Product design scheme evaluation method and system combining customer preferences and expert knowledge

Abstract

This invention provides a product design scheme evaluation method and system that integrates customer preferences and expert knowledge, relating to the technical field of product design evaluation. This invention proposes a new comprehensive scheme evaluation method that incorporates the synergy between parameters into the scheme evaluation system based on fuzzy integrals, correcting the bias of evaluation methods based on indicator additivity. This provides decision support for enterprises to optimize design schemes and develop new products. Furthermore, this invention comprehensively considers customer preferences and expert knowledge, resulting in a more systematic and rational scheme evaluation system.

Description

Technical Field

[0001] The present invention relates to the technical field of product design evaluation, and in particular to a product design scheme evaluation method and system combining customer preferences and expert knowledge. Background Art

[0002] Growing pressure from the global economy compels manufacturers to design products with short lead times, low costs, and high quality. During the product design phase, companies often prepare multiple alternative designs and select the best one to finalize. This process impacts product development cycles and performance. Therefore, accurately and systematically evaluating and ranking these alternatives has become a crucial task in product design. Advanced design evaluation methods can not only help companies quickly finalize design solutions and improve product performance, but also maximize customer satisfaction.

[0003] Current design scheme evaluation methods are primarily divided into those based on an indicator system and those based on feature learning. The core of indicator-based scheme evaluation is determining the evaluation system and the weights of each evaluation metric. Therefore, it involves two steps: first, designing a scheme evaluation system based on customer needs and preferences; second, determining the weights of the evaluation metrics and aggregating the performance of each metric to evaluate the design scheme. In the first step, methods such as quality function deployment, house of quality, niche theory, and dual-objective soft sets are used to analyze customer preferences and determine the evaluation system. In the second step, methods such as TOPSIS, adaptive conjoint analysis, and VIKOR are used to calculate the weights of each metric. Feature learning-based methods have emerged in recent years. Their core concept is to select appropriate evaluation features and predict scheme evaluation values ​​based on machine learning. The most commonly used model is a neural network.

[0004] However, both indicator-based and feature-learning-based scheme evaluation methods ignore the synergistic effects between indicators and evaluate schemes based on their additive nature. With economic development and increasing user demand, products are becoming increasingly complex. Product design parameters intersect and influence each other. This synergistic effect of design parameters results in indicators no longer being additive, making traditional scheme evaluation methods irrational and biased. Summary of the Invention

[0005] (1) Technical problems solved

[0006] In response to the deficiencies of the prior art, the present invention provides a product design scheme evaluation method and system that combines customer preferences and expert knowledge, solving the problem of how to improve the evaluation accuracy of product design schemes.

[0007] (2) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] In a first aspect, a product design scheme evaluation method combining customer preferences and expert knowledge is provided, the method comprising:

[0010] Obtain product parameter data and review data; the review data includes review text and star rating;

[0011] Constructing a design parameter evaluation function based on the parameter data and the comment data; the design parameter evaluation function is used to convert the design parameter values ​​of the design scheme into design parameter evaluation values;

[0012] Obtain independent weights for each design parameter based on design parameter evaluation values, star ratings, and artificial neural networks;

[0013] The independent weights of design parameters and their synergy are integrated to calculate the joint weight of the parameter alliance;

[0014] Based on the design parameter evaluation function and joint weight, a comprehensive evaluation is conducted on the evaluation schemes.

[0015] Furthermore, the acquisition of product parameter data and review data also includes:

[0016] For parameter data, digitize text-type design parameters;

[0017] For the comment data, data cleaning, sentence segmentation, word segmentation and part-of-speech tagging are performed.

[0018] Furthermore, constructing a design parameter evaluation function based on the parameter data and the comment data includes:

[0019] Extract characteristic opinion pairs from the review text and classify them into corresponding design parameters;

[0020] Calculating the semantic tendency of the characteristic viewpoint pair, and calculating the design parameter evaluation value based on the semantic tendency;

[0021] For different types of design parameters, corresponding design parameter evaluation functions are constructed based on the design parameter evaluation values; the types of design parameters include discrete design parameters and continuous design parameters.

[0022] Furthermore, the calculating of the design parameter evaluation value based on the semantic tendency includes:

[0023]

[0024] in, The gth value of the jth design parameter of design parameter i i,jCharacteristic viewpoint pairs;

[0025] Indicates the semantic tendency of the feature viewpoint pair;

[0026] N i,j Indicates fo i,j the number of

[0027] And for different types of design parameters, corresponding design parameter evaluation functions are constructed based on the design parameter evaluation values, including:

[0028] The design parameter evaluation function of discrete design parameters is modeled using piecewise functions;

[0029] The design parameter evaluation function of continuous design parameters is modeled using fitting theory.

