CN113343573B - User perception evaluation method based on backtracking search algorithm and electronic equipment - Google Patents

Abstract

The invention discloses a user perception evaluation method based on a backtracking search algorithm, which comprises the following steps: s1, defining a QoE variable structure model, inputting a variable which is a KQI index of a service to be evaluated, and outputting a variable which is the comprehensive user perception of the service; s2, adopting a binary tree coding relation prediction model; s3, improving a BSA (bovine serum albumin) algorithm to adapt to a cross mutation strategy of an operator, wherein the cross mutation strategy comprises a mutation mode realized based on a mutation step length, and a single-point factor and combination factor combined mutation mode is provided; s4, realizing the crossing strategy based on the mapping value Map, and providing a single-point and multi-point crossing factor retaining strategy. The method aims at the construction research of the KQI-QoE numerical value relation model to improve the quality and the construction efficiency of the relation model, and is applied to the user experience degree evaluation of the hanging device service scene under the platform of the Internet of things. The invention also provides corresponding electronic equipment.

Description

User perception evaluation method based on backtracking search algorithm and electronic equipment

Technical Field

The invention belongs to the technical field of communication, and particularly relates to a user perception evaluation method based on a backtracking search algorithm and electronic equipment.

Background

The user perception metric method generally includes two methods, one is to actually investigate the user experience quality of the application, and the other is to infer the user experience quality of the application by analyzing the service index. The two modes of actual investigation and business index analysis are insufficient to visually acquire user perception in real time. Currently, with the wide application of communication technologies and internet of things platforms, based on a large amount of collected and stored high-quality and useful data, various services and applications urgently need to improve the application experience of users. Due to the diversity of users, actual scenes and performance parameters, the problem of solving the user perception model is regarded as an NP-hard difficulty problem. Currently, most of the applications of fixed models or one-time multiple linear regression fitting prediction result in large deviation between prediction perception and actual perception. In addition, some scholars use GA (Genetic Algorithm) based complex fitting function construction, but the Algorithm cross mutation strategy is weak in guaranteeing solution diversity and has inherent defects of early maturity, low efficiency and the like.

Disclosure of Invention

Aiming at the defects or improvement requirements of the prior art, the invention provides a user perception evaluation method based on a backtracking search algorithm, which aims at the construction research of a KQI-QoE numerical value relation model to improve the quality and construction efficiency of the relation model and is applied to the user experience evaluation of a hanging device service scene under an Internet of things platform.

In order to achieve the above object, according to an aspect of the present invention, there is provided a user perception evaluation method based on a backtracking search algorithm, including:

s1, defining a QoE variable structure model, inputting a variable which is a KQI index of a service to be evaluated, and outputting a variable which is the comprehensive user perception of the service;

s2, adopting a binary tree coding relation prediction model;

s3, improving a BSA (bovine serum albumin) algorithm to adapt to a cross mutation strategy of an operator, wherein the cross mutation strategy comprises a mutation mode realized based on a mutation step length, and a single-point factor and combination factor combined mutation mode is provided;

s4, realizing the crossing strategy based on the mapping value Map, and providing a single-point and multi-point crossing factor retaining strategy.

In an embodiment of the present invention, in step S1, the QoE variable structure model is:

given N sets of sample data, the objective function of the integrated user perception evaluation problem with T sub-term perceptibility is represented as follows:

wherein, y_itDenotes the t-th sub-term perceptibility, y, of the i-th sample_itRepresenting the t-th sub-term perceptibility, x, of the ith sample as evaluated by the independent variable_it1The first key quality index of the t sub-item perceptibility of the ith sample, and so on, f_tAnd predicting the relation for the sub-item perception.

In an embodiment of the present invention, the step S2 specifically includes:

considering T sub-tree as a feasible solution to the problem, where each sub-tree T is encoded using a binary tree, and the binary tree satisfies the following model constraints: (1) the binary tree satisfies a maximum depth constraint; (2) the binary tree consists of an operator, an independent variable and a constant; (3) selecting the range of each node element of the binary tree as an operator, an independent variable or a constant; (4) the root node and the intermediate node of the binary tree must be operators, and the end node must be an independent variable or a constant; (5) and the fitness value of the prediction model of the sub-item corresponding to the binary tree is not less than a preset value.

In an embodiment of the present invention, the step S3 includes: initializing a population, selecting an operation I, performing mutation operation, performing cross operation, performing legalization operation and selecting an operation II, wherein two populations are initialized, and P is respectively_oldAnd P_initThe population size is S, and each individual in the population comprises T sub-term prediction models.

