A software quality evaluation method based on fuzzy matter element and MEREC empowerment

Abstract

The application discloses a software quality evaluation method based on fuzzy matter-element and MEREC empowerment. The application converts the structural characteristics of a UML class diagram into a quantifiable analysis fuzzy matter-element model, overcomes the processing bottleneck of index information fuzziness and uncertainty in a traditional measurement method, and significantly improves the logical consistency between original measurement data and comprehensive evaluation conclusion. An objective empowerment mode based on a removal effect measure is adopted, so that the weight is completely driven by the internal structure of data, subjective assignment interference on the evaluation result is effectively avoided, and the direct corresponding relationship between the weight and the final ranking is ensured. In the ranking stage, the Euclidean distance calculation is replaced by the algebraic deviation accumulation, the robustness of the evaluation result to the increase and decrease of the scheme set is greatly improved, a feedback type weight correction mechanism is introduced, the weight can be adaptively adjusted according to the ranking discriminability, and the ranking failure problem caused by score centralization is effectively prevented.

Description

Technical Field

[0001] This invention relates to the field of software quality evaluation technology, specifically to a software quality evaluation method based on fuzzy matter-element and MEREC weighting. Background Technology

[0002] In the design and development of object-oriented software systems, UML (Unified Modeling Language) class diagrams, as the core modeling tool for describing the static structure of the system, directly impact the understandability, maintainability, and testability of the software. With the continuous expansion of software scale, the number of elements such as classes, associations, and inheritance in class diagrams increases dramatically, leading to increasingly complex system structures and posing a serious challenge to software quality assessment. How to objectively and effectively evaluate the structural complexity of UML class diagrams has become an important research topic in the field of software engineering.

[0003] Currently, research on the complexity evaluation of UML class diagrams mainly falls into two categories. One category relies on software metrics theory, quantifying complexity by defining and calculating metrics such as size, coupling, and cohesion of the class diagram. Classic methods like the CK metric suite and MOOD metric set are widely used in the industry. However, these methods typically treat each metric as independent and equally important, rarely considering the differentiated contributions of different metrics to the overall complexity, leading to evaluation results that fail to accurately reflect class diagram quality. The other category introduces multi-attribute decision-making methods, using weighting techniques such as the analytic hierarchy process (AHP) and entropy weighting to determine metric weights, and combining these with ranking algorithms like TOPSIS and VIKOR to comprehensively evaluate the class diagram. This approach somewhat compensates for the shortcomings of the equal-weight assumption, but the weight determination process often suffers from strong subjectivity and a lack of direct logical connection to the evaluation conclusions. Furthermore, existing methods are insufficient in handling the fuzziness and uncertainty of metric values, making them unsuitable for practical scenarios in software quality evaluation where data is incomplete or measurement errors occur.

[0004] Furthermore, in the ranking stage of the comprehensive evaluation, the traditional TOPSIS method uses Euclidean distance to calculate the distance between the proposed solution and the ideal solution, which is prone to nonlinear amplification of subtle differences. This results in a lack of robustness in the ranking when the set of proposed solutions is expanded or reduced. Simultaneously, most existing methods employ an open-loop process of "weighting first, then ranking," meaning that once the weights are determined, they are not adjusted based on the ranking effect. When improper weight settings lead to a concentration of scores among proposed solutions, existing methods lack an effective feedback correction mechanism. Therefore, there is an urgent need for a UML class diagram-based software quality evaluation method that can balance objective weighting of indicators, fuzzy information processing, robust ranking, and adaptive weight adjustment, to more scientifically and comprehensively support the quantitative assessment of software structural complexity. Summary of the Invention

[0005] To address the technical problems mentioned above, this invention provides a software quality evaluation method based on fuzzy matter-element and MEREC weighting, comprising the following steps: S1. Obtain several UML class diagrams to be evaluated as the evaluation objects, and construct a composite fuzzy matter-element matrix; S2. Determine the objective weights of each evaluation index based on the composite fuzzy matter-element matrix; S3. Construct a weighted fuzzy complex element matrix based on the objective weights and the composite fuzzy matter element matrix; S4. Determine the overall utility of each evaluated item based on the weighted fuzzy complex meta-matrix; S5. Based on the overall utility, rank the multiple evaluation items and output the ranking of software quality.

