Complex product multi-subject demand mining and importance analysis method

Abstract

The invention relates to a complex product multi-subject demand mining and importance analysis method. Determining a demand subject of the complex product based on the subject judgment element; constructing an information quality framework; performing noise filtering on the low-quality data based on a natural language processing technology; performing multi-subject demand fusion based on the semantic similarity and the demand subject weight; constructing a possibility degree model combining the hesitant fuzzy set and Kano; calculating a technical initial weight in combination with information entropy; and finally, the overall weight of the multi-subject demand is calculated based on combined weighting, and a multi-subject demand weight sorting result is obtained. According to the method, multiple technologies such as natural language processing, deep learning and Kano model improvement are combined, a technical scheme for carrying out complex product multi-subject demand mining and importance analysis is realized, and a decision basis is provided for enterprises to carry out product demand development and product design and manufacturing.

Description

Technical Field

[0001] The present invention relates to the field of product customization design, and in particular to a complex product customization design method based on demand uncertainty. Background Art

[0002] Complex products have a long service life, and their entire life cycle involves many demand entities. Their demands include the generalized demands of all groups in the entire life cycle of the product. Scholars have conducted in-depth research on the definition, evaluation, mining and analysis of multiple demand entities. However, demand data has the characteristics of multi-source heterogeneity and uneven quality. Current research on complex product demand mining does not adequately consider the impact of the quality of multi-subject demand data on demand mining; demands are ambiguous and emotionally uncertain, and there are complex interactive influence relationships between demands. Different demand entities have different factors. Current research on the importance analysis of complex product demands does not adequately consider the impact of the above factors on the results of demand importance analysis. Therefore, how to reliably mine and analyze the importance of complex product multi-subject demands is still a problem that needs to be solved. Summary of the invention

[0003] The purpose of the present invention is to conduct multi-subject demand mining and importance analysis on complex products at present. When the demand subject, demand data source and data characteristics can be determined, the collected data is first preprocessed, and the data quality is classified through the SVM model. Then, demand mining is performed on data of different qualities respectively, and the standardized expression of multi-subject demand is obtained through demand fusion. Finally, the importance of multi-subject demand is analyzed. By calculating the relative weights between multi-subject demands, user demand weights and technical demand weights, the overall weight value of multi-subject demands is obtained through combined weighting, which provides important design decision-making basis for subsequent customized design of products.

[0004] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0005] A method for mining and importance analysis of multi-subject requirements of complex products includes the following steps:

[0006] Step S1: Determine the main body of complex product demand and collect demand data;

[0007] Step S2: Classification of demand data quality and analysis of the association between demand subjects; Based on the information quality (IQ) framework and the support vector machine (SVM) method, the data quality of the collected multi-source and multi-type data is classified into low-quality data, medium-quality data and high-quality data; the demand subject is identified by character matching after data sampling to achieve matching correspondence between data sources and multiple subjects;

[0008] Step S3: low-quality demand data mining based on natural language processing;

[0009] Step S4: Medium-quality demand data mining based on FP-Growth;

[0010] Step S5: high-quality demand data mining based on BERT-BiLSTM-CRF;

[0011] Step S6: Fusion of multi-agent requirements;

[0012] Step S7: Analyze the importance of multi-subject requirements of complex products; divide the multi-subject requirements into two parts: technical requirements and user requirements, and calculate the importance weight of the requirements. Based on the impact of the differences of the demand subjects on the demand weights, analyze the weights of the demand subjects to weight the user demand weights and the technical demand weights, and obtain the overall importance weights of the multi-subject requirements;

[0013] Step S8: Organize the results of demand mining and importance analysis.

[0014] Optionally, the complex product demand subject determination and demand data collection includes the following steps:

[0015] S11: Identify complex product stakeholders

[0016] Demand subjects are part of the stakeholders of complex products. The stakeholders of complex products are classified and determined from two perspectives: the product life cycle and the product environment.

[0017] S12: Determine the judgment factors of the demand subject

[0018] Determine the main elements of demand judgment from four perspectives: product, producer, customer and environment;

[0019] S13: Grading of Demand Subject Judgment Factors

[0020] S14: Determine the demand subject through demand subject judgment factors

[0021] Conduct needs assessments using elements of judgement relevant to stakeholders;

[0022] S15: Using internal enterprise data, online public data and other data sources as data sources, demand data is collected through manual extraction, web crawlers and other methods.

[0023] S21: Determine quality assessment criteria

[0024] The multi-subject demand data is divided into high-quality data, medium-quality data and low-quality data; high-quality data must provide as complete and accurate demand information as possible about complex products; medium-quality data is related to complex product demand but has low information density; low-quality data does not contain product demand information;

[0025] S22: Information quality framework construction

[0026] An information quality framework with three information quality dimensions and six information quality characteristics is constructed; the dimensions and their characteristics are as follows:

[0027] Intrinsic quality dimensions: credibility characteristics, information value density characteristics, signal-to-noise ratio characteristics;

[0028] Characterization quality dimensions: completeness characteristics, description accuracy characteristics;

[0029] Contextual quality dimensions: contextual quality characteristics;

[0030] S23: Data quality classification based on support vector machine

[0031] Based on the constructed information quality framework, each multi-subject demand data is transformed into a feature vector, and support vector machine is used to classify the data quality;

[0032] S24: Correlation analysis of multiple demand entities

[0033] The demand subject is identified by character matching after data sampling.