In an embodiment of the present invention, the operation of selecting I specifically includes:

judging the population P through random probability_oldWhether the existing population information needs to be kept or not, if the existing population information does not need to be kept, the population P is used_initUpdating the current population P_oldThe random probability includes that random numbers a and b obey [0,1 ]]The normal distributions within the range are denoted as a to U (0,1), b to U (0,1), respectively;

randomly changing the population P_oldI.e. the prediction model order.

In an embodiment of the present invention, the mutation operation specifically includes:

by P_oldGuide P_initIndividual variation, i.e. variation of the prediction model, results in a population of variations P of size S_mutTo enhance the diversity of the prediction model, realize the prediction model variation according to the statistical variation step length, and realize the prediction model variation according to the variationThe different factor types are different, and variation constraints adapted to a prediction model are defined so as to enhance the quality of the variant individuals.

In one embodiment of the invention, the statistical variation step size process satisfies the following constraints:

the positions of the variation factors are different and interchangeable;

elements of different classes are not interchanged, namely operators are of one class, and independent variables and constants are of one class;

directly exchanging unary operators and binary operators inevitably generates illegal operators, and exchanging the operators by using combined variation factors;

and if the length of the mapping position of the variation factor exceeds the maximum value of the position, selecting any element different from the original value from the same type factor for replacement.

In an embodiment of the present invention, the legalization operation includes a direct clipping method, a minimum legal tree method, and a random new binary tree method, where:

direct clipping method, clipping the part beyond the limit directly, and combining the original termination point x₁₁And x₁₂Directly replacing the binary tree into a trimmed binary tree termination point;

a minimum legal tree method randomly generates a minimum binary tree, namely replacing an illegal part of the binary tree by a binary tree with the depth of 1;

and (5) randomly updating the binary tree, and discarding the current illegal binary tree.

In one embodiment of the invention, the select II operation comprises:

test population P_trialTest individuals with better fitness value in the population P are used for eliminating the population P_initThe prediction model with poor medium Fitness compares the sub-item Fitness, Fitness (P)_{init k,t}) Represents P_initThe Fitness value of the kth prediction model in (1) at the sub-term t, Fitness (P)_{trial k,t}) Represents P_trialThe Fitness value of the kth prediction model in (1) at the sub-term t if Fitness (P)_{trial k,t}) Superior to Fitness (P)_{init k,t}) Then use P_trialThe scheme of the kth prediction model in the sub-item t eliminates the corresponding P_initThe solution of the kth predictive model in subentry t; otherwise, the original P is kept_initThe scheme of the kth prediction model in the sub-item t is unchanged, and finally, the updated population P with the size of S is formed_init。

In one embodiment of the present invention, the interleaving policy includes: the variant population P obtained by the variant operation_mutAs and group P_initCross-over objects of the individuals, i.e. of the prediction models, resulting in a cross-over test population P of size S_trialFurther enhancing the population diversity, providing a single-point and multi-point cross factor retention strategy, and retaining individual characteristics to a certain extent, namely partial characteristics of a prediction model.

In an embodiment of the present invention, the process of the crossing policy specifically includes:

create a Map value Map, producing two [0-1 s ]]Random numbers d and e in the ranges are denoted d to U (0,1), e to U (0,1), respectively, if d<e, calculating the cross probability P_mixed、[0-1]The product of the random number and the element number is rounded up to determine the number of mapping elements; if d is>e, calculating [0-1 ]]The product of the random number and the number of the elements is rounded up, the index of the mapping elements is determined, the initial mapping value Map is 1, and the randomness of the crossing strategy and the diversity of crossing results are enhanced by two different mapping modes;

by P_init、P_mutAnd creating a test population P by using the corresponding mapping value Map_trialAnd keeping the branch of the binary tree where the element with the corresponding value of 0 is positioned unchanged, and performing cross change on the rest elements according to the variant prediction model to generate a test individual, namely a test prediction model.

According to another aspect of the present invention, there is also provided an electronic apparatus including:

at least one processor; and the number of the first and second groups,

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the above-described backtracking search algorithm-based user perception assessment method.