[0006] Preferably, S1 specifically includes: Obtain several UML class diagrams to be evaluated as the evaluation items; Using several evaluation indicators as evaluation features, the original values ​​of each evaluated item on each evaluation indicator are statistically analyzed. Construct a composite fuzzy matter-element matrix based on the original values; In this system, the rows of the composite fuzzy matter element correspond to each evaluated object, the columns correspond to each evaluated index, and the matrix elements are the original values ​​of each object on each index.

[0007] Preferably, S2 specifically includes: The original values ​​are preprocessed by translation to obtain the translated values. Construct a MEREC normalized matrix based on the translated values; calculate the global comprehensive performance value of each evaluated entity based on the MEREC normalized matrix; After removing each evaluation indicator one by one, the performance value is recalculated, and the removal effect value of each indicator is calculated based on the recalculated performance value and the overall performance value. The objective weights of each evaluation index are obtained by normalizing the removal effect value.

[0008] Preferably, S3 specifically includes: multiplying the objective weight of each evaluation index by the corresponding preferential membership degree, and using the product as a weighting element to form a weighted fuzzy complex element matrix.

[0009] Preferably, S4 specifically includes: Obtain the preferred membership matrix from the composite fuzzy matter-element matrix; construct the CRADIS normalized decision matrix from the preferred membership matrix; The best and worst benchmark values ​​are determined based on the CRADIS normalized decision matrix; Based on the excellent benchmark value, the worst benchmark value, the CRADIS normalized value, and the objective weights, calculate the weighted algebraic deviation of each evaluated item from the excellent benchmark value and the weighted algebraic deviation from the worst benchmark value. Calculate the positive and negative utility function values ​​based on the weighted algebraic deviation; calculate the overall utility value based on the positive and negative utility function values.

[0010] Preferably, S5 specifically includes: arranging the evaluation items in descending order of comprehensive utility, with the evaluation item with the highest comprehensive utility corresponding to the best software quality.

[0011] Preferably, the translation preprocessing is to add a preset positive number to each original value.

[0012] Preferably, the method for constructing the MEREC normalization matrix includes: for each evaluation index, using the minimum value of the index after shifting among all evaluation items as the denominator, and using the shifted value of each evaluation item as the numerator, calculating the proportion value, and the proportion values ​​constitute the MEREC normalization matrix.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention overcomes the bottleneck of handling fuzzy and uncertain index information in traditional measurement methods by transforming the structural characteristics of UML class diagrams into quantifiable fuzzy matter-element models, significantly improving the logical consistency between the original measurement data and the comprehensive evaluation conclusions. By employing an objective weighting method based on a removal effect measure, the weights are entirely driven by the inherent structure of the data, effectively avoiding the interference of subjective assignment on the evaluation results, while ensuring a direct correspondence between the weights and the final ranking.

[0014] During the ranking phase, algebraic bias accumulation replaces Euclidean distance calculation, significantly improving the robustness of the evaluation results to changes in the solution set. A feedback-based weight correction mechanism is introduced, enabling weights to adaptively adjust based on the ranking discrimination, effectively preventing ranking failures caused by score centralization. Overall, this invention significantly improves the objectivity, robustness, and data adaptability of UML class diagram software quality evaluation. Attached Figure Description

[0015] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are 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.

[0016] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0017] 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.

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] Example like Figure 1 The diagram shown is a schematic representation of the method flow in this embodiment, and the steps include: S1. Obtain several UML class diagrams to be evaluated as evaluation objects, and construct a composite fuzzy matter-element matrix.

[0020] Pending evaluation A UML class diagram is used to evaluate things. ( ), using 7 complexity evaluation metrics—NDep ( ), NASSOC ), NAgg ( ), NGen ( ), NM ( NA ( ), NC ( — for evaluation characteristics Statistical analysis of the original index values ​​of various types of graphs Construct according to formula (1) 1st order composite fuzzy matter-element matrix For situations where the value of an indicator is difficult to determine precisely or is partially missing, it can be represented in the form of a confidence interval or a fuzzy number. This fully demonstrates the compatibility of fuzzy matter-element analysis with the fuzziness and uncertainty of indicators.