[0034] S31: Data Preprocessing

[0035] Write filtering rules to remove explicit noise; use support vector machine model to remove implicit noise;

[0036] S32: Demand Item Mining

[0037] The low-quality data itself has time attributes and a certain input order, and the hierarchical clustering algorithm is used for cluster analysis;

[0038] S33: Attention Analysis

[0039] The attention of the demand item is calculated based on the four indicators of the number of texts, forwardings, likes and comments of the demand item. For data without forwarding, likes and comments, the attention of the demand item can be calculated based on the number of texts and the frequency of appearance in the text.

[0040] S34: Multi-agent demand expression

[0041] The requirement template is constructed in the form of a triple, expressed as B1 = <requirement number, requirement subject, requirement attention>.

[0042] Optionally, the FP-Growth-based medium-quality demand data mining includes the following steps:

[0043] S41: data preprocessing;

[0044] Delete abnormal and redundant data; screen high-value data features; discretize numerical feature items;

[0045] S42: Define the rule antecedent, consequent and their associated rules

[0046] Before mining association rules, it is necessary to define the antecedents and consequents of the algorithm and their association rules; introduce the five elements of total quality management, which include people, machines, materials, methods, and environment, divide the design objects, resources, environment, activities and other factors into antecedents and consequents, and determine the constraints of association rules;

[0047] S43: Association rule mining based on FP-Growth

[0048] Input the transaction set P of the complex product usage process, the minimum support Minsup, and the minimum confidence Mincon, use the FP-Growth algorithm to mine the strong association rules that meet the constraints, and output the association rules between the design object and various resources and activities in the manufacturing, operation, and maintenance stages;

[0049] S44: Rule-based RAMS attribute mapping

[0050] The association relationship can reflect the requirements on reliability, availability, maintainability and safety proposed by the relevant subjects; define RAMS-related design attributes, build mapping rules between association rules and design attributes, map the association rules of manufacturing, operation and maintenance stages to the design stage, and convert them into requirement descriptions based on expert experience;

[0051] S45: Multi-agent demand expression

[0052] The requirement template is constructed in the form of a triple, expressed as B2 = <requirement number, requirement subject, requirement description>.

[0053] Optionally, the high-quality demand data mining based on BERT-BiLSTM-CRF includes the following steps:

[0054] S51: Document preprocessing

[0055] Use statistical machine learning models and combine them with professional dictionaries in the target product demand field to learn and master the rules of word segmentation; for some demand information stored in tables, perform standardization checks on the table format; for standardized documents with hierarchical structures, use regular expressions to identify chapter information and secondary title information in high-quality data, and use text similarity calculations to match the document structure with the target product family metastructure tree;

[0056] S52: Demand word extraction

[0057] S53: Multi-agent demand expression

[0058] The requirement template is constructed in the form of a triple, expressed as B3 = <requirement subject, requirement item, requirement value>.

[0059] Optionally, the multi-agent demand fusion includes the following steps:

[0060] S61: Build a standardized demand expression template

[0061] S62: Requirement item fusion based on semantic similarity

[0062] For repeated or similar requirements, the requirement items are converted into word vectors, and the cosine similarity is calculated to define the semantic similarity between the requirement items. When the semantic similarity is not less than the threshold β, the two requirement items are combined into one, and then the requirement values ​​are merged. When the semantic similarity is less than the threshold β, the two requirement items are retained.

[0063] S63: Demand Value Fusion

[0064] For the demands of multiple entities, if the demand items and demand values ​​are the same, they will be directly merged into one item, and the demand entities will be multiple demand entities; if the demand items are the same but the demand values ​​are different, the one with a higher overall weight of the multiple entity demands will be selected to cover the other.

[0065] Optionally, the complex product multi-subject demand importance analysis includes the following steps:

[0066] S71: Relative weight calculation among multiple subjects based on improved Kano

[0067] Based on the impact of differences between demand subjects on demand weights, combined with hesitant fuzzy sets,

[0068] S72: Calculation of technical requirements weights based on information entropy and fuzzy DEMATEL

[0069] S73: User demand weight calculation based on emotion-cloud model

[0070] S74: Multi-agent demand weight fusion based on combined empowerment

[0071] According to the calculation results in S71, S72, S73, and S74 above, combined with the relative weights between demand entities, the technical demand weights, and the user demand weights, the final overall weight of the multi-subject demand is calculated based on a combined weighting method.

[0072] Optionally, the demand mining and importance analysis result collation comprises the following steps:

[0073] Returning to step S63, according to the calculated overall weight of the multi-subject demands, for the case where the demand items are the same but the demand values ​​are different, the multi-subject demand with a higher overall weight is selected to cover the other to obtain the final demand mining result.