Generally, compared with the prior art, the technical scheme of the invention has the following beneficial effects:

the QoE evaluation scheme based on the BSA algorithm is constructed. Firstly, the QoE variable structure model is defined, the input variable is a KQI index of a service to be evaluated, and the output variable is the comprehensive user perception of the service. Then, the system adopts a binary tree coding relation prediction model. Meanwhile, the improved BSA algorithm is suitable for a cross mutation strategy of an operator, and comprises a mutation mode realized based on a mutation step length and a combined mutation mode of a single-point factor and a combined factor; and realizing a cross strategy based on the mapping value Map, and providing a single-point and multi-point cross factor retention strategy. In addition, the system provides three mixed legalization strategies to be applied to the legalization relation prediction model. The method can establish a reasonable mapping relation for unknown characteristic data, provide at least one relation prediction model, and quickly obtain an approximate optimal relation prediction model through evolution iteration. The obtained approximate optimal relationship prediction model is beneficial to realizing real-time inference of user perception by the service or the application, can assist the user to find out service or application problems in time, and improves the performance and user experience of the service or the application.

Drawings

FIG. 1 is an overall framework of user perception assessment in an embodiment of the present invention;

FIG. 2 is a variable structure model according to an embodiment of the present invention;

FIG. 3 is sample data according to an embodiment of the present invention;

FIG. 4 is a solution relationship prediction model framework according to an embodiment of the present invention;

FIG. 5 is a block diagram of a possible solution scheme in an embodiment of the present invention;

FIG. 6 is a possible solution scheme of sub-item 1 according to an embodiment of the present invention;

FIG. 7 is a diagram illustrating statistical variation step sizes according to an embodiment of the present invention;

FIG. 8 is a relationship prediction model after mutation according to an embodiment of the present invention;

FIG. 9 is a diagram illustrating the creation of a Map value Map according to an embodiment of the present invention;

FIG. 10 is a cross-over relationship prediction model in an embodiment of the invention;

FIG. 11 illustrates a validation strategy in accordance with an embodiment of the present invention;

FIG. 12 is a diagram of a relationship prediction model after operation II is selected, in accordance with an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

How to use Key Quality Indicators (KQI) data of acquisition services or applications to realize fast inference of user perception (Quality of Experience, QoE) and guarantee the inference Quality. In fact, a reasonable mapping relation is established for unknown characteristic data such as KQI-QoE, and the construction method is suitable for various service application scenarios. In order to better solve the optimal relationship model, the invention provides a novel QoE evaluation scheme based on a Backtracking Search Algorithm (BSA), which specifically comprises the following steps:

firstly, the invention establishes a mathematical model aiming at the user perception evaluation problem and constructs a comprehensive user perception evaluation model.

For any service, a variable structure and a variable relation exist to represent the user perception. Generally, the service to be evaluated includes multiple performances, and each performance corresponds to a set of KQI indexes and a sub-item of user perceptibility. The quality of the KQI index directly affects the quality of the perception of the sub-item user, and the quality of the perception of the sub-item directly affects the quality of the perception of the service comprehensive user. Thus, the overall user perception is expressed as follows:

among them, QoE_tRepresenting the perception degree of the tth sub-item; lambda [ alpha ]_tA weight representing the perception of the tth sub-item; t represents the number of perceptibility of the included sub-items, f in the formula_tNamely, the scheme needs to solve the service subitem perceptibility evaluation, and the corresponding optimal KQI-QoE relationship model.

Solving the problem of the relation model of unknown characteristic data of the KQI-QoE, namely, obtaining each sub-item perception relation model f by considering T sub-item perception metrics_tThe optimal variable relationship of (c). The system adopts the minimum residual sum of squares as an optimization target, namely a legal variable relation model is obtained under the condition of meeting known constraint conditions, so that the sum of the evaluation errors of the T sub-item perceptibility is minimum. Therefore, knowing N sets of sample data, the objective function of the integrated user perception assessment problem with T sub-term perceptibility can be expressed as follows:

wherein, y_itDenotes the t-th sub-term perceptibility of the i-th sample. y is_itAnd expressing the t sub-item perceptibility of the ith sample obtained by independent variable evaluation. x is the number of_it1The first key quality index of the t sub-item perceptibility of the ith sample is expressed, and the like. Because the natural selection process usually determines the advantages and disadvantages of individuals by the fitness value, the invention takes the reciprocal of the objective function value as the fitness value of the individual to be used in the selection process of decision-making feasible solutions.