[0021] In the matter-element analysis framework, the evaluation object is... (Things), evaluation indicators (Characteristics of things) and corresponding values The (characteristic values) combined together form the basic unit describing the thing, called the matter element, denoted as . ,in , , These are called the three elements of a matter-element system. If the characteristic value... If it possesses fuzziness (such as being expressed as interval numbers or fuzzy numbers), it is called a fuzzy matter-element. When describing an object... have Features The corresponding values ​​are as follows: When this happens, the corresponding fuzzy matter element is called... 3D fuzzy matter element.

[0022] Will a thing 3D fuzzy matter elements combine together to form a thing 3D composite fuzzy matter element, denoted as , represented as (1) In formula (1), ( ) is the first One UML software class diagram to be evaluated; ( ) is the first One evaluation indicator; For the first The class diagram in the first... The original observed values ​​for each indicator; The total number of class diagrams participating in the evaluation; This represents the total number of evaluation indicators.

[0023] The software evaluation objects involved in this embodiment include a total of There are 10 UML class diagrams, and the evaluation metrics are as follows: There are: (NDep, number of dependencies) (NAssoc, ordinary association number) (NAgg, cluster relation number) (NGen, generalization relation number) (NM, number of methods in a class) (NA, number of attributes within the class) (NC, number of classes).

[0024] In this embodiment, among the seven class graph metrics, count metrics such as NDep, NAssoc, and NGen are naturally likely to appear. In this case, the subsequent MEREC weight calculation uses proportional normalization, and its denominator requires the original values ​​of each indicator; if a certain indicator in a certain category is exactly equal to the minimum value of that indicator in the entire sample, and the minimum value itself is 0, then... Normalized ratio exist The time can still be calculated, but if This will lead to Meaningless.

[0025] To completely avoid the aforementioned zero-value denominator problem, before proceeding to the MEREC weighting step, all original index values ​​are preprocessed by shifting, let: (2) This embodiment takes All subsequent MEREC weighting calculations are based on the shifted values. The dimensionless processing is still based on the original values. In this process, zero values ​​directly participate in min-max normalization without causing division by zero.

[0026] Since applying the same constant to all evaluation objects of the same indicator does not change the size relationship between any two evaluation objects under that indicator, the translation preprocessing does not change the objective comparison basis based on relative order and does not introduce new subjective information.

[0027] Since the dimensions and orders of magnitude of various evaluation indicators usually differ significantly, directly substituting the original values ​​into the comprehensive evaluation calculation may lead to dimensional effects that interfere with the objectivity of the evaluation conclusions. Therefore, it is necessary to perform dimensionless processing on the values ​​of each indicator, that is, to eliminate the influence of dimensions through mathematical transformations and map indicators with different dimensions to a unified comparable scale.

[0028] For the Each evaluation index is calculated to its maximum value according to formulas (3) and (4). With minimum value : (3) (4) Boundary conditions: If (i.e., a certain indicator in all) If all values ​​in the class diagram are exactly the same, then the indicator lacks discriminatory power. In this case, all preferred membership degrees in the column should be uniformly set to the same constant (in this article, we take...). (to indicate that the indicator does not provide effective distinguishing information), and to allow subsequent MEREC to remove the corresponding effect value. Weight This indicator is not included in the overall ranking.

[0029] In software quality evaluation, a smaller complexity index generally indicates a simpler class diagram structure and higher software quality. Therefore, all seven evaluation indices in this embodiment belong to the "smaller is better" type (inverse indices). For inverse indices, the following formula is used for dimensionless calculation to determine the... The first class diagram Each indicator corresponds to the degree of preferential membership. : (5) In the formula, For the first The first class diagram The dimensionless result of the evaluation index value is called the best membership degree. Determined by formula (3); Determined by formula (4); For formula (1) The corresponding element. The larger the value, the stronger the first... The class diagram in the first... The better the quality on each metric (the lower the complexity); when At that time, the type of graph is in the first... Achieving the optimal value (minimum value) on each indicator; when At that time, the type of graph is in the first... The worst value (maximum value) was achieved on each indicator.