[0074] The beneficial effects of the present invention are:

[0075] (1) Complex products have long service cycles, and their entire life cycle involves many demand entities. Understanding and meeting the needs of multiple demand entities is crucial to successfully completing projects and providing valuable products or services. However, the demand entities involved in different industries and different objects are different, and it is difficult to define the demand entities with a unified definition standard. The present invention takes into account both the product life cycle and the product environment, and determines the six types of demand entities in the complex product design process by setting the basic judgment factors, demand preference factors, and overall decision-making factors of the demand entity for evaluation.

[0076] (2) The quality of complex product demand data varies. Current research on complex product demand mining has not considered the impact of multi-subject demand data quality on demand mining. This paper proposes a multi-subject demand data quality classification method based on the information quality (IQ) framework and the support vector machine (SVM) model. First, the IQ framework is used to build the model to obtain three information quality dimensions with 6 information quality features; the multi-subject demand data is converted into feature vectors, and the SVM model is used for quality classification prediction. According to the obtained data quality scores, the multi-subject demand data is divided into low-quality data, medium-quality data and high-quality data for subsequent demand mining process.

[0077] (3) Currently, most multi-subject demand data are in the form of unstructured text (such as technical bidding documents). The description of technical indicators is in the form of natural language description, which lacks standardization. At present, most of them are manually extracted by designers to obtain the required demand items or reuse empirical values. The present invention combines natural language processing (NLP), deep learning and other technologies to propose different data mining methods for data of different qualities. For low-quality data, sample preprocessing, noise filtering and feature word extraction are performed by NLP method and SVM technology, and demand items are clustered by Single-pass clustering algorithm, and then the demand attention is calculated; for medium-quality data, after data cleaning, filling, screening and discretization, demand mining and association rule mining are performed by FP-Growth algorithm; for high-quality data, word segmentation and stop word removal are implemented based on demand domain dictionary, and the hierarchical structure of the document is extracted based on regular expression, so that the document structure is aligned with the product family meta-structure tree, and a deep network model based on BERT-BiLSTM-CRF is constructed to identify demand entities, and the demand information is matched with the demand meta-model through the demand entity alignment algorithm. Finally, a standardized expression template for requirements is constructed, and the cosine similarity is used to calculate the semantic similarity between the main requirement items, and requirement fusion is achieved based on semantic similarity.

[0078] (4) User needs are ambiguous and emotionally uncertain. There are complex interactive relationships between needs, and there are differences between demand subjects. These factors have a great impact on the results of demand importance analysis. Current research has not considered the impact of comprehensive consideration of user preferences, relationships between needs, and differences between demand subjects on demand importance analysis. The present invention comprehensively considers technical needs and user needs to calculate weights. First, based on the improved Kano model, the possibility matrix of demand subjects for demand items is constructed, the similarity of the possibility vectors between demand subjects is calculated, and the similarity matrix is ​​normalized to obtain the relative weights between multiple demand subjects; the initial weights of technical needs are calculated based on information entropy, and the influence utility between needs is calculated by introducing triangular fuzzy numbers and DEMATEL methods to construct a comprehensive weight function CRW of technical needs; sentiment analysis is performed on subjective texts containing emotional preferences, and the subjective and objective weights of user needs are calculated by the S-cloud model integrated with sentiment analysis. User satisfaction is calculated by combining subjective and objective calculations, and then the user demand weights are calculated in combination with demand attention; finally, the relative weights between multiple demand subjects, technical demand weights, and user demand weights are combined and weighted to obtain the overall weight of multi-subject demand. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 A flowchart of multi-subject demand mining and importance analysis for complex products;

[0080] Figure 2 Determine the process diagram for the demand subject;

[0081] Figure 3 A schematic diagram of the technical route for data quality classification;

[0082] Figure 4 Develop a network diagram for the impact of technical requirements; DETAILED DESCRIPTION

[0083] To make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be described clearly and completely in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0084] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the invention claimed for protection, but merely represents some embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0085] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features and technical solutions in the embodiments may be combined with each other.

[0086] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0087] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the invention product is usually placed when in use, or the orientation or positional relationship commonly understood by those skilled in the art. Such terms are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0088] Example 1, see Figure 1-Figure 4 As shown, the embodiment of the present application provides a method for mining and analyzing the importance of multiple subjects of complex products, and by combining multiple technologies such as natural language processing, deep learning, and improved Kano model, a technical solution for mining and analyzing the importance of multiple subjects of complex products is realized, providing a decision-making basis for enterprises to carry out product demand development and product design and manufacturing.

[0089] A method for mining and analyzing the importance of complex product multi-subject requirements includes the following steps:

[0090] Step S1: Determine the main demand of complex products and collect demand data

[0091] In the demand analysis stage of complex product design, designers need to weigh and adopt the needs of some stakeholders to carry out subsequent customized design. The demand subject refers to the group or organization that puts forward requirements such as function, performance, user experience, reliability, etc. for complex product customized design. The process of determining the complex product demand subject is as follows:

[0092] S11: Identify complex product stakeholders

[0093] Demand subjects are part of the stakeholders of complex products. Classify and identify stakeholders of complex products from two perspectives: the product life cycle and the product environment.

[0094] S12: Determine the judgment factors of the demand subject

[0095] The customization needs of complex products are mainly affected by customer needs and environmental requirements, but the actual realization of the function depends on the technical capabilities of the producer and the inherent properties of the product itself. Therefore, the main judgment factors of demand are determined from four perspectives: product, producer, customer and environment.