And solving the optimal solution scheme of the problem by using a BSA algorithm. The present system considers T subtrees as a feasible solution to the problem, also commonly referred to as an individual or a chromosome. Each subitem tree t adopts binary tree coding, the root nodes and the middle nodes of the binary tree are operators, and the end nodes are independent variables or constants, namely each subitem perception relation prediction model f_tBy independent variablesSymbol set, operator symbol set and constant. The system takes the inverse of the objective function value as the individual fitness value. The specific algorithm flow comprises population initialization, selection I operation, mutation operation, crossover operation and selection II operation. Wherein two populations, each P, are initialized_oldAnd P_initThe population size is S.

Warp of warp P_oldAnd P_initVariation generating a variant population P of size S_mut. This document is based on the population P_oldAnd (3) guiding individual variation, and increasing the diversity of variation modes through the variation modes realized based on the variation step length. Considering special operator factor characteristics, the system defines variation constraints, provides a variation mode combining single point factors and combination factors, and enhances the quality of variation individuals.

Warp of warp P_mutAnd P_initCross-producing test population P of size S_trial. The variant population P obtained by the variant operation_mutAs and group P_initCross-subjects of individuals. The system realizes a cross strategy based on the Map value Map, provides a reasonable single-point and multi-point cross factor retention strategy, and retains individual characteristics to a certain extent.

Test population P generated by legalization cross variation based on mixed legalization strategy_trialAnd (4) illegal individuals. The system provides three legalization schemes, and a cutting strategy is implemented on the filial generation of a new binary tree generated by cross variation, wherein the maximum tree depth of the new binary tree exceeds the limit. The first method comprises the following steps: directly cutting leaves exceeding the maximum limit part, and then changing the middle node of the last layer after cutting into a terminal node; and the second method comprises the following steps: relative to the root node, cutting the left/right subtrees exceeding the maximum limit part, and then randomly generating minimum legal subtrees according to cutting elements for completion; and the third is that: discarding the new illegal individual, and randomly generating a legal individual again. The strategy effectively controls the complexity of the generated binary tree, namely the complexity of a relation prediction model.

Obtaining a legal test population P_trialEliminating population P by using test population_initThe individuals with lower moderate fitness satisfy the fitnessPrinciple of survival. The system sequentially circulates the sub-items, each sub-item iterates ITER times, each iteration keeps the sub-items with higher fitness values, and the sub-items with lower fitness values are eliminated. The system initially sets ITER 1000.

All the sub-items are evolved and iterated, and the population P_initThe individual with the highest fitness value is the optimal individual, namely, the optimal relation prediction model f is solved_tAnd obtaining a comprehensive perception evaluation solution scheme.

Example 1:

the invention provides a user perception evaluation method based on a BSA (bovine serum albumin) algorithm, which comprises two parts of constructing a comprehensive user perception evaluation variable model and solving a KQI-QoE (KQI-QoE) relation prediction model based on the BSA algorithm as shown in figure 1.

In step 201, an integrated user perception assessment model is constructed. Generally, the service to be evaluated includes multiple performances, and each performance corresponds to a set of KQI indexes and a sub-item of user perceptibility. As shown in fig. 2, the perception evaluation impact is bottom-up, the better the sub-term perception, the better the overall perception. Thus, the integrated user perceptibility can be expressed as follows:

among them, QoE_tRepresenting the perception degree of the tth sub-item; lambda [ alpha ]_tA weight representing the perception of the tth sub-item; t represents the number of perceptibility of the included sub-items, f in the formula_tNamely, the service subitem perception degree evaluation model which needs to be solved corresponds to the optimal KQI-QoE relation model.

For example, taking the solution of the user perception model of the video streaming service as an example, the known sub-item perception evaluation of the video streaming service includes: whether the video file can be played satisfactorily, the smoothness of video playing pictures and the smoothness of video playing soundsThree sub-items of degree, respectively called sub-item 1, sub-item 2, and sub-item 3 for short, and their sub-item perceptibility is respectively recorded as QoE₁、QoE₂、QoE₃. The quality of the perception of the sub-items directly influences the whole user perception of the video stream playing service. Then, initially setting the weight of the sub-term to λ₁＝0.4、λ₂＝0.3、λ₃When the video stream playing service is equal to 0.3, the integrated user perception QoE of the video stream playing service is expressed as:

in step 202, each sub-item perceptibility evaluation model f_tA mapping relationship representing KQI-QoE.