[0030] The original values ​​in formula (1) After dimensionless processing using formula (5), the optimal membership fuzzy matter-element matrix is ​​obtained. , is represented as: (6) In formula (6), To optimize the membership degree fuzzy matter-element matrix, the first... Line 1 Column elements It is determined by formula (5). It eliminates the dimensional differences between the original data, making different indicators have a unified and comparable scale, and serves as the basic data matrix for subsequent weight calculation and comprehensive ranking.

[0031] S2. Determine the objective weights of each evaluation index based on the composite fuzzy matter-element matrix.

[0032] In multi-indicator comprehensive evaluation, the method of determining weights directly affects the rationality and credibility of the evaluation conclusions. This embodiment introduces the Removal Effect Weighting (MEREC) method to determine the objective weights of each evaluation indicator. The core idea of ​​MEREC is to measure the actual importance of an indicator by observing the change in the overall evaluation result after removing each evaluation indicator one by one from the calculation—the greater the change in the overall evaluation result after removing an indicator, the more significant the influence of that indicator on the comprehensive evaluation conclusion, and the higher its weight should be. MEREC is an effect-oriented measurement method, and the determined weights have a direct logical correspondence with the evaluation conclusions.

[0033] The MEREC method's weight calculation consists of five steps, all based on the translated values. conduct.

[0034] S201. Construct the MEREC normalization matrix. The MEREC method uses proportional normalization, which is computationally independent of the maximum-minimum dimensionless normalization described in Section 2.2. Since all seven indices in this embodiment are inverse indices (smaller is better), the MEREC normalization formula is: (7) Equation (7), For the first The first class diagram MEREC normalized values ​​of each evaluation indicator; For the first Each indicator in all The minimum value of the quantity after translation in each class diagram; due to Therefore The denominator is strictly positive. If and only if hour , The larger the value, the higher the value. The class diagram in the first... The lower the relative complexity of each indicator.

[0035] S202. Calculate the global comprehensive performance values ​​of various graph types. Based on the MEREC normalized matrix. Calculate the first Each class diagram considers all Global comprehensive performance value for each evaluation metric : (8) In the formula, For the first The overall performance value of each class diagram ( ); The total number of evaluation indicators (in this embodiment, we take...) ); Determined by formula (7); due to Therefore Take the absolute value To ensure the summation term is non-negative; adding 1 to the outer layer and then taking the logarithm ensures... It also plays a certain role in compressing and regulating extreme values.

[0036] S203, Remove the first The performance value will be recalculated after the first indicator. One evaluation indicator was removed from the calculation, based on the remaining... The evaluation index is recalculated. Performance values ​​of class diagrams : (9) In the formula, To remove the first After the first indicator Recalculate performance values ​​for each class diagram; summation of indices. Traversing except the first All except for these indicators One indicator; Determined by formula (7). For all The above operations are performed sequentially for each indicator, requiring a total calculation. indivual value.

[0037] S204, Calculate the... The removal effect value of each indicator. For all The class diagram is accumulated and removed. The absolute difference of the change in the overall performance value of the first indicator before and after is obtained. The removal effect value of each indicator : (10) In the formula, The larger it is, the more likely it is to be the first The more significant the change in the overall evaluation result after an indicator is removed, the more important the indicator's impact on the comprehensive evaluation conclusion, and the higher its weight should be assigned.

[0038] S205, Normalization yields the final weights. For all... The removal effect value of each evaluation indicator After performing proportional normalization, the objective weights of each indicator are obtained. : (11) In the formula, For the first The MEREC objective weights of each evaluation indicator; the obtained weights satisfy the normalization constraint. ,and .

[0039] Will The weights of each indicator constitute a weight vector. The corresponding fuzzy matter-element weight matrix of preferred membership Represented as: (12) In the formula, To optimize the membership degree fuzzy matter-element weight matrix, the first... Line 1 The elements of the column are ;product This comprehensively reflects the characteristics of this type of diagram in the first... The weighted contribution and quality level of each indicator are the direct basis for constructing the subsequent decision matrix.

[0040] S3. Construct a weighted fuzzy complex element matrix based on the objective weights and the composite fuzzy element matrix.

