step 1. Calculate the positive foreground value of the proposed solution:

[0212]

[0213] Step 2. Calculate the negative foreground value of the scheme:

[0214]

[0215] Step 3. Calculate the overall prospect value of the proposed solution:

[0216] V(Γ(L ij (p)))=δV(Γ + (L ij (p)))+(1-δ)V(Γ - (L ij (p)));

[0217] Table 6 Overall Prospect Value

[0218]

[0219]

[0220] Step 4: Calculate the scheme in attribute C j The following harmony indicators:

[0221] Step 5: Calculate the scheme in attribute C j The following are the indicators of disharmony:

[0222] Step 6: Calculate the harmony index of the scheme:

[0223]

[0224] Step 7: Calculate the harmony index of the scheme:

[0225]

[0226] Step 8: Calculate the disharmony index of the scheme:

[0227]

[0228] Step 9: Calculate the net confidence level of the scheme.

[0229] ψ(A i )=ψ + (A i )-ψ - (A i );

[0230] in, and Representing attribute C respectively j Minimum and maximum values ​​of the score function under Γ; + (L ij (p)) and Γ - (L ij (p)) represent the loss and gain values ​​relative to the reference point, respectively; V(Γ) + (L ij (p))) and V(Γ) - (L ij (p))) represent positive and negative prospect values ​​respectively; α and β represent risk attitude coefficients, satisfying 0 < α and β < 1; λ represents the loss avoidance factor, satisfying λ > 1; δ represents the balance coefficient, satisfying 0 ≤ δ ≤ 1; q j p j and v j These are the difference threshold, preference threshold, and rejection threshold, respectively, and 0 < q. j <p j <v j <1; O j(A i A k ) indicates that in attribute C j Option A i For option A k Consistency index; D j (A i A k ) indicates that in attribute C j Option A i For option A k Inconsistency index; O(A) i A k ) represents scheme A i For option A k The overall consistency index; the credibility index Z(A) i A k (This indicates case A) i For option A k Priority, J(A) i A k ) is all D j (A i A k )≤O(A i A k A collection of attributes; ψ + (A i ), ψ - (A i ) and ψ(A i ) represent scheme A respectively i Net confidence level for consistency, confidence level for inconsistency, and net confidence level.

[0231] Step 10: Establish a goal programming model and solve for the prospect theory parameters:

[0232]

[0233] Table 7 Prospect Theory Parameters

[0234]

[0235] The net confidence level of the scheme can be obtained from the solved α, β and λ.

[0236] Table 8 Reliability

[0237]

[0238]

[0239] Step 11: Solve for the conditional probability based on the concept of relative proximity:

[0240]

[0241] In the formula, maxψ(A) i ) and minψ(A i ) represent the maximum and minimum net confidence levels for each scheme, respectively.

[0242] Table 9 Conditional Probabilities

[0243]

[0244] Step 7: Compare the conditional probabilities and thresholds to determine the classification results of the schemes, and rank the schemes in each domain according to the net confidence index:

[0245] The first step is to compare the conditional probability with the threshold to determine the classification result of the scheme:

[0246] (P0) If Pr(X|[A i ])≥x i Then A i ∈POS(X),

[0247] (B0)y i <Pr(X|[A i ])<x i Then A i ∈BND(X),

[0248] (N0) If Pr(X|[A i ])≤y i Then A i ∈NEG(X).

[0249] Based on the above results, the final three-branch segmentation is obtained. In this embodiment, the positive region represents potential customers. The three-branch decision model yields the segmentation results for all customers. Based on these segmentation results, different decisions can be made. Specifically, A5 is a potential customer, while A1, A2, A3, and A4 are not potential customers.

[0250] The second step is to rank the proposals within each region according to their net confidence level.

[0251] In the potential customer group, A5 ranked first, while in the non-potential customer group, A3 ranked first, followed by A1 and A2, and A4 ranked last.

[0252] Based on the same inventive concept, the present invention provides a telecommunications potential customer assessment method based on probabilistic language three-branch decision-making, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the aforementioned three-branch multi-attribute potential customer assessment method based on prospect theory and ELECTRE.