[0030] Furthermore, the independent weight of each design parameter is obtained based on the design parameter evaluation value, star rating and artificial neural network, including:

[0031] Using the design parameter evaluation value as input and the star rating as output, an artificial neural network is constructed and trained;

[0032] The total weight of the design parameters in the artificial neural network when explaining the output is used as the independent weight of the design parameters.

[0033] Furthermore, the total weight of the design parameter interpretation output in the artificial neural network is used as the independent weight of the design parameter, including:

[0034] The absolute values ​​of all connection weights of the connection attributes from the input layer to the hidden layer to the output layer are summed and normalized; and the calculation formula of the independent weight of the design parameter i is:

[0035]

[0036] Among them, ind_w i represents the independent weight of design parameter i;

[0037] |w inp，hid | represents the absolute value of the connection weight between the input layer neuron inp and the hidden layer neuron hid;

[0038] |w hid，out | represents the absolute value of the connection weight between the hidden layer neuron hid and the output layer neuron out;

[0039] INP and HID represent the input layer neuron set and the hidden layer neuron set, respectively.

[0040] Furthermore, the independent weights of the fusion design parameters and the joint weight of the synergistic calculation parameter alliance include:

[0041] Obtain the synergy evaluation of the parameter alliance by each expert based on the synergy degree label;

[0042] Aggregate the evaluation results of each expert and obtain the comprehensive evaluation results of the synergy of each parameter alliance based on the weighted aggregation results;

[0043] The joint weight of the parameter alliance is calculated based on the fuzzy measure fusion of independent weights and synergy.

[0044] Furthermore, the joint weight of the parameter alliance calculated based on the fuzzy measure fusion independent weights and synergy includes:

[0045] The independent weight is taken as the weight μ(A p ), p = 1;

[0046] For parameter alliance A of p≥2 p , whose joint weight μ(A p ) is calculated as:

[0047] μ(A p )=μ(X)+μ(Y)+λμ(X)μ(Y)

[0048] in,

[0049] μ(X)=max(μ(X g ))

[0050] μ(Y)=μ(A p -X g )

[0051]

[0052]

[0053]

[0054] λ max =abs(λ min )

[0055]

[0056] in,

[0057] represents the aggregated results of the synergy assessment results of each expert;

[0058] n j Represents the parameter alliance A pLabel the degree of collaboration the number of experts;

[0059] represents the jth collaborative tag, j = 1, 2, ... J;

[0060] J represents the number of collaborative tag types;

[0061] Indicates the comprehensive evaluation result after weighted averaging of the aggregation results;

[0062] n j X g ∈A p , g = p-1;

[0063] For μ(A1) and {μ(A p ), p=2,…,K} for normalization.

[0064] Furthermore, the comprehensive evaluation of the scheme to be evaluated based on the design parameter evaluation function and the joint weight includes:

[0065] Convert the design parameters of the scheme to be evaluated into design parameter evaluation values ​​and arrange them in order from small to large;

[0066] The scheme S to be evaluated is comprehensively evaluated using the generalized Choquet fuzzy integral combined with the joint weight.

[0067] The comprehensive evaluation calculation method of the scheme to be evaluated is:

[0068]

[0069] Among them, mr (i) represents the evaluation value of the i-th design parameter arranged from small to large;

[0070] A (i) ={dp (i) , dp (i+1) ,…,dp (k)}, dp (i) represents the design parameter corresponding to the i-th design parameter evaluation value in the sequence; and μ(A (K+1) )=0.

[0071] Secondly, a product design scheme evaluation system combining customer preferences and expert knowledge is provided, including:

[0072] one or more processors;

[0073] Memory; and

[0074] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including a method for executing the above-mentioned product design solution evaluation method combining customer preferences and expert knowledge.

[0075] (3) Beneficial effects

[0076] The present invention provides a product design solution evaluation method and system that combines customer preferences and expert knowledge. Compared with existing technologies, it has the following advantages:

[0077] This paper proposes a new comprehensive solution evaluation method that incorporates the synergy between parameters into the solution evaluation system based on fuzzy integrals. This method corrects the bias of evaluation methods based on indicator additivity, providing decision support for companies to optimize design solutions and develop new products. Furthermore, this paper comprehensively considers customer preferences and expert knowledge, resulting in a more systematic and rational solution evaluation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0079] Figure 1 is a flow chart of an embodiment of the present invention;

[0080] Figure 2 Schematic diagram of the design parameter evaluation function of NEDC comprehensive fuel consumption according to an embodiment of the present invention. DETAILED DESCRIPTION

[0081] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0082] The embodiments of the present application solve the problem of improving the evaluation accuracy of product design solutions by providing a product design solution evaluation method and system that combines customer preferences and expert knowledge.