[0041] After obtaining the weight vectors of each indicator With the preferred membership matrix Then, the two are combined to construct a weighted fuzzy complex element matrix for software quality characteristic evaluation. This will comprehensively display weight information and quality level information, and provide a reference basis for the CRADIS method.

[0042] set up The Line 1 Column elements Defined as the first The first class diagram Weighted membership degree of each evaluation indicator: (13) In the formula, For the first The first class diagram The weighted average score of each evaluation indicator ( ; ); The objective weights of MEREC are determined by formula (11); The degree of preferential membership is determined by formula (5). The larger the value, the more significant the effect. The class diagram in the first... The better the weighted quality contribution on each indicator, the lower the relative complexity.

[0043] From formula (13), the weighted fuzzy complex element matrix The complete form is represented as: (14) In formula (14), for The element-matrix of the weighted fuzzy complex is a matrix of order, where each element is a fuzzy complex. Determined by formula (13), it comprehensively reflects information on both weight contribution and quality level. It should be particularly noted that: Primarily used to visually display comprehensive quality information; in subsequent CRADIS ranking, to ensure that the weights have a substantial impact on the final ranking, the normalization step is based on the best membership degree. The weights are explicitly introduced through the weighted distance formula, which fundamentally avoids the problem of weights canceling each other out in the normalized numerator and denominator.

[0044] S4. Determine the overall utility of each evaluated item based on the weighted fuzzy complex meta-matrix.

[0045] Complete the weighted fuzzy complex meta-matrix After the construction is completed, further processing of all components is required. The class diagrams are comprehensively sorted to provide a final evaluation of the software quality.

[0046] The Ideal Solution Distance Compromise Ranking Method (CRADIS) is adopted, and a weight vector is explicitly introduced into the distance formula to ensure that the weights calculated by MEREC effectively affect the final ranking. CRADIS is a comprehensive improvement on the ARAS, MARCOS, and TOPSIS methods. It introduces the concept of a utility function and replaces the Euclidean distance with the sum of algebraic deviations, avoiding the nonlinear amplification effect of square root operations on subtle differences. The ranking results have good robustness when the set of solutions is expanded or reduced. The ranking calculation of the CRADIS method consists of six steps.

[0047] S401. Determine the input data. The preferred membership matrix obtained from formula (6) elements in ( ; As a direct input to the CRADIS normalization step, this design ensures that subsequent normalization does not introduce weights, making the weights uniquely effective only through the weighted distance step.

[0048] S402. Construct the CRADIS normalized decision matrix. Perform CRADIS-specific scaling normalization to obtain normalized values. .because As a positive indicator (the larger the better), the normalization formula is: (15) In the formula, For the first The first class diagram CRADIS normalized values ​​of each indicator; For the first Each indicator in all The maximum preferred membership degree in each class graph is used as the denominator for proportional normalization. ,and .

[0049] Boundary conditions: If a certain column , so that all of the columns Weight It is not involved in distance calculation.

[0050] S403. Determine the best and worst benchmarks based on the CRADIS normalization matrix. Determine the first Excellent benchmark values ​​for each indicator and worst-case reference value : (16) (17) In the formula, we know from formula (15) , representing the The excellent reference level of each indicator. In formula (17), , representing the The worst-case reference level for each indicator, the specific value of which depends on the actual data distribution of various charts. This constitutes the benchmark vector of excellence. and worst reference vector These serve as upper and lower reference benchmarks for measuring the quality performance of various types of diagrams.

[0051] S404. Calculate the weighted algebraic deviation between various graphs and the ideal solution. The key difference between CRADIS and TOPSIS lies in the use of algebraic summation instead of Euclidean square root distance; this paper further explicitly introduces a weight vector into the algebraic deviation. This ensures that the weights calculated by MEREC have a substantial impact on the ranking results.

[0052] Calculate the first Weighted algebraic deviation of each class diagram from the benchmark of excellence : (18) Calculate the first Weighted algebraic deviation of each class diagram from the worst-case benchmark : (19) In the formula, For the first The weighted total algebraic deviation of each class diagram from the benchmark of excellence ( ); As determined by formula (10), the weights have an explicit effect here; The smaller the value, the closer the graph is to the ideal level, and the better the software quality. In formula (19), The larger the value, the further the graph is from the worst solution, and the better the software quality. and It portrays the first from two complementary perspectives. The relative quality performance of each class diagram.