For example, taking the satisfaction degree of the sub-item 1 video file, the corresponding set of KQI indicators includes the packet loss rate x₁₁Delay x₁₂Transmission rate x₁₃. Thus, the relationship model f of sub-item 1_tThe method is formed by combining an independent variable symbol set and an operator. Wherein the independent variable symbol set comprises { x₁₁、x₁₂、x₁₃The operator set includes unary operators { log, exp, sin, cos, tan, fabs }, binary operators { +, -,/, }. From this, f is₁＝ln(x₁₁)+(x₁₂^0.5)*ln(x₁₂)+(x₁₃^2) a prediction model of user perceptibility between key index KQI-objective QoE, y) that can be regarded as subentry 1_itRepresenting the perception degree of the tth sub-item of the ith sample; the f-value is a predicted value calculated from the sample index. However, there may be a deviation between the actual user perceptibility and the predicted user perceptibility obtained by the prediction model, and therefore an optimal evaluation model needs to be solved.

Solving the optimal relationship model problem of unknown characteristic data of KQI-QoE, namely, considering T sub-item perception metrics to obtain each sub-item perception relationship model f_tThe optimal variable relationship of (c). The system adopts the minimum residual sum of squares as an optimization target, namely a legal variable relation model is obtained under the condition of meeting known constraint conditions, so that the sum of the evaluation errors of the T sub-item perceptibility is minimum. Therefore, knowing N sets of sample data, the objective function of the integrated user perception assessment problem with T sub-term perceptibility can be expressed as follows:

wherein, y_itDenotes the t-th sub-term perceptibility of the i-th sample. y is_itAnd expressing the t sub-item perceptibility of the ith sample obtained by independent variable evaluation. x is the number of_it1The first key quality index of the t sub-item perceptibility of the ith sample is expressed, and the like.

For example, as shown in fig. 3, two sets of data in the collected sample data are taken as an example. If the 1 st sub-item perception QoE₁Is f₁＝ln(x₁₁)+(x₁₂^0.5)*ln(x₁₂)+(x₁₃^2), the predicted user perception value of the first set of data is: y is₁₁＝f₁(x₁₁，x₁₂，x₁₃) -ln (0.36) + (10^0.5) × ln (10) + (1.2^2) ═ -1.02+3.16 ^ 2.3+1.44 ^ 7.69; the predicted user perception values for the second set of data are: y is₂₁＝f₁(x₁₁，x₁₂，x₁₃) -ln (0.72) + (5^0.5) × ln (5) + (2.1^2) ═ -0.33+2.24 ^ 1.61+4.41 ^ 7.68. Then, the deviation F from the actual user perceptibility factor is (8-7.69) ^2+ (6-7.68) ^ 2^ 0.0961+2.8224 ^ 2.9185. Obviously, the smaller the prediction bias, the more accurate the characterization prediction model.

Example 2:

the embodiment of the invention is to explain the implementation process of the invention by combining a specific scene on the basis of the technical content of the embodiment 1. Solving a KQI-QoE optimal relationship prediction model based on BSA (bovine serum albumin), as shown in FIG. 4, comprises the following specific steps:

in step 301, the encoding scheme and model constraints of the KQI-QoE prediction model are determined.

For example, as shown in fig. 4, the present system encodes the sub-item prediction model using a binary tree coding scheme, and the binary tree satisfies the following model constraints: (1) binary tree satisfying maximum depth constraint(ii) a (2) The binary tree consists of an operator, an independent variable and a constant; (3) selecting the range of each node element of the binary tree as an operator, an independent variable or a constant; (4) the root node and the intermediate node of the binary tree must be operators, and the end node must be an independent variable or a constant; (5) the fitness value of the prediction model of the corresponding sub-item of the binary tree is not less than a preset value (in the embodiment of the invention, 1e is taken as^-6)。

In step 302, a population, i.e., a binary tree population, is initialized. Randomly generating two groups of initial binary tree solution schemes with the size of S, namely an initial population P_initAnd P_old. Each individual in the population includes T sub-term prediction models. The system initially sets the scale size S to 50.

For example, as shown in fig. 5, the prediction models of sub-item 1, sub-item 2, and sub-item 3 constitute a feasible solution for evaluating the comprehensive user perceptibility of the video streaming service, which is also referred to as an individual. S such individuals constitute the initial population. Thus, in the example T-3.

Next, the prediction models of sub-item 1, sub-item 2, and sub-item 3 are iterated in sequence until a set maximum number of iterations is reached. The following steps and examples take as an example a prediction model of whether a video file can play the satisfaction sub-item 1.