[0083] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0084] Example 1:

[0085] like Figure 1 As shown, the present invention provides a product design scheme evaluation method combining customer preferences and expert knowledge, the method comprising:

[0086] Obtain product parameter data and review data; the review data includes review text and star rating;

[0087] Constructing a design parameter evaluation function based on the parameter data and the comment data; the design parameter evaluation function is used to convert the design parameter values ​​of the design scheme into design parameter evaluation values;

[0088] Obtain independent weights for each design parameter based on design parameter evaluation values, star ratings, and artificial neural networks;

[0089] The independent weights of design parameters and their synergy are integrated to calculate the joint weight of the parameter alliance;

[0090] Based on the design parameter evaluation function and joint weight, a comprehensive evaluation is conducted on the evaluation schemes.

[0091] The beneficial effects of this embodiment are:

[0092] This embodiment of the present invention proposes a new comprehensive solution evaluation method. This method, based on fuzzy integrals, incorporates the synergy between parameters into the solution evaluation system. This method corrects the bias of evaluation methods based on indicator additivity, providing decision support for enterprises to optimize design solutions and develop new products. Furthermore, this invention comprehensively considers customer preferences and expert knowledge, resulting in a more systematic and rational solution evaluation system.

[0093] The implementation process of the embodiment of the present invention is described in detail below:

[0094] S1. Obtain product parameter data and review data.

[0095] Specifically, it includes two steps: data collection and preprocessing.

[0096] S1.1. Data collection specifically includes: First, we crawl parameter data and review data (including review text and star ratings) of different product models from vertical websites. Review text refers to the textual opinions expressed by customers about the product, and star ratings refer to the customer's overall satisfaction rating for a particular product model.

[0097] In the specific implementation, you can write crawler code based on Python, extract and parse web page data through XPath and CSS expressions, and use Navicat 15 database as a data repository to crawl parameter data and comment data of products in vertical websites.

[0098] The dimension of parameter data is the dimension that customers are concerned about. By counting K frequently appearing product features in the reviews, and based on professional knowledge, the product features are mapped to design parameters.

[0099] {dp} k ={dp1, ..., dp k} represents the design parameter set;

[0100] dp i represents the i-th design parameter;

[0101] dp i，j represents the j-th parameter value of design parameter i;

[0102] Indicates the parameter data of the cth product crawled;

[0103] Represents the corresponding parameter value of design parameter i of product c.

[0104] For example:

[0105] Taking automobiles as an example, we collected review data and parameter data from the Autohome website for 137 gasoline-only vehicles priced between 80,000 and 130,000 yuan in 2019 and 2020. These gasoline-only vehicles included microcars, small cars, compact cars, mid-size cars, mid-to-large cars, and large cars. We then analyzed nine popular product features from customer reviews and mapped these features to design parameters based on our expertise, identifying key product design parameters for subsequent analysis.

[0106] That is, {dp}9 = {dp1, ..., dp9}, and dp1 to dp9 respectively represent: NEDC comprehensive fuel consumption, wheelbase, engine maximum torque, seat material, rear suspension type, central control screen size, parking brake system, engine maximum horsepower and number of speakers.

[0107] S1.2 Data Preprocessing. Further process the crawled data to obtain high-quality and valid data. For parameter data, digitize the text-based design parameters to facilitate subsequent calculations. For comment data, filter out valuable comment texts, perform sentence segmentation, word segmentation, and part-of-speech tagging to facilitate subsequent feature point pair extraction and sentiment analysis.

[0108] S1.2.1. Parameter data preprocessing.

[0109] Numerical design parameters (such as "Manufacturer's Suggested Price") are easier to calculate, while text-based design parameters (such as "Seat Material") are more difficult to calculate. Therefore, it's necessary to convert text-based parameters into numerical ones to facilitate subsequent calculations. Different numerical conversion methods are used for different text-based parameters.

[0110] In specific implementation, for design parameters with different values, the parameter values ​​can be sorted by quality and converted into integer data. Alternatively, design parameters with only two parameter values ​​can be converted into logical data (the data field only contains 0 and 1), etc.

[0111] For example:

[0112] For example, the design parameter "power steering system" contains three parameter values ​​of "mechanical hydraulic power steering", "electronic hydraulic power steering" and "electric power steering". The parameter values ​​are converted into {0, 1, 2} respectively according to the order of merit (mechanical hydraulic power steering < electronic hydraulic power steering < electric power steering).

[0113] S1.2.2. Comment data preprocessing.

[0114] The preprocessing of review data mainly focuses on the review text.