Claims

1. A method for evaluating potential telecommunications customers based on probabilistic three-way decision-making, characterized in that, Includes the following steps: S1. Preprocess the target information to obtain the probabilistic language decision matrix, and construct the state set and action set of the target in decision theory; The state set of a target indicates whether a target belongs to the potential customer set X, and is expressed as follows: The action set of a target represents the actions of taking action, conducting further research, and not taking action. The probability of each target belonging to set X is calculated based on the decision matrix, as shown in the following expression: and * (*=P,B,N) In the formula, ac P ac B and ac N These represent the actions of acceptance, delay, and rejection, respectively. S2. Based on the characteristics of the probabilistic language terminology set, construct a probabilistic language scoring function based on expectation-reliability, as shown in the following expression: SC(L ij (p))=E(L ij (p)) ω ·RD(L ij (p)) 1-ω ; In the formula, L ij (p) represents scheme A i In attribute C j The probabilistic linguistic evaluation value is given below, where p is the probability of the corresponding linguistic item s; ω is the probability of E(L) linguistic evaluation. ij (p)) and RD(L ij The tradeoff coefficient between (p) and E(L) ranges from 0 to 1; ij (p)) represents the expected value of the probabilistic language terminology set, RD(L ij (p)) represents the reliability of the probabilistic language terminology set; S3. Calculate the score matrix corresponding to the probabilistic language decision matrix based on the probabilistic language matrix. Its expression is as follows: SC=[SC(L ij (p))] n×m In the formula, L ij (p) corresponds to scheme A i (i = 1, 2, ..., n) in attribute C j Probabilistic linguistic term evaluation values ​​under (j=1,2,...,m); SC(L ij (p)) is the expected-reliability score corresponding to this evaluation value; S4. Calculate the subjective and objective weights of the attributes using the analytic hierarchy process (AHP) and the coefficient of variation method. Weight the subjective and objective weights and the weights to be solved onto the score matrix to obtain the weighted score vector of each scheme. Establish an optimization model that maximizes the cosine similarity between the weighted score vector of the scheme corresponding to the attribute weights to be solved and the score vectors corresponding to the subjective and objective weights. Determine the final weights of the attributes by solving the model. S5. Construct the relative loss function of the schemes based on the score matrix and attribute weights, and calculate the threshold x for each scheme. i and y i ; S6. Under each attribute, calculate the comprehensive prospect value of the scheme based on the score matrix. During the calculation process, the reference points are selected as the maximum and minimum values ​​of the score function under each attribute. Based on the ELECTRE concept, and using the (non)harmony index and net confidence index of the comprehensive prospect value calculation scheme, a conditional probability calculation method based on the prospect value-harmony index is proposed. S7. Combine the conditional probability and the threshold x i and y i The comparisons are made; based on this, a series of decision criteria are proposed to classify the objectives; the schemes are ranked within each class according to the net credibility index to obtain the final decision result.

2. The method for evaluating potential telecommunications customers based on probabilistic language three-branch decision-making as described in claim 1, characterized in that, In S1, the target information is preprocessed to obtain the probabilistic linguistic decision matrix, including the following steps: S1.1 Construct the original information matrix, expressed as follows: H=[H ij ] n×m In the formula, H ij For scheme A i (i = 1, 2, ..., n) in attribute C j Original evaluation information for (j = 1, 2, ..., m); Construct the decision matrix in the probabilistic language, as shown in the following expression: F=[L ij (p)] n×m In the two formulas above, 1≤i≤n, where n represents the number of schemes; 1≤j≤m, where m represents the number of attributes; S1.2 Given a set of language terms, divide the data range corresponding to each attribute into five equal-length intervals s. -2 ,s -1 The expression is as follows: s0, s1, s2; S={s -2 ,s -1 ,s0,s1,s2} Each interval corresponds to a linguistic term in S, and the ratio of the number of data points in each interval to the total number of data points for the corresponding attribute of the given alternative is considered as the probability of that linguistic term; thus, a probabilistic linguistic decision matrix is ​​formed.