[0096] S13: Demand Subject Judgment Factor Classification

[0097] Determine the demand-related subject factors related to each stakeholder. According to different subject evaluation issues, the demand subject judgment factors are divided into three levels. Basic judgment factors: the basic conditions and prerequisites for evaluating whether specific needs can be met; demand preference factors: the evaluation criteria for judging the preference degree or obviousness of specific needs; overall decision factors: the decision basis for comprehensively evaluating the possibility of meeting specific needs and the implementation strategy.

[0098] S14: Determine the demand subject through demand subject judgment factors.

[0099] Conduct a needs assessment using judgement elements relevant to stakeholders. Steps include:

[0100] S141: Use basic judgment factors to conduct preliminary problem analysis and evaluate the basis for the existence of demand

[0101] S142: Use demand preference factors to deeply analyze selectivity issues and evaluate demand tendencies

[0102] S143: Finally, a final evaluation is conducted through the overall decision-making factors to determine the feasibility of the requirements.

[0103] S144: Add stakeholders who have passed the judgment factor of the needs assessment as demand subjects.

[0104] like Figure 2 As shown, stakeholders in the complex product design process and their related demand subject judgment factors are identified.

[0105] Demand assessment is conducted through the demand subject judgment factors: first, basic judgment factors are used to conduct preliminary problem analysis to evaluate the basis for the existence of demand. If the demand exists, proceed to the next step; then, demand preference factors are used to conduct in-depth analysis of selectivity issues and evaluate the tendency of demand. If the tendency of demand is obvious, proceed to the next step; finally, a final assessment is conducted through the overall decision-making factors to determine the feasibility of the demand. If the demand can be realized, the demand assessment is passed and the stakeholder is added as the demand subject.

[0106] S15: Using internal enterprise data, online public data and other data sources as data sources, demand data is collected through manual extraction and other methods.

[0107] Step S2: Demand data quality classification and demand subject correlation analysis

[0108] like Figure 3 As shown, this step is based on the information quality (IQ) framework and the support vector machine (SVM) method to classify the data quality of the collected multi-source and multi-type data, and perform demand subject association analysis to achieve matching correspondence between data sources and multiple subjects.

[0109] For multi-subject demand data, an IQ framework with three information quality dimensions and six information quality features is constructed. Experts conduct feature evaluation between [0,1] and convert each demand data into a feature vector for data quality classification. A classification model is constructed and SVM is used for data quality classification. For the demand data collected from multiple subjects, data samples are obtained by random sampling. Based on the information quality feature vectors obtained by expert evaluation, the sample data is manually labeled, 80% of the data samples are selected for training the SVM model, and the remaining 20% ​​are used for classification prediction of data quality. The SVM classification model constructed above is used to classify the multi-subject demand data into low-quality data, medium-quality data, and high-quality data. The specific steps are as follows:

[0110] S21: Determine quality assessment criteria

[0111] The multi-subject demand data is divided into high-quality data, medium-quality data and low-quality data. High-quality data must provide as complete and accurate demand information as possible about complex products; medium-quality data is related to complex product demand but has low information density; low-quality data does not contain product demand information.

[0112] S22: Information quality (IQ) framework construction

[0113] An IQ framework consisting of three information quality dimensions with six information quality characteristics is constructed. The dimensions and their characteristics are as follows:

[0114] Intrinsic quality (D1) dimension: credibility (f1) characteristics, information value density (f2) characteristics, signal-to-noise ratio (f3) characteristics;

[0115] Representational quality (D2) dimension: completeness (f4) characteristics, description accuracy (f5) characteristics;

[0116] Context quality (D3) dimension: context quality (f6) features.

[0117] S23: Data quality classification based on Support Vector Machine (SVM)

[0118] Based on the constructed IQ framework, each multi-subject demand data is converted into a feature vector, and SVM is used to classify the data quality. The specific process is as follows:

[0119] S231: Data sampling. Random sampling is used to obtain data samples for data quality classification from the demand data collected from multiple subjects.

[0120] S232: Data feature vectorization. Based on the six information quality features of the IQ framework, experts conduct feature evaluation between [0,1] and convert each demand data into a feature vector.

[0121] S233: Classification model construction. Based on the information quality feature vector obtained by expert evaluation, the sample data is manually labeled, 80% of the data samples are selected for training the SVM model, and the remaining 20% ​​are used for classification prediction of data quality.

[0122] S234: Data quality classification: Use the SVM classification model constructed above to classify the multi-subject demand data into low-quality data, medium-quality data, and high-quality data.

[0123] S24: Correlation analysis of multiple demand entities

[0124] The method of identifying the subject of requirements is to perform character matching after data sampling. For example, if an emotion word is identified, it is considered that the data is online review data and the requirement is the user's requirement; if characters such as "fault" and "maintenance" are identified, it is considered that the data is a fault maintenance record and the requirement is the requirement of maintenance personnel; if characters such as "bidding" and "contract" are identified, it is considered that the data is the bidding technical conditions and the requirement is the requirement of the operator.