[0115] First, remove ineffective reviews that lack value. These reviews often contain only symbols, numbers, or gibberish, such as "jjjjj5.11 4.76 180 37.82." Ineffective reviews lack customer opinions and sentiment, and therefore fail to reflect customer preferences, so they need to be deleted.

[0116] Secondly, the valuable comments are divided into clauses, because clauses are the basic analysis units when extracting characteristic viewpoints in the present invention.

[0117] In specific implementation, we can first build a set of punctuation marks covering both Chinese and English, such as "。.", ",," "::", etc. Then, based on the punctuation mark set, we use regular expressions to cut long text comments into a set of short text clauses.

[0118] Finally, the segmented clauses are segmented and tagged with parts of speech, which is the basis for subsequent domain dictionary construction, opinion extraction and sentiment analysis.

[0119] In the specific implementation, the jieba library in Python can be called to achieve word segmentation and part-of-speech tagging of comment clauses.

[0120] S2. Construct a design parameter evaluation function.

[0121] The market response to a new design proposal needs to be determined by weighting the design parameter evaluation values. A design parameter evaluation function is constructed for each design parameter. Its function is to convert the design parameter values ​​of the design proposal into a design parameter evaluation value, which serves as the decision-making basis for proposal evaluation. The essence of the design parameter evaluation function is to fit the mapping relationship between the design parameter values ​​and the design parameter evaluation values. Therefore, it is necessary to first extract characteristic viewpoints from the review data and classify them based on the design parameters. Then, based on these characteristic viewpoints, sentiment analysis is used to mine the average market evaluation value of the design parameters. Finally, an appropriate function form is selected to construct the mapping relationship.

[0122] S2.1. Extract characteristic viewpoint pairs and classify them into corresponding design parameters.

[0123] Online reviews contain rich feature-opinion pairs that show customers’ evaluations of various features (i.e., design parameters).

[0124] For example, the feature-opinion pair (fo) <shock absorption, strong> consists of the feature (fea) "shock absorption" (i.e., the design attribute "suspension") and the opinion (opi) "strong." In other cases, opinions also include intensifiers (int, such as "very") and negators (neg, such as "not").

[0125] Obviously, there is an indirect relationship between the feature viewpoint pair and the design parameter. Based on the domain knowledge, the present invention constructs the design parameter feature word set according to the one-to-one correspondence between the design parameter and the feature word. Among them, fea i，nf Indicates the nfth feature word mapped to design parameter i, NF i Represents the number of feature words mapped to design parameter i. In this way, the feature viewpoint pairs can be classified according to the design parameter feature word set.

[0126] For example:

[0127] Taking the “NEDC comprehensive fuel consumption” design parameter as an example, its design parameter feature word set is {oil, fuel consumption}.

[0128] However, due to the richness of Chinese vocabulary and the complexity of its grammar, it is impossible to directly generate feature-opinion pairs.

[0129] First, the feature dictionary, opinion dictionary, degree dictionary and negation dictionary are manually constructed.

[0130] In the specific implementation, the feature dictionary consists of common nouns related to products in reviews; the last three use the public Chinese vocabulary HOWNET.

[0131] Secondly, the present invention summarizes a series of grammatical patterns of characteristic opinion pairs as shown in Table 1. Subsequently, by locating characteristic words and searching backward / forward for opinions containing intensifiers and negation words from the position of the characteristic words according to the dictionary and grammatical patterns, characteristic opinion pairs are automatically generated.

[0132] In the specific implementation, we first locate the position of the feature word in the clause according to the feature dictionary, and then search for opinion words, negative words and degree adverbs in the detection window of six words before and after the feature word according to the grammatical pattern of the feature opinion pair shown in Table 1 to form a feature opinion pair; if we encounter other feature words or the beginning and end of the clause or punctuation marks, the search ends.

[0133] Table 1. Grammatical patterns of feature viewpoint pairs

[0134]

[0135] S2.2. Calculate the evaluation values ​​of the design parameters.

[0136] Based on sentiment analysis, the semantic tendency of feature opinion pairs is calculated to extract the evaluation value of design parameters. Sentiment analysis is mainly divided into two types: (1) polarity-based sentiment analysis and (2) intensity-based sentiment analysis. Compared with the former, the latter has a better ability to identify the position and attitude of the reviewer. We follow the latter sentiment analysis and use a dictionary annotated with the semantic polarity and semantic intensity of words.

[0137] The core of dictionary-based sentiment analysis is to assign a semantic orientation value to each word. In the feature dictionary, benefit-related features have a semantic orientation value of +1, while cost-related features have a semantic orientation value of -1. The semantic orientation values ​​in the opinion dictionary range from -5 to +5.