3. The method for evaluating potential telecommunications customers based on probabilistic language three-branch decision-making as described in claim 1, characterized in that, In S2, based on the characteristics of the probabilistic language term set, a probabilistic language scoring function based on expectation-reliability is constructed. Includes the following steps: The probabilistic language terminology set is expressed as follows: In the formula, Let s be the l-th linguistic item in the probabilistic linguistic dataset L(p). t ;p l For the l-th language item s t The corresponding probability; #L(p) is the probability language dataset L(p) containing language terms s t The number of; S2.1 Construct a probabilistic language expectation function to calculate the expected value of the i-th target under the j-th attribute. Its expression is as follows: In the formula, L ij (p) corresponds to scheme A i (i = 1, 2, ..., n) in attribute C j Probabilistic linguistic terminology evaluation values ​​under (j = 1, 2, ..., m); #L ij (p) represents scheme A i (i = 1, 2, ..., n) in attribute C j The number of linguistic terms contained in the probabilistic linguistic term evaluation set under (j=1,2,...,m); Represents the l-th language item s t Language scaling conversion functions; For scheme A i In attribute C j The probabilistic language evaluation value of the l-th language item s t The corresponding probability; That The expression is as follows: In the formula, t represents the language term s. t The corresponding subscript; τ is the maximum value of the language term set; γ is the value of s. t The corresponding membership degree; S2.2 Construct a probabilistic language reliability function to calculate the reliability of the i-th target under the j-th attribute, where the reliability includes hesitancy hd, clarity dd, and completeness cd; Among them, Scheme A i In attribute C j The following probabilistic language terminology set is expressed in the form of: In the formula, For the probabilistic language dataset L ij The l-th language item s in (p) t ;p l For the l-th language item s t The corresponding probability; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of; The reliability function expression in probabilistic language is as follows: In the formula, hd, dd, and cd represent the probabilistic language term set L, respectively. ij (p) Corresponding hesitation, clarity, and completeness; The hesitancy function hd in probabilistic languages ​​is expressed as follows: In the formula, γ l This represents the membership degree corresponding to the l-th language item; For the probabilistic language terminology set L ij (p) The average membership degree corresponding to the language terms included; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of; Among them, the average membership degree The expression is as follows: In the formula, γ l This represents the membership degree corresponding to the l-th language item; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of; The probabilistic language clarity function is expressed as follows: In the formula, For scheme A i (i = 1, 2, ..., n) in attribute C j The probability corresponding to the l-th evaluation term under (j=1,2,...,m); For the probabilistic language terminology set L ij (p) The average probability of the language terms included; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of; Among them, average probability The expression is as follows: In the formula, p l This represents the probability corresponding to the l-th language item; #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of; The concentration function in probabilistic languages ​​is expressed as follows: In the formula, For scheme A i (i = 1, 2, ..., n) in attribute C j The probability corresponding to the l-th evaluation term in (j=1,2,...,m); #L ij (p) represents the probabilistic language terminology set L. ij (p) contains language items t The number of; S2.

3. Construct the expected-reliability score function based on the functions in S2.1-S2.2, with the following expression: where E(L ij (p)) and RD(L ij (p) represents the expected value function and reliability function of the probabilistic language term set, respectively. The coefficient ω links the two to obtain the expected-reliability-based score function SC(L). ij (p)), this formula can be seen as an extension of the expected value function; SC(L ij (p))=E(L ij (p)) ω ·RD(L ij (p)) 1-ω In the formula, L ij (p) represents scheme A i In attribute C j The probabilistic language terminology evaluation set below; ω is E(L ij (p)) and RD(L ij The tradeoff coefficient between (p)) ranges from 0 to 1; E(L) ij (p)) represents the expected value of the probabilistic language terminology set, RD(L ij (p) represents the reliability of the probabilistic language terminology set.