[0125] Step S3: Mining of low-quality requirement data based on natural language processing

[0126] Different quality data adopts different requirement mining methods. This step is based on the natural language processing (Natural Language Processing, NLP) method to mine low-quality requirement data, and its process includes:

[0127] S31: Data preprocessing

[0128] Write filtering rules to remove explicit noise; use the SVM model to remove implicit noise. The process includes:

[0129] S311: Identify special characters such as "#", "【】", "http: / / ", single emoji, and duplicate data, and filter out explicit noise such as advertisements, overly short invalid comments, and duplicate information.

[0130] S312: Data sampling and marking. In the data after eliminating explicit noise, randomly select a certain number of samples and mark each of them as valid or invalid.

[0131] S313: Sample preprocessing. Mainly for word segmentation and stop word removal, etc., to split the continuous language information in the sample into individual words, and filter out frequently occurring but meaningless words such as "de", "wo", "ba", etc. in the word-segmented sample with the help of the stop word list.

[0132] S314: Use the method of mutual information (MI) to extract feature words. Manually select a set of seed words s-words containing implicit noise, calculate the mutual information MI(t-word, s-words) between the target word t-word and s-words, sort the words according to the mutual information MI from large to small, and select the top h words as feature words.

[0133] S315: Text vectorization. According to the feature values, form an adjusted word vector for the feature words of each text, and use the Boolean logic model for text vectorization. Assume that the target text is T. If the feature word appears in the target text T, the feature value in the vector is 1, otherwise it is 0. The feature word vector of the text can be expressed as Tj = {t1, t2,..., tm}, ti ∈ {0, 1}.

[0134] S316: SVM classification model filters implicit noise. Use SVM to build a classification model and select a linear kernel function to map the text into a high-dimensional space in order to process nonlinearly separable data. Filter the data for implicit noise.

[0135] S32: Demand Item Mining

[0136] Low-quality data itself has time attributes and a certain input order, and a hierarchical clustering algorithm is used for clustering analysis. This step uses the classic Single-pass clustering algorithm to perform clustering analysis on the low-quality data after preprocessing to obtain demand keywords. The process includes:

[0137] S321: Text vectorization: Use word frequency-inverse text frequency to perform text vectorization.

[0138] S322: Text clustering. Input the feature vector of the first text and establish the first cluster; input the feature vector of the next text and use the angle cosine value to calculate the distance between each feature word; if the maximum cosine similarity of the two feature vectors is greater than the threshold, the second input text is classified into the first cluster, otherwise a new cluster is established, and the vector is the center vector of the new cluster; update the center vector of the cluster, and end a cluster.

[0139] S323: Clustering loop. Input the feature value vectors of the entire text one by one in chronological order, and repeat the above process until it ends to obtain the final clustering result.

[0140] S324: naming the clusters; counting the most frequently appearing feature words in the clusters, and matching the characters with the demand word library; the feature words with the highest matching degree are the demand words of the clusters, and the clusters become demand items.

[0141] S33: Attention Analysis

[0142] The demand item attention is calculated based on the four indicators of the number of texts, forwardings, likes and comments. For data without forwarding, likes and comments, the demand attention can be calculated based on the number of texts and the frequency of appearance in the text.

[0143] S34: Multi-agent demand expression

[0144] The requirement template is constructed in the form of a triple, expressed as MR1 ​​= <requirement subject A, requirement item B, requirement attention C1>.

[0145] Step S4: Mining medium-quality demand data based on FP-Growth

[0146] Medium-quality data has characteristics such as time attributes, discretization, and statistical properties. The subject needs reflected by such data cannot be directly extracted or mined from a semantic perspective and are implicit needs. This step uses the FP-Growth algorithm to mine association rules, obtain feature items and association relationships, and reflect related subject needs. The process includes:

[0147] S41: Data preprocessing.

[0148] It mainly includes deleting abnormal and redundant data; screening high-value data features; and discretizing numerical feature items.

[0149] S42: Define the rule antecedent, consequent and their associated rules

[0150] Before mining association rules, it is necessary to define the antecedents, consequents and association rules of the algorithm. Introduce the five elements of total quality management (man-machine-material-method-environment), divide the design objects, resources, environment, activities and other factors into antecedents and consequents, and determine the association rule constraints.

[0151] S43: Association rule mining based on FP-Growth

[0152] Input the transaction set P of the complex product usage process, the minimum support Minsup, and the minimum confidence Mincon, use the FP-Growth algorithm to mine the strong association rules that meet the constraints, and output the association rules between the design object and various resources and activities in the manufacturing, operation, and maintenance stages.

[0153] S44: Rule-based RAMS attribute mapping

[0154] The association relationship can reflect the requirements for reliability, availability, maintainability and safety (RAMS) proposed by the relevant subjects. Define RAMS-related design attributes, build mapping rules between association rules and design attributes, map the association rules of the manufacturing, operation and maintenance stages to the design stage, and convert them into requirement descriptions based on expert experience.

[0155] S45: Multi-agent demand expression

[0156] The requirement template is constructed using a triple form, expressed as MR2 = <requirement subject A, requirement item B, requirement description C2>.