[0138] It's important to note that some opinion words exhibit semantic ambiguity. For example, the opinion word "high" exhibits a negative semantic orientation when paired with the characteristic word "price," while it exhibits a positive semantic orientation when paired with the characteristic word "height." WSA(opi) = 1 indicates an opinion word with semantic ambiguity, while WSA(opi) = 0 indicates an opinion word without semantic ambiguity.

[0139] In the intensifier dictionary, each intensifier has a percentage associated with it. For example, the degree word "most" has a semantic tendency value of 100%, while the degree word "slightly" has a semantic tendency value of -10%. In the negation dictionary, semantic tendency values ​​range from +1 to +5.

[0140] Then, the semantic tendency value of the feature opinion pair is calculated by the following formula:

[0141]

[0142] Wherein:

[0143] SO(fo) represents the semantic tendency value of the feature-opinion pair fo;

[0144]

[0145] F, O, I, and N respectively represent the semantic tendencies of the feature word fea, opinion word opi, intensity word int, and negation word neg.

[0146] Illustrate with an example:

[0147] Take the feature-opinion pair <fuel consumption, very, high> as an example.

[0148] Among them, "fuel consumption" belongs to the cost-type feature, and its semantic tendency is "-1".

[0149] The semantic tendency of "high" is "3"; "high" has semantic ambiguity (WSA(high) = 1), and it is paired with the cost-type feature "fuel consumption" (SO(fuel consumption) = -1), so SWI = -1.

[0150] "Very" is the adverb of degree describing "high" in this sentence, and its intensification degree is "1".

[0151] Calculate its sentiment tendency to be "-6" (-6 = -1×[3×(1 + 1)]).

[0152] Next, the evaluation value of each design parameter value is calculated through the following formula:

[0153]

[0154] Wherein, represents the g i，j th feature-opinion pair of the jth design parameter value of the design parameter i, and N i，j represents the quantity of fo i，j .

[0155] S2.3, Construction of the design parameter evaluation function.

[0156] Due to the complexity of the product, the product often includes two types of design parameters, namely discrete design parameters (such as "seat material") and continuous design parameters (such as "fuel consumption"). Therefore, the present invention adopts two different methods to construct the design parameter evaluation function for discrete design parameters and continuous design parameters respectively.

[0157] Since the values of discrete design parameters are limited, the evaluations of all design parameter values by customers can be mined from a large number of online reviews. Therefore, based on the design parameter evaluation values calculated in S2.2, a piecewise function is used to model the design parameter evaluation function for discrete design parameters:

[0158]

[0159] In contrast, continuous design parameters have infinite parameter values. Therefore, we use fitting theory to model the design parameter evaluation function to predict those design parameter evaluation values ​​that cannot be mined from online reviews. However, the type of function is difficult to determine. For each continuous design parameter, the one with the best fitting effect is selected from the preparatory function types such as linear, quadratic, cubic, logarithmic, growth curve and sigmoid function as its fitting function type. The present invention uses the coefficient of determination (R 2 ) to evaluate the fitting effect. When the coefficient of determination of the models is the same, the Akaike Information Criterion (AIC) of the models is further compared. This can determine the continuous design parameter dp i The design parameter evaluation function is CF i (dp i )express.

[0160] In this way, the design parameter evaluation value of the design parameter can be obtained by the following formula:

[0161]

[0162] For example:

[0163] Taking the design parameter "NEDC comprehensive fuel consumption" as an example, it is a continuous design parameter. Therefore, the linear, quadratic, cubic, logarithmic, growth curve and sigmoid functions are used to construct the design parameter evaluation function of "NEDC comprehensive fuel consumption", and the index R is used. 2 The fitting effect is evaluated by AIC. After comparison, the cubic curve has the best fitting effect. Therefore, the design parameter evaluation function of NEDC comprehensive fuel consumption is as follows: Figure 2 As shown, its function expression is:

[0164]

[0165] S3. Based on the design parameter evaluation value, star rating and artificial neural network, the independent weight of each design parameter is obtained.

[0166] That is, an artificial neural network (ANN) is constructed, with the design parameter evaluation value as input and the star rating as output, and the nonlinear relationship between the design parameter evaluation value and the star rating is fitted to obtain the independent weight of each design parameter.

[0167] S3.1. Build and train artificial neural networks (ANNs).

[0168] Technically, an ANN consists of an input layer, a hidden layer, and an output layer. Design parameter evaluation values ​​are assigned as input layer neurons, and star ratings are assigned as output layer neurons. The number of hidden layer neurons is equal to 2 / 3 (n_out + n_inp), where n_inp represents the number of input layer neurons and n_out represents the number of output layer neurons. Furthermore, the ReLU function is selected as the activation function for the hidden layer. Finally, the root mean square error (RMSE) is used to evaluate the performance of the BPNN:

[0169]

[0170] in, Indicates the predicted value of the product's c-star rating;

[0171] or c Indicates the true value of the product's c-star rating;

[0172] C represents the amount of data used for training.