In step 303, select I operation, enhance population P_oldGuiding the randomness of the evolution process. The method mainly comprises two steps: (1) first obey [0,1 ] by random probability (random numbers a and b)]Normal distributions within the range are denoted as a to U (0,1) and b to U (0,1), respectively, and the population P is judged_oldWhether the existing population information needs to be kept or not, if the existing population information does not need to be kept, the population P is used_initUpdating the current population P_old. (2) Randomly changing the population P_oldI.e. the prediction model order.

For example, in the iterative process of the prediction model of sub-term 1, if the random number a is 0.6 and b is 0.5, i.e. a>b, keeping the existing population P_oldIs not changed, otherwise the population P is used_initSub item 1 of (1) updates the current population P_oldSub item 1 of (1). Then, the population P_oldSub-item 1 prediction model of (1) order random typingAnd disorder, so that the randomness of natural selection evolution is ensured.

In step 304, the strategy is mutated. By P_oldGuide P_initIndividual variation, i.e. variation of the prediction model, results in a population of variations P of size S_mutTo enhance the diversity of the prediction model. Different from the previous variation strategy, the system realizes the prediction model variation according to the statistical variation step length, and defines the variation constraint suitable for the prediction model according to the different types of the variation factors so as to enhance the quality of the variation individuals.

The system needs to satisfy the following constraints in the process of counting the variation step length: (1) the positions of the variation factors are different and interchangeable; (2) elements of different classes are not interchanged, namely operators are of one class, and independent variables and constants are of one class; (3) directly exchanging unary operators and binary operators inevitably generates illegal operators, and exchanging the operators by using combined variation factors; (4) if the length of the mapping position of the variation factor exceeds the maximum value of the position, selecting any element different from the original value from the same type factor for replacement; (5) the repetition factors are interchanged in a one-to-one correspondence mode according to the appearing sequence, and the repetition factors are not interchanged if the corresponding sequence is absent.

In step 3041, P is selected_oldAnd P_initThe number of times of interchange E occurring is counted through one-to-one mapping by sequentially corresponding prediction models_swap。

For example, as shown in FIG. 6, the sub-term 1 prediction models are selected to be f_old1＝exp(x₁₁)+ln (x₁₂)+ln(x₁₃) And f_init1＝ln(x₁₁)+(x₁₂^2)+exp(x₁₃). As shown in fig. 7, the selected prediction model is subjected to a middle-order traversal, and operators are compared in sequence. Such as the independent variable x₁₁Are all in position 1, not interchangeable; e.g. operator exp at position 2 and position 9, P respectively_initOperators ln and exp corresponding to the middle position 2 and the position 9 can be interchanged; e.g. operator ln at position 5 and position 9, P respectively_initOperators corresponding to the middle position 5 and the position 9 can be interchanged, but because the dyadic operator ^ right node is a constant and the monobasic operator ln has no right node, the direct exchange of the operators ^ and ln easily causes illegal, a combined variation factor needs to be set. Thus, in binary operationSymbol ^ for example, swap their right node combinations, i.e., swap the combination variance factor ^2 with the unary operator ln. Such as P_oldThe operator ln corresponding to the middle position 8 is the second position of its occurrence, and P_initDoes not appear once, has no interchange mapping and is not interchangeable. By parity of reasoning, finally obtaining the interchange times E_swapThe size is 2.

In step 3042, a method is performed by generating [0,1 ]]The uniform random variable c in the range is designated as c-U (0,1), then c and E are used_swapThe product of the above steps determines the variation step length, and finally obtains the variant individual, namely the variant prediction model.

For example, based on the number of interchanges E obtained in step 3041_swap2, if [0,1 ]]The uniform random variable c in the range is 0.51, the actual number of times of interchange is the product of the variable c and the number of times of interchange, i.e. c × E_swapRounding up to 0.51 × 2 to 1.02 results in an actual rotational variation step of 2. Thus, P in FIG. 6_initThe mutation results in a new prediction model for sub-item 1, as shown in FIG. 8, i.e., a new relational prediction model f_mut1＝exp(x₁₁)+ln(x₁₂)+x₁₃^2。

In step 305, the strategy is crossed. The variation population P obtained by the variation operation in the step 304_mutAs and group P_initCross-over objects of the individuals, i.e. of the prediction models, resulting in a cross-over test population P of size S_trialAnd further enhancing the population diversity, namely the diversity of the prediction model. The system provides a single-point and multi-point cross factor retention strategy, and individual characteristics, namely partial characteristics of a prediction model, are retained to a certain extent. The method comprises the following specific steps:

the first step is as follows: a Map value Map is created. As shown in FIG. 9, two [0-1 ] s are generated]Random numbers d and e in the ranges are denoted d to U (0,1), e to U (0,1), respectively, if d<e, calculating the cross probability P_mixed、 [0-1]The product of the random number and the element number is rounded up to determine the number of mapping elements; if d is>e, calculating [0-1 ]]And (4) multiplying the random number and the number of the elements, rounding up, and determining the index of the mapping element. The initial mapping values Map are all 1. Two different mapping modes enhance intersectionRandomness of the cross strategy, diversity of the cross results.