[0157] Step S5: High-quality demand data mining based on BERT-BiLSTM-CRF

[0158] Different quality data use different demand mining methods. This step uses the named entity recognition method to extract multi-subject requirements from high-quality data. Its main process includes:

[0159] S51: Document preprocessing

[0160] Use statistical machine learning models, combined with professional dictionaries in the target product demand field, to learn and master the rules of word segmentation; for some demand information stored in tables, perform standardization checks on the table formats; for standardized documents with hierarchical structures, use regular expressions to identify chapter information and secondary title information in high-quality data, and use text similarity calculations to match the document structure with the target product family meta-structure tree.

[0161] S52: Demand word extraction

[0162] Based on the BERT pre-trained model, a BERT-BiLSTM-CRF network model is built to extract structured information. The main process includes:

[0163] S521: Build BERT embedding layer. Build an embedding layer based on the BERT model, integrate the contextual information on the left and right sides of the text to build a deep bidirectional representation, which can capture more realistic contextual information and learn the coherent relationship between text fragments.

[0164] S522: Constructing a BiLSTM layer. The BiLSTM network uses the n-dimensional word vector obtained by the BERT layer as the input of each time step, and its output is the label score, that is, the predicted score of all labels for each word in the sentence.

[0165] S523: Construct CRF layer: The Conditional Random Field (CRF) layer is used to complete the sequence labeling of sentences. The prediction score of each label output by the BiLSTM layer is used as input, and through normalization, loss function adjustment, model decoding prediction, Viterbi algorithm to solve the optimal path and other processes, the labeled result set is output, that is, the labeled sequence with the maximum probability, as the data source of the demand metamodel matching module.

[0166] S524: Requirement metamodel matching: Match the original requirement items extracted from the normalized document with the requirement items in the component metamodel, and store the requirement items and values ​​whose similarity is greater than a threshold.

[0167] S53: Multi-agent demand expression

[0168] The requirement template is constructed using a triple form, expressed as MR3 = <requirement subject A, requirement item B, requirement value C3>.

[0169] Step S6: Multi-agent demand fusion

[0170] Different demand subjects have certain overlaps in their requirements for the product's usage process and functions, and have different focus angles and interest positions, which leads to the existence of identical or similar redundant demands among multiple subjects, as well as conflicts among multiple subject demands. After completing the mining and extraction of multiple subject demands, this step eliminates redundancy and conflict in demands and realizes the standardized expression of demands.

[0171] S61: Build a standardized demand expression template

[0172] The standardized requirement expression template is: MR=<N,A,B,C,D> ,N=requirement number,A=subject,A∈{A1,A2,…,A6},B=requirement item,C=additional content,C∈{C1,C2,C3},D=requirement weight.

[0173] Among them, the requirement number N is the unique identifier of a single requirement item; subject A refers to the requirement subject that proposes the requirement, which can be one or more; B is the requirement item obtained by demand mining and fusion; additional content C is the valuable content other than the requirement item and the requirement subject after the three multi-subject requirement mining methods mentioned above are used for the requirement data of different qualities, including one or more of the requirement attention C1, requirement description C2 and requirement value C3; the requirement weight D refers to the relative importance of the requirements after weight calculation, which is used to support designers in making subsequent design decisions and resource allocation. The calculation method of the requirement weight will be explained in the subsequent steps.

[0174] S62: Requirement item fusion based on semantic similarity

[0175] For repeated or similar requirements, the requirements are converted into word vectors, and the cosine similarity is calculated to define the semantic similarity between the requirements. When the semantic similarity is not less than the threshold β, the two requirements are combined into one, and the requirements values ​​are merged; when the semantic similarity is less than the threshold β, the two requirements are retained.

[0176] S63: Demand Value Fusion

[0177] For the demands of multiple subjects, if the demand items and demand values ​​are the same, they are directly merged into one item, and the demand subject is multiple demand subjects; if the demand items are the same but the demand values ​​are different, the one with a higher overall weight of the multi-subject demand is selected to cover the other. The calculation method of the overall weight of the multi-subject demand will be explained in the subsequent steps.

[0178] Step S7: Analysis of importance of multi-subject requirements for complex products

[0179] This step divides the multi-subject requirements into two parts: technical requirements and user requirements, and calculates the weight of the requirements. At the same time, considering the impact of the differences in the requirements on the weight of the requirements, the weight of the requirements is analyzed to weight the user requirements weight and the technical requirements weight, and the overall importance weight of the multi-subject requirements is obtained. The main process includes:

[0180] S71: Relative weight calculation among multiple subjects based on improved Kano

[0181] Considering the impact of differences between demand entities on demand weights, this step combines hesitant fuzzy sets to construct an improved Kano model to calculate the relative weights between multiple entities. The process includes:

[0182] S711: Preparation for improving the Kano model. Combine the Kano model with the hesitant fuzzy set to establish an improved Kano questionnaire; combine the five types of demand classification, namely basic type (M), expected type (O), exciting type (A), irrelevant type (I) and reverse type (R), to develop a Kano evaluation table.

[0183] S712: Construct a possibility matrix. According to the results of the improved Kano questionnaire, the survey matrix P, N is obtained, and the interaction matrix S is calculated; the possibility vector Y of each demand category is calculated according to the Kano evaluation table; and the possibility matrix T of the demands of multiple demand subjects is obtained by combination.