[0173] For example:

[0174] Step S1.1 determines nine key product design parameters, so the number of input layer neurons is 9; the star rating is used as an output layer neuron, so the number of output layer neurons is 1. Therefore, the number of hidden layer neurons can be determined to be 7 (2 / 3(9+1)=20 / 3≈7).

[0175] S3.2. Calculate the independent weights of the design parameters.

[0176] The total weight of the design parameters in the BPNN when explaining the output is taken as the independent weight of the design parameters. In order to obtain the total weight when the attribute explains the output, the absolute values ​​of the connection weights of the connection attributes from the input layer to the hidden layer to the output layer are summed and normalized:

[0177]

[0178] Among them, ind_w i represents the independent weight of design parameter i;

[0179] |w inp，hid | represents the connection weight between the input layer neuron inp and the hidden layer neuron hid;

[0180] |w hid，out | represents the connection weight between the hidden layer neuron hid and the output layer neuron out;

[0181] INP and HID represent the input layer neuron set and the hidden layer neuron set, respectively.

[0182] For example:

[0183] The connection weights between the input layer and hidden layer of the ANN trained in step S3.1 are as follows:

[0184]

[0185] The connection weights between the hidden layer and the output layer are as follows:

[0186] {-0.0133, -0.0327, -0.0016, 0.0398, -0.0257, 0.4589, 0.4898}

[0187] The independent weight of each design parameter is calculated as follows:

[0188] {ind_w}9=

[0189] {0.1157, 0.1335, 0.1058, 0.0917, 0.0888, 0.1234, 0.1111, 0.0959}.

[0190] S4. Calculate the joint weight of the parameter alliance based on fuzzy measure fusion independent weights and synergy

[0191] S4.1. Obtain the degree of coordination of the parameter alliance.

[0192] First, determine the collaboration level label

[0193] Then, expert e will classify the parameter alliance based on the collaboration degree label. Evaluate the synergy Where p = 2, 3, ..., K represents the number of parameters in the alliance.

[0194] It should be noted that this patent only considers the synergy between two parameters, namely A2={dp i , dp j Only the obvious degree of synergy is necessary; the rest are considered additive.

[0195] Then aggregate the evaluation results of each expert, where n j Represents the parameter alliance A p Label the degree of collaboration number of experts.

[0196] Finally, the weighted average of the aggregation results is performed, and the formula is as follows:

[0197]

[0198] S4.2. Based on the independent weights, the degree of fusion coordination is continuously iteratively calculated to form the joint weights of the higher-dimensional parameter alliance.

[0199] Synergistic effect is defined in this patent as [λ min ,λ max ] range. min It is derived from the monotonicity of fuzzy integral, that is, it is determined by the two rules μ(X+Y)>μ(X) and μ(X+Y)>μ(Y). max For the purpose of simplifying the actual situation, considering that there are very few extreme positive interactions in reality, λ is limited to a reasonable range to facilitate calculation.

[0200] The independent weights {ind_w} obtained in step S3.2 are k As the joint weight μ(A p )(p=1).

[0201] For a parameter alliance with p≥2, the weight is calculated as follows:

[0202] μ(X)=max(μ(X g ))

[0203] μ(Y)=μ(A p -X g )

[0204]

[0205] λ max =abs(λ min )

[0206]

[0207]

[0208] μ(A p )=μ(X)+μ(Y)+λμ(X)μ(Y)

[0209] Among them, X g ∈A p , g=p-1. When g=1, X g =dp i . Treat the independent weights as special joint weights and {μ(A p ), p=2,…,K} are normalized together.

[0210] For example:

[0211] Set the collaboration level label to:

[0212] Wcom ={irrelevant, very weak, weak, moderate, strong, very strong}

[0213] And numerically label the degree of coordination:

[0214] If the degree of synergy is positively correlated, then:

[0215]

[0216] If the degree of synergy is negatively correlated, then:

[0217]

[0218] Four experts evaluated the degree of coordination of the parameter alliance, aggregated the evaluation results of each expert, and finally weighted the aggregation results to obtain a comprehensive evaluation result of the degree of coordination, as shown in Table 2 below.