The second step is that: by P_init、P_mutAnd creating a test population P by using the corresponding mapping value Map_trial. The branch of the binary tree where the element with the corresponding value of 0 is located is kept unchanged, and the rest elements are subjected to cross change according to the variant prediction model to generate a test individual, namely a test prediction model. As shown in fig. 10, the argument packet loss rate x₁₁Operator ln, operator + three elements remain unchanged, then cross-delaying x₁₂Transmission rate x₁₃The subtree where the argument factor is located. E.g. comprising a factor delay x₁₂The subtree and the variant prediction model of (1) include a factor delay x₁₂The subtrees of (a) are subject to cross-over changes. Meanwhile, it needs to be satisfied that the root nodes of two crossed subtrees must be equal operator priority elements. The operator priorities in order from high to low are: unary operators/exponentiation/multiplication/division, addition/subtraction. For example

f_trial1The fitness value of (A) is:

obviously, f_trial1Has a fitness value of more than f_init1Fitness value of (a), all culling of_init1The prediction model of retention test f_trial1. Thus, after one iteration of sub-term 1 of the prediction model, the prediction model of FIG. 5 evolves into a new prediction model, as shown in FIG. 12.

And (4) carrying out next iteration on the sub-item 1 through the S prediction models after the iteration until the iteration upper limit is reached. And after the sub-item 1 is iterated, sorting the sub-items in a descending order according to the fitness value of the sub-item 1. The system initially sets the number of iterations to 1000.

In step 308, until all sub-items (sub-item 1, sub-item 2, sub-item 3) have passed through the above-described stepsAfter the iteration of step 303-307, P is selected_initAnd taking the prediction model with the highest medium fitness value as an approximate optimal prediction model. And taking the inverse of the F value of the objective function as the fitness value of the final prediction model. Thus, the higher the fitness value, the better the prediction model. And determining whether the sub-item 1 video file can be played to a satisfactory degree, the sub-item 2 video playing picture fluency satisfactory degree and the sub-item 3 video playing sound fluency satisfactory degree in sequence, and determining a comprehensive user perception model of the video stream playing service.

Further, the present invention also provides an electronic device, comprising:

at least one processor; and the number of the first and second groups,

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the above-described backtracking search algorithm-based user perception assessment method.

It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (6)

1. A user perception evaluation method based on a backtracking search algorithm is characterized by comprising the following steps:

s1, defining a QoE variable structure model, inputting a variable which is a KQI index of a service to be evaluated, and outputting a variable which is the comprehensive user perception of the service; the QoE variable structure model in step S1 is:

given N sets of sample data, the objective function of the integrated user perception evaluation problem with T sub-term perceptibility is represented as follows:

wherein, y_itDenotes the t-th sub-term perceptibility, y, of the i-th sample_itRepresenting the t-th sub-term perceptibility, x, of the ith sample as evaluated by the independent variable_it1The first key quality index of the t sub-item perceptibility of the ith sample, and so on, f_tPredicting a relation for the subitem perception;

s2, adopting a binary tree coding relation prediction model; the method specifically comprises the following steps: considering T sub-tree as a feasible solution to the problem, where each sub-tree T is encoded using a binary tree, and the binary tree satisfies the following model constraints: (1) the binary tree satisfies a maximum depth constraint; (2) the binary tree consists of an operator, an independent variable and a constant; (3) selecting the range of each node element of the binary tree as an operator, an independent variable or a constant; (4) the root node and the intermediate node of the binary tree must be operators, and the end node must be an independent variable or a constant; (5) the fitness value of the sub item prediction model corresponding to the binary tree is not smaller than a preset value;