[0184] S713: Constructing a similarity matrix between demand entities. Using cosine similarity, the similarity of the possible degree vectors between demand entities is calculated; the similarities between all demand entities are combined to obtain a similarity matrix.

[0185] S714: Calculate the relative weights between demand entities: sum the similarity matrix to obtain the relative weights of other demand entities to demand entity A j The overall similarity of other demand entities indicates that demand entity A j The support degree of the demand subject A is obtained by normalization j The credibility of the demand subject is calculated to obtain the relative weight θ between the demand subjects of the demand subject Aj

[0186] S72: Calculation of technical requirements weights based on information entropy and fuzzy DEMATEL

[0187] Technical requirements often have complex hierarchical nature, and there are complex interactions within and between levels, which affect the calculation results of demand weights. This step combines information entropy and fuzzy DEMATEL method to analyze the mutual influence relationship between technical requirements, obtain the importance of the requirements, and calculate the comprehensive demand weight by combining information entropy. The main process includes:

[0188] S721: Calculate the initial weight of the technology requirement based on the information entropy. According to the possibility vector Y obtained in step S712, calculate the information entropy value E of the technology requirement, and further calculate the initial weight ω of the technology requirement. v .

[0189] S722: Initial evaluation of the impact relationship between requirements. The 5-level scale language term ID is used, namely, huge impact {E: (0.7, 0.9, 1.0)}, high impact {H: (0.5, 0.7, 0.9)}, medium impact {M: (0.3, 0.5, 0.7)}, low impact {E: (0.1, 0.3, 0.5)}, and no impact {E: (0.0, 0.1, 0.3)}; the expert group members construct the initial evaluation matrix IM by analyzing the internal impact relationship between technical requirements.

[0190] S723: Construct a comprehensive influence matrix. Combine the initial evaluation matrix IM with triangular fuzzy numbers to obtain the fuzzy direct influence matrix M, normalize it to obtain the normalized influence matrix N, and then calculate the fuzzy comprehensive influence matrix. After defuzzification, a clear comprehensive influence matrix T is obtained.

[0191] S724: Calculate the impact factors. Calculate the impact degree D based on the comprehensive impact matrix v and the degree of influence C v ; Further calculation to obtain the centrality M v and the degree of cause R v ; Further calculate the impact importance of demand η v .

[0192] S725: Draw an impact relationship network diagram of technical requirements. According to the above calculation results, the requirements are divided into a cause group and an effect group; draw an impact relationship network diagram of technical requirements, with arrows pointing from the cause group requirements to the effect group requirements, and mark the initial weights corresponding to the technical requirements.

[0193] like Figure 4 As shown, according to the cause R v The calculation results divide the requirements into cause groups and effect groups; draw the impact relationship network diagram of technical requirements, Figure 4 The middle circle represents the technical requirements and their corresponding initial weights ω, and the arrows point from the cause group requirements to the effect group requirements; t ij Indicates technical requirements CR i Technical requirements CR j The algorithm corresponds to step S4. First, the original requirement value of the module extracted from the technical bid is matched with the requirement item in the module requirement metamodel, the value is converted to a unit check, the value domain is selected according to the rules, and for the null value and abnormal value attribute items, the mean value of the attribute is calculated and assigned.

[0194] S726: Construct a comprehensive weight function. Considering the direct impact of demand on the initial weight, and the promotion or inhibition of the indirect impact of demand on the demand weight, a comprehensive weight function CRW of technical demand is constructed according to the above calculation results to obtain the final technical demand weight.

[0195] S73: User demand weight calculation based on emotion-cloud model;

[0196] User needs have linguistic ambiguity and emotional uncertainty, which affect the calculation results of demand weights. This step quantifies user emotional preferences based on the emotion-cloud model to achieve weight calculation of user needs. The process includes:

[0197] S731: Sentiment analysis. Analyze subjective texts with emotional preferences through sentiment dictionaries, subdivide the emotional intensity from positive to negative into seven levels, and correspond emotional degree adverbs to quantifiers; calculate the sentiment score by comprehensively considering the number of emotional words in the text, the total number and the frequency of degree adverbs.

[0198] S732: Conversion between language information and cloud. Based on the improved golden section five-cloud generation method, the adjustment coefficient is calculated in combination with the given language term set, a multi-level cloud model is constructed, and the upper and lower bounds of the interval are clarified.

[0199] S733: Calculation of user demand weight. The constructed cloud model is used to process the data, and the subjective weight of the demand is calculated according to the middle value of the interval number of the cloud model, and the objective weight is calculated according to the interval width. The user demand satisfaction is obtained by combining subjective and objective calculations; the user demand weight is obtained by combining the demand attention and normalization.

[0200] S74: Multi-agent demand weight fusion based on combined empowerment

[0201] According to the above calculation results, combined with the relative weights between demand entities, technical demand weights and user demand weights, the final overall weight of multi-subject demand is calculated based on the combined weighting method.