[0219] Table 2. Evaluation results of parameter alliance coordination degree

[0220]

[0221] The independent weights of design parameters and their synergy are integrated to calculate the joint weight of the parameter alliance;

[0222] The joint weights of the partial parameter alliance are as follows:

[0223] A2({rear suspension type, NEDC comprehensive fuel consumption})=0.21592565;

[0224] A3({rear suspension type, NEDC combined fuel consumption, wheelbase})=0.39484364;

[0225] A4({rear suspension type, NEDC comprehensive fuel consumption, wheelbase, center console screen size})=0.50101055;

[0226] And normalize the joint weight and independent weight together. The partially normalized weights are as follows:

[0227] A1({NEDC comprehensive fuel consumption})=0.10146678;

[0228] A2({rear suspension type, NEDC comprehensive fuel consumption})=0.19811254;

[0229] A3({rear suspension type, NEDC combined fuel consumption, wheelbase})=0.36227042;

[0230] A4({rear suspension type, NEDC comprehensive fuel consumption, wheelbase, central control screen size})=0.45967894.

[0231] S5. Based on the constructed design parameter evaluation function and joint weight, the generalized Choquet fuzzy integral is used to conduct a comprehensive evaluation of the evaluation scheme S.

[0232] The generalized Choquet fuzzy integral is a more general measure than the classical additivity measure, which uses μ to handle non-additive complex relationships between criteria (redundancy, synergy, etc.). Therefore, by inputting the valuation of each criterion, it not only considers their respective importance but also the degree of possible interaction.

[0233] When implementing:

[0234] First, the constructed design parameter evaluation function is used to convert the design parameters of the scheme to be evaluated into design parameter evaluation values, and then the values ​​are arranged in order from small to large to obtain a new sequence:

[0235] mr (1) ≤mr (2) ≤…≤mr (K)

[0236] Combined with the joint weight, the generalized Choquet fuzzy integral is used to conduct a comprehensive evaluation of the evaluation scheme S.

[0237] The calculation formula is as follows:

[0238]

[0239] Among them, A (i) ={dp (i) , dp (i+1) ,…,dp (k)} μ(A (K+1) )=0.

[0240] For example:

[0241] Suppose there are two solutions S to be evaluated about cars 1 and S 2 , the values ​​of its nine design parameters are:

[0242]

[0243]

[0244] Use the design parameter evaluation function to convert the design parameters into design parameter evaluation values:

[0245]

[0246]

[0247] Combined with the joint weight of step S4.2, the generalized Choquet fuzzy integral calculation can be obtained, and the solution S 1 The score is 1.9322, and the solution S 2 The score of S is 2.1983. 2 Better than plan S 1 .

[0248] Example 2:

[0249] A product design scheme evaluation system combining customer preferences and expert knowledge, the system comprising:

[0250] one or more processors;

[0251] Memory; and

[0252] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs comprising steps for performing the following steps:

[0253] Obtain product parameter data and review data; the review data includes review text and star rating;

[0254] Constructing a design parameter evaluation function based on the parameter data and the comment data; the design parameter evaluation function is used to convert the design parameter values ​​of the design scheme into design parameter evaluation values;

[0255] Obtain independent weights for each design parameter based on design parameter evaluation values, star ratings, and neural networks;

[0256] The synergy between design parameters is integrated to calculate the joint weight of parameter alliance;

[0257] Based on the design parameter evaluation function and joint weight, a comprehensive evaluation is conducted on the evaluation schemes.

[0258] It can be understood that the product design scheme evaluation system based on combined customer preferences and expert knowledge provided in the embodiment of the present invention corresponds to the above-mentioned product design scheme evaluation method based on combined customer preferences and expert knowledge. The explanation, examples, beneficial effects, etc. of its relevant contents can refer to the corresponding contents in the product design scheme evaluation method based on combined customer preferences and expert knowledge, and will not be repeated here.

[0259] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0260] 1) This paper proposes a new comprehensive solution evaluation method. This method, based on fuzzy integrals, incorporates the synergy between parameters into the solution evaluation system, correcting the bias of evaluation methods based on indicator additivity. This provides decision support for companies to optimize design solutions and develop new products. Furthermore, this paper comprehensively considers customer preferences and expert knowledge, resulting in a more systematic and rational solution evaluation system.

[0261] 2) Discrete and continuous design parameter evaluation functions are constructed by considering the different characteristics of design parameters; product evaluation comes from a large amount of review data on the Internet, which reduces the cost of customer opinion mining and the randomness of small samples.

[0262] 3) The weights are determined using a data-driven approach, with high-precision fitting results used to calculate the weights of each design parameter.

[0263] It should be noted that, through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiment. In this article, relational terms such as first and second are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. Without further constraints, an element defined by the phrase "comprises a..." does not preclude the existence of additional identical elements in the process, method, article or apparatus that includes the element.