s3, improving a BSA (bovine serum albumin) algorithm to adapt to a cross mutation strategy of an operator, wherein the cross mutation strategy comprises a mutation mode realized based on a mutation step length, and a single-point factor and combination factor combined mutation mode is provided; the method comprises the following steps: initializing a population, selecting an operation I, performing mutation operation, performing cross operation, performing legalization operation and selecting an operation II, wherein two populations are initialized, and P is respectively_oldAnd P_initThe size of the population is S, and each individual in the population comprises T sub-item prediction models; wherein:

the mutation operation specifically comprises the following steps: by P_oldGuide P_initIndividual variation, i.e. variation of the prediction model, results in a population of variations P of size S_mutThe diversity of the prediction model is enhanced, the prediction model variation is realized according to the statistical variation step length, and the variation constraint which is suitable for the prediction model is defined according to the difference of the types of the variation factors, so that the quality of the variant individual is enhanced; the statistical variation step size process satisfies the following constraints: the positions of the variation factors are different and interchangeable; elements of different classes are not interchanged, namely operators are of one class, and independent variables and constants are of one class; direct interchange of unary and binary operatorsIf the illegal condition occurs, the combined variation factors are used for interchange; if the length of the mapping position of the variation factor exceeds the maximum value of the position, selecting any element different from the original value from the same type factor for replacement;

the legalization operation comprises a direct cutting method, a minimum legal tree method and a random new binary tree method, wherein: direct clipping method, clipping the part beyond the limit directly, and combining the original termination point x₁₁And x₁₂Directly replacing the binary tree into a trimmed binary tree termination point; a minimum legal tree method randomly generates a minimum binary tree, namely replacing an illegal part of the binary tree by a binary tree with the depth of 1; random new binary tree method, abandon present illegal binary tree;

s4, realizing the crossing strategy based on the mapping value Map, and providing a single-point and multi-point crossing factor retaining strategy.

2. The method for evaluating user perceptibility based on backtracking search algorithm according to claim 1, wherein the operation of selecting I specifically comprises:

judging the population P through random probability_oldWhether the existing population information needs to be kept or not, if the existing population information does not need to be kept, the population P is used_initUpdating the current population P_oldThe random probability includes that random numbers a and b obey [0,1 ]]The normal distributions within the range are denoted as a to U (0,1), b to U (0,1), respectively;

randomly changing the population P_oldI.e. the prediction model order.

3. The backtracking search algorithm-based user perceptibility evaluation method of claim 1, wherein said selecting II operation comprises:

test population P_trialTest individuals with better fitness value in the population P are used for eliminating the population P_initThe prediction model with poor medium Fitness compares the sub-item Fitness, Fitness (P)_{init k,t}) Represents P_initThe Fitness value of the kth prediction model in (1) at the sub-term t, Fitness (P)_{trial k,t}) Represents P_trialAt sub-term tIf Fitness (P)_{trial k,t}) Superior to Fitness (P)_{init k,t}) Then use P_trialThe scheme of the kth prediction model in the sub-item t eliminates the corresponding P_initThe solution of the kth predictive model in subentry t; otherwise, the original P is kept_initThe scheme of the kth prediction model in the sub-item t is unchanged, and finally, the updated population P with the size of S is formed_init。

4. The backtracking search algorithm-based user perceptibility evaluation method of claim 1, wherein said crossover strategy comprises: the variant population P obtained by the variant operation_mutAs and group P_initCross-over objects of the individuals, i.e. of the prediction models, resulting in a cross-over test population P of size S_trialFurther enhancing the population diversity, providing a single-point and multi-point cross factor retention strategy, and retaining individual characteristics to a certain extent, namely partial characteristics of a prediction model.

5. The user perception evaluation method based on the backtracking search algorithm according to claim 4, wherein the process of the cross strategy specifically comprises:

create a Map value Map, producing two [0-1 s ]]Random numbers d and e in the ranges are denoted d to U (0,1), e to U (0,1), respectively, if d<e, calculating the cross probability P_mixed、[0-1]The product of the random number and the element number is rounded up to determine the number of mapping elements; if d is>e, calculating [0-1 ]]The product of the random number and the number of the elements is rounded up, the index of the mapping elements is determined, the initial mapping value Map is 1, and the randomness of the crossing strategy and the diversity of crossing results are enhanced by two different mapping modes;

by P_init、P_mutAnd creating a test population P by using the corresponding mapping value Map_trialAnd keeping the branch of the binary tree where the element with the corresponding value of 0 is positioned unchanged, and performing cross change on the rest elements according to the variant prediction model to generate a test individual, namely a test prediction model.

6. An electronic device, comprising:

at least one processor; and the number of the first and second groups,

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

Images