[0053] S405. Calculate the positive and negative utility functions. Let... For all The minimum weighted algebraic deviation from the benchmark in each class diagram. For all The maximum weighted algebraic deviation from the worst-case benchmark in each class diagram: (20) (twenty one) Based on the above reference standards, the positive utility function values ​​for each type of graph are calculated. With the inverse utility function value : (twenty two) (twenty three) In the formula, ,when hour This means that the diagram is closest to the benchmark of excellence among all candidate solutions.

[0054] In formula (23), ,when hour This means that the graph of this type is the furthest from the worst reference point among all candidate solutions.

[0055] S406. Calculate and rank the combined utility functions. Combining the positive and negative utility functions, calculate the... Overall utility of each class diagram : (twenty four) In the formula, For the first The overall utility of each category diagram is used to balance and integrate the quality performance in terms of "approaching the best benchmark" and "deviating from the worst benchmark". The larger the value, the more likely it is to be the first. The better the software quality (the lower the complexity) of each class diagram. Sort them from largest to smallest to get all of them. Ranking of software quality based on UML class diagrams.

[0056] S5. Based on the overall utility, rank the multiple evaluation items and output the ranking of software quality.

[0057] In the single-round process of MEREC weighting + CRADIS sorting, the weights are fixed once determined by MEREC, and subsequent sorting results do not affect the weights. However, when the initial MEREC weights lead to excessive concentration of the overall utility of various graph types (i.e., most graph types have similar scores and are difficult to distinguish), it indicates that the current weights lack discriminative power, and it is necessary to adjust the weights based on the sorting results. To address this, this paper introduces an adaptive feedback adjustment mechanism for weights after CRADIS sorting, forming an iterative closed loop of "MEREC weighting → CRADIS sorting → feedback adjustment of weights → CRADIS sorting again" until the sorting results converge.

[0058] S501. Calculate the utility dispersion coefficient to determine whether feedback is triggered. Let the first... The overall utility vector after rounds of iteration is Calculate its utility coefficient of variation. : (25) In the formula, For the first The standard deviation of the overall utility of the wheel; The mean; The smaller the value, the more concentrated the scores of each type of graph are, and the lower the sorting discrimination. When (Discrimination threshold, in this embodiment, is taken as...) When ), weight feedback correction is triggered; when If the current sorting resolution is sufficient, no correction is needed, and the result can be output directly.

[0059] S502. Calculate the distinguishing contribution of each indicator. For the positive and negative bias matrices in the current CRADIS results, calculate the... Differentiating contribution of each indicator : (26) In the formula, For the first Round The positive utility function value of each class diagram; Normalized value; and The first The best and worst benchmark values ​​for each indicator; To prevent numerically stable terms with a denominator of zero. The larger the value, the higher the value. The more significant an indicator's contribution to distinguishing different types of graphs is in the current round, the higher its weight should be assigned in the next round.

[0060] S503, Weight Feedback Update. Based on Distinguishing Contribution. Feedback correction to the weight vector: (27) (28) In the formula, To provide feedback on the step size parameter and control the weight update magnitude (in this embodiment, we take...), ); The larger the weight, the more sensitive the weight is to distinguish the contribution, but the convergence speed may be slower; The smaller the value, the more conservative the update, and the more stable the convergence. Formula (26) normalizes the updated weights to ensure... and It always holds true.

[0061] S504. Iteration Termination Criterion. The change in the infinite norm of the weight vector is used as the convergence criterion: (29) In the formula, The convergence threshold (in this embodiment, it is taken as...) ).

[0062] If formula (29) holds, the iteration terminates, using the comprehensive utility vector of the current round. As the final sorting criterion; otherwise, use replace Return to S402 to re-execute CRADIS normalization and subsequent steps, and begin the next step. Round iteration. In practice, this embodiment sets the maximum number of iterations to an upper limit of [number]. This is to prevent infinite loops from occurring under extreme data conditions.