[0264] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A product design scheme evaluation method combining customer preferences and expert knowledge, characterized by: The method includes: Obtain product parameter data and review data; the review data includes review text and star rating; Constructing a design parameter evaluation function based on the parameter data and the comment data; the design parameter evaluation function is used to convert the design parameter values ​​of the design scheme into design parameter evaluation values; Obtain independent weights for each design parameter based on design parameter evaluation values, star ratings, and artificial neural networks; The independent weights of design parameters and their synergy are integrated to calculate the joint weight of the parameter alliance; Based on the design parameter evaluation function and joint weight, a comprehensive evaluation is conducted on the evaluation schemes; The independent weights of the fusion design parameters and the joint weights of the synergistic calculation parameters alliance include: Obtain the synergy evaluation of the parameter alliance by each expert based on the synergy degree label; Aggregate the evaluation results of each expert and obtain the comprehensive evaluation results of the synergy of each parameter alliance based on the weighted aggregation results; The joint weight of the parameter alliance is calculated based on the fuzzy measure fusion of independent weights and synergy; The method of calculating the joint weight of the parameter alliance based on the fuzzy measure fusion independent weights and synergy includes: Treat the independent weight as the weight of the parameter union containing only one parameter , p =1; for Parameter Union , its joint weight The calculation formula is: in, in, represents the aggregated results of the synergy assessment results of each expert; Represents a parameter union Label the degree of collaboration the number of experts; Indicates the j Collaborative tags, ; J Indicates the number of collaborative tag types; Indicates the comprehensive evaluation result after weighted averaging of the aggregation results; ， ； right and Perform normalization; The comprehensive evaluation of the scheme to be evaluated based on the design parameter evaluation function and the joint weight includes: Convert the design parameters of the scheme to be evaluated into design parameter evaluation values ​​and arrange them in order from small to large; Combined with joint weights, the generalized Choquet fuzzy integral is used to evaluate the scheme. Conduct comprehensive evaluation; The comprehensive evaluation calculation method of the scheme to be evaluated is: in, Indicates the order from small to large i Evaluation values ​​of design parameters; , Indicates the first i The design parameter corresponding to the design parameter evaluation value; and , .

2. The product design scheme evaluation method combining customer preferences and expert knowledge as claimed in claim 1, characterized in that: The acquisition of product parameter data and review data also includes: For parameter data, digitize text-type design parameters; For the comment data, data cleaning, sentence segmentation, word segmentation and part-of-speech tagging are performed.

3. The product design scheme evaluation method combining customer preferences and expert knowledge as claimed in claim 1, characterized in that: The constructing of a design parameter evaluation function based on the parameter data and the comment data includes: Extract characteristic opinion pairs from the review text and classify them into corresponding design parameters; Calculating the semantic tendency of the characteristic viewpoint pair, and calculating the design parameter evaluation value based on the semantic tendency; For different types of design parameters, corresponding design parameter evaluation functions are constructed based on the design parameter evaluation values; the types of design parameters include discrete design parameters and continuous design parameters.

4. The product design scheme evaluation method combining customer preferences and expert knowledge as claimed in claim 3, characterized in that: Calculating the design parameter evaluation value based on the semantic tendency includes: in, Indicates design parameters No. The first design parameter value Characteristic viewpoint pairs; Indicates the semantic tendency of the feature viewpoint pair; express the number of And for different types of design parameters, corresponding design parameter evaluation functions are constructed based on the design parameter evaluation values, including: The design parameter evaluation function of discrete design parameters is modeled using piecewise functions; The design parameter evaluation function of continuous design parameters is modeled using fitting theory.

5. The product design scheme evaluation method combining customer preferences and expert knowledge as claimed in claim 1, characterized in that: The independent weight of each design parameter is obtained based on the design parameter evaluation value, star rating and artificial neural network, including: Using the design parameter evaluation value as input and the star rating as output, an artificial neural network is constructed and trained; The total weight of the design parameters in the artificial neural network when explaining the output is used as the independent weight of the design parameters.

6. The product design scheme evaluation method combining customer preferences and expert knowledge as claimed in claim 5, characterized in that: The method of using the total weight of the design parameter interpretation output in the artificial neural network as the independent weight of the design parameter includes: The absolute values ​​of all connection weights from the input layer to the hidden layer to the output layer connection attributes are summed and normalized; and the design parameters The calculation formula of the independent weight is: in, Indicates design parameters The independent weight of Represents the input layer neurons and hidden layer neurons The absolute value of the connection weight; Represents hidden layer neurons and output layer neurons The absolute value of the connection weight; and They represent the input layer neuron set and the hidden layer neuron set respectively.

7. A product design scheme evaluation system combining customer preferences and expert knowledge, characterized by: include: one or more processors; Memory; as well as One or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, the programs including a method for executing a product design solution evaluation method combining customer preferences and expert knowledge as described in any one of claims 1-6.

Images