[0202] Step S8: Demand mining and importance analysis results collation

[0203] Return to step S63, according to the calculated overall weight of the multi-subject demands, for the case where the demand items are the same but the demand values ​​are different, the multi-subject demand with a higher overall weight is selected to cover the other, and the final demand mining result is obtained.

[0204] The calculated overall weight of multi-subject demands reflects the overall ranking of multi-subject demands from high to low. The demands at the top are the demands that need to be given priority and satisfied in subsequent product design. The ranking results of multi-subject demand weights can provide decision-making basis for enterprises to carry out product demand development and product design and manufacturing.

[0205] Embodiment 2: This embodiment provides a medium and a device for storing and executing a method for mining and importance analysis of multi-agent requirements of complex products.

[0206] A computer storage medium stores a program executable by a processor, wherein the program executable by the processor is used to implement the method for mining and importance analysis of multi-agent requirements of complex products when executed by the processor.

[0207] A complex product multi-subject demand mining and importance analysis device, the device includes: a sensor for collecting complex product demand subjects and demand data; at least one processor, capable of processing various data related to the complex product demand subjects and demand data collected by the sensor; at least one memory, for storing various data related to the complex product demand subjects and demand data collected by the sensor and at least one program; when the at least one program is executed by the at least one processor, so that the at least one processor implements the complex product multi-subject demand mining and importance analysis method, the processor and the memory can be connected through a bus or other means to store at least one program; the memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory can include a high-speed random access memory, and can also include a non-transient memory, such as at least one disk storage device, a flash memory device or other non-transient solid-state storage device. In some embodiments, the memory optionally includes a memory remotely arranged relative to the processor, and these remote memories can be connected to the device through a network. Examples of the above-mentioned network include but are not limited to the Internet, corporate intranet, local area network, mobile communication network and combinations thereof. When the at least one program is executed by the at least one processor, the at least one processor implements the complex product multi-subject demand mining and importance analysis method.

[0208] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0209] The above embodiments are only used to illustrate the present invention and are not intended to limit the technical solutions described in the present invention. Although the present invention has been described in detail with reference to the above embodiments, the present invention is not limited to the above specific implementation methods. Therefore, any modification or equivalent replacement of the present invention; and all technical solutions and improvements thereof that do not depart from the spirit and scope of the invention are included in the scope of the claims of the present invention.

Claims

1. A method for mining and importance analysis of multi-subject requirements of complex products, the method comprising the following steps: Step S1: Determine the main body of complex product demand and collect demand data; Step S2: Classification of demand data quality and analysis of the association between demand subjects; Based on the information quality framework and support vector machine method, the data quality of the collected multi-source and multi-type data is classified into low-quality data, medium-quality data and high-quality data; the demand subject is identified by character matching after data sampling to achieve matching correspondence between data sources and multiple subjects; Step S3: low-quality demand data mining based on natural language processing; Step S4: Medium-quality demand data mining based on FP-Growth; Step S5: high-quality demand data mining based on BERT-BiLSTM-CRF; Step S6: Fusion of multi-agent requirements; Step S7: Analyze the importance of multi-subject requirements of complex products; divide the multi-subject requirements into two parts: technical requirements and user requirements, and calculate the importance weight of the requirements. Based on the impact of the differences of the demand subjects on the demand weights, analyze the weights of the demand subjects to weight the user demand weights and the technical demand weights, and obtain the overall importance weights of the multi-subject requirements; Step S8: Organize the results of demand mining and importance analysis.

2. The complex product multi-agent demand mining and importance analysis method according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21: Determine quality assessment criteria; The multi-subject demand data is divided into high-quality data, medium-quality data and low-quality data; high-quality data provides as complete and accurate demand information as possible about complex products; medium-quality data is related to complex product demand but has low information density; low-quality data does not contain product demand information; S22: Information quality framework construction; An information quality framework with three information quality dimensions and six information quality characteristics is constructed; the dimensions and their characteristics are as follows: Intrinsic quality dimensions: credibility characteristics, information value density characteristics, signal-to-noise ratio characteristics; Characterization quality dimensions: completeness characteristics, description accuracy characteristics; Contextual quality dimensions: contextual quality characteristics; S23: Data quality classification based on support vector machine; Based on the constructed information quality framework, each multi-subject demand data is transformed into a feature vector, and support vector machine is used to classify the data quality; S24: Correlation analysis of multiple demand entities; The demand subject is identified by character matching after data sampling; Step S23 specifically includes the following steps: S231: Data sampling: for the demand data collected from multiple subjects, random sampling is used to obtain data samples for data quality classification; S232: Data feature vectorization: Based on the six information quality features of the information quality framework, feature evaluation is performed between [0, 1] to convert each demand data into a feature vector; S233: Classification model construction: Based on the information quality feature vector obtained by the evaluation, the sample data is manually labeled, 80% of the data samples are selected for training the support vector machine model, and the remaining 20% ​​are used for classification prediction of data quality; S234: Data quality classification; use the support vector machine classification model constructed above to classify the multi-subject demand data into low-quality data, medium-quality data and high-quality data.

3. The complex product multi-agent demand mining and importance analysis method according to claim 2, characterized in that: Step S3 specifically includes the following steps: S31: data preprocessing; Write filtering rules to remove explicit noise; use support vector machine model