[0063] Boundary conditions: If (That is, the combined utility of all class diagrams is zero, which usually does not occur), let And trigger feedback; if the weight of a certain indicator approaches zero after feedback correction ( If the value is not stable, then the indicator will be removed from subsequent calculations and renormalized to avoid numerical instability.

[0064] According to the eventual convergence Sort by size from largest to smallest to get all. Ranking of software quality based on UML class diagrams The larger the value, the higher the value. The better the software quality (the lower the complexity) of each class diagram, the better. Simultaneously, output the final iteration weights. This serves as an adaptive weighting result that reflects the inherent structure of the data.

[0065] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A software quality evaluation method based on fuzzy matter-element and MEREC weighting, characterized in that, Includes the following steps: S1. Obtain several UML class diagrams to be evaluated as the evaluation objects, and construct a composite fuzzy matter-element matrix; S2. Determine the objective weights of each evaluation index based on the composite fuzzy matter-element matrix; S3. Construct a weighted fuzzy complex element matrix based on the objective weights and the composite fuzzy matter element matrix; S4. Determine the overall utility of each evaluated item based on the weighted fuzzy complex meta-matrix; S5. Based on the overall utility, rank the multiple evaluation items and output the ranking of software quality.

2. The software quality evaluation method based on fuzzy matter-element and MEREC weighting as described in claim 1, characterized in that, S1 specifically includes: Obtain several UML class diagrams to be evaluated as the evaluation items; Using several evaluation indicators as evaluation features, the original values ​​of each evaluated item on each evaluation indicator are statistically analyzed. Construct a composite fuzzy matter-element matrix based on the original values; In this system, the rows of the composite fuzzy matter element correspond to each evaluated object, the columns correspond to each evaluated index, and the matrix elements are the original values ​​of each object on each index.

3. The software quality evaluation method based on fuzzy matter-element and MEREC weighting according to claim 1, characterized in that, S2 specifically includes: The original values ​​are preprocessed by translation to obtain the translated values. Construct a MEREC normalized matrix based on the translated values; calculate the global comprehensive performance value of each evaluated entity based on the MEREC normalized matrix; After removing each evaluation indicator one by one, the performance value is recalculated, and the removal effect value of each indicator is calculated based on the recalculated performance value and the overall performance value. The objective weights of each evaluation index are obtained by normalizing the removal effect value.

4. The software quality evaluation method based on fuzzy matter-element and MEREC weighting according to claim 1, characterized in that, S3 specifically includes: multiplying the objective weight of each evaluation index with the corresponding preferential membership degree, and using the product as a weighting element to form a weighted fuzzy complex element matrix.

5. The software quality evaluation method based on fuzzy matter-element and MEREC weighting according to claim 1, characterized in that, S4 specifically includes: Obtain the preferred membership matrix from the composite fuzzy matter-element matrix; construct the CRADIS normalized decision matrix from the preferred membership matrix; The best and worst benchmark values ​​are determined based on the CRADIS normalized decision matrix; Based on the excellent benchmark value, the worst benchmark value, the CRADIS normalized value, and the objective weights, calculate the weighted algebraic deviation of each evaluated item from the excellent benchmark value and the weighted algebraic deviation from the worst benchmark value. Calculate the positive and negative utility function values ​​based on the weighted algebraic deviation; calculate the overall utility value based on the positive and negative utility function values.

6. The software quality evaluation method based on fuzzy matter-element and MEREC weighting according to claim 1, characterized in that, Specifically, S5 includes: arranging the evaluation items in descending order of overall utility, with the evaluation item with the highest overall utility corresponding to the best software quality.

7. The software quality evaluation method based on fuzzy matter-element and MEREC weighting according to claim 3, characterized in that, The translation preprocessing involves adding a preset positive number to each original value.

8. The software quality evaluation method based on fuzzy matter-element and MEREC weighting according to claim 3, characterized in that, The method for constructing the MEREC normalization matrix includes: for each evaluation index, using the minimum value of the index after shifting among all evaluation items as the denominator, and using the shifted value of each evaluation item as the numerator, calculating the proportion value, and the proportion values ​​constitute the MEREC normalization matrix.

Images