4. The method for evaluating potential telecommunications customers based on probabilistic language three-branch decision-making as described in claim 1, characterized in that, In S4, subjective weights are calculated using the analytic hierarchy process (AHP), objective weights are calculated using the coefficient of variation method, and the overall weights are calculated by maximizing the objective function, as detailed below: S4.1 Calculate the subjective weights using the Analytic Hierarchy Process (AHP), as shown in the following expression: W sj (j=1,2,…,m) In the formula, W sj This represents the subjective weight of the j-th attribute; there are a total of m attributes. S4.2 Calculate the subjective weighted score vector, as shown in the following expression: In the formula, W sm SC(L) represents the subjective weight of the m-th attribute; im (p) represents scheme A i In attribute C m The following is based on the expected-reliability score; Scheme A i In attribute C m The score is based on expectation-reliability after subjective weighting. S4.3 Calculate the objective weights using the coefficient of variation method, as shown in the following expression: First, calculate the mean of the expected-reliability score function for each attribute, as shown in the following expression: In the formula, SC ij Scheme A i (i = 1, 2, ..., n) in attribute C j The following is based on the expected-reliability score; Indicates attribute C j The mean is based on the expected-reliability score function; Then, the standard deviation of the score for each attribute is calculated, as shown in the following expression: In the formula, SC ij Scheme A i (i = 1, 2, ..., n) in attribute C j The following is based on the expected-reliability score; Indicates attribute C j The mean value is based on the expected-reliability score function; n represents the number of alternatives; St j Indicates attribute C j Based on the standard deviation of the expected-reliability score function; Attribute C was obtained j After calculating the mean and standard deviation of the expected-reliability score function, the C of the attribute is then calculated. j The coefficient of variation is expressed as follows: Finally, assign each attribute C j By standardizing the coefficient of variation of (j=1,2,...,m), we can obtain attribute C. j The objective weight is expressed as follows: In the formula, Va j Indicates attribute C j coefficient of variation; W oj This represents the objective weight of the j-th attribute, and there are a total of m attributes; S4.4 Calculate the objective weighted score vector, as shown in the following expression: In the formula, W om SC(L) represents the objective weight of the m-th attribute; im (p) represents scheme A i In attribute C m The following is based on the expected-reliability score; Scheme A i In attribute C m The score is based on expectation-reliability after being weighted by objective weights. S4.5 Finally, calculate the overall weight, as shown in the following expression: W j (j=1,2,…,m) In the formula, W j This represents the final weight of the j-th attribute, and there are a total of m attributes; The comprehensive weighted score vector is calculated as follows: In the formula, W m SC(L) represents the final weight of the m-th attribute; im (p) represents scheme A i In attribute C m The following is based on the expected-reliability score; Scheme A i In attribute C m The score is calculated based on expected reliability after weighting. S4.6 Calculate the cosine similarity between the comprehensive weighted score vector and the subjective (objective) weighted score vector, respectively, as shown in the following expression: In the formula, Scheme A i In attribute C j The expected-reliability score after subjective weighting for (j=1,2,...,m); Scheme A i In attribute C j The expected-reliability score after objective weighting for (j=1,2,...,m); Scheme A i In attribute C j The expected-reliability score after weighting for (j=1,2,...,m); The overall weight is calculated by maximizing the objective function, as shown in the following expression: In the formula, Scheme A i (i = 1, 2, ..., n) with weights W to be solved j (j=1,2,...,m) Weighted expectation-reliability-based scoring function And subject to subjective weighting W oj (j=1,2,…,m) Weighted score function based on expectation-reliability Cosine similarity between them; Scheme A i (i = 1, 2, ..., n) with weights W to be solved j (j=1,2,…,m) Weighted score function based on expectation-reliability And by objective weight W sj (j=1,2,…,m) Weighted score function based on expectation-reliability Cosine similarity between them.

5. The method for evaluating potential telecommunications customers based on probabilistic language three-branch decision-making as described in claim 1, characterized in that, In S5, the relative loss function of the scheme is constructed based on the score matrix and attribute weights, and the threshold x of each scheme is calculated. i and y i The specific steps are as follows: S5.1, Calculation scheme in state X and Take action ac P ,ac B and ac N The aggregate relative loss function is expressed as follows: and P : and B : and N : In the formula, SC(L) ij (p) represents scheme A i In attribute C j The score below; and Representing attribute C respectively j The minimum and maximum values ​​of the score function under the given conditions; η∈[0,0.5] is the risk aversion coefficient; and Representing scheme A respectively i Take action ac in state X P ,ac B and ac N The aggregation relative loss function value; similarly, and Representing scheme A respectively i In state Take action ac P ,ac B and ac N The aggregate relative loss function value; Based on the aggregation relative loss function, calculate the threshold x for each scheme. i and y i The expression is as follows; In the formula, and Representing scheme A respectively i Take action ac in state X P ,ac B and ac N The aggregate relative loss function value; similarly, and Representing scheme A respectively i In state Take action ac P ,ac B and ac N The aggregate relative loss function value.

6. The method for evaluating potential telecommunications customers based on probabilistic language three-branch decision-making as described in claim 1, characterized in that, In S6, the conditional probability calculation method based on the foreground value-confidence index includes the following steps: S6.1 Calculate the comprehensive foreground value based on the scoring function matrix, as shown in the following expression: Calculate the positive foreground value of the scheme: Calculate the negative foreground value of the scheme: Calculate the overall prospect value of the scheme: V(Γ(L ij (p)))=δV(Γ + (L ij (p)))+(1-δ)V(Γ - (L ij (p))); In the above formula, and Representing attribute C respectively j Minimum and maximum values ​​of the score function under Γ; + (L ij (p)) and Γ - (L ij (p)) represent the loss and gain values ​​relative to the reference point, respectively; V(Γ) + (L ij (p))) and V(Γ) - (L ij (p))) represent positive and negative prospect values ​​respectively; α and β represent risk attitude coefficients, satisfying 0 < α and β < 1; λ represents loss avoidance factor, satisfying λ > 1; δ represents balance coefficient, satisfying 0 ≤ δ ≤ 1; S6.

2. Using the ELECTRE concept, the (non)harmony index and net confidence index of the comprehensive prospect value calculation scheme are expressed as follows: The calculation scheme is in attribute C j Harmony indicators: The calculation scheme is in attribute C j Disharmony indicators: Calculate the net harmony index of the scheme: The harmonious dominance value index of the calculation scheme: Calculate the non-uniform dominance index of the scheme: Net confidence index of the calculation scheme: ψ(A i )=ψ + (A i )-ψ - (A i ); In the above formula, q j p j and v j These are the difference threshold, preference threshold, and rejection threshold, respectively, and 0 < q. j <p j <v j <1; W j Indicates attribute weight; O j (A i A k ) indicates that in attribute C j Option A i For option A k The harmony index; D j (A i A k ) indicates that in attribute C j Option A i For option A k Disharmony indicators; O(A) i A k ) represents scheme A i For option A k The comprehensive harmony index; the net harmony index Z(A) i A k (This indicates case A) i For option A k Priority, J(A) i A k ) is all D j (A i A k )≤O(A i A k A collection of attributes; ψ + (A i ) represents scheme A i The harmony dominance index; ψ - (A i ) represents scheme A i The disharmony dominance index; ψ(A i Scheme A i Net credibility index; S6.

3. Establish a goal programming model and solve for the prospect theoretical parameters α, β, and λ; the net confidence level of the solution can then be obtained; the expression is as follows: In the formula, ψ(A) i ) represents scheme A i The net dominance index, where α and β represent the decision-maker's sensitivity to gains and losses, respectively; λ is the loss aversion coefficient. S6.

4. Solve for the conditional probability based on the idea of ​​relative proximity. The expression is as follows: In the formula, ψ(A) i ) represents scheme A i Net credibility index; maxψ(A) i ) and minψ(A i ) represent the maximum and minimum values ​​of the net confidence index for each scheme, respectively.

7. The method for evaluating potential telecommunications customers based on probabilistic language three-branch decision-making as described in claim 1, characterized in that, In S7, the conditional probability is related to the threshold x. i and y i The objectives are compared, classified, and ranked within each category based on the net credibility index to obtain the final decision result. Includes the following steps: S7.

1. Compare the conditional probability with the threshold to determine the classification result of the scheme, as shown in the following expression: (P0) If Pr(X|[A i ])≥x i Then A i ∈POS(X), (B0)y i <Pr(X|[A i )<x i , then A i ∈BND(X), (N0) If Pr(X|[A i ])≤y i Then A i ∈NEG(X). In the formula, Pr(X|[A i ]) represents the conditional probability that a solution belongs to state X; POS(X), BND(X), and NEG(X) represent the acceptance region, boundary region, and rejection region, respectively; S7.2 Within each region, sort the proposals according to their net confidence level.

8. A method for evaluating potential telecommunications customers based on probabilistic language three-branch decision-making, as described in any one of claims 1-7, characterized in that: It also includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor; when the computer program is loaded onto the processor, it implements the aforementioned probabilistic language-based three-way decision-making method for evaluating potential telecommunications customers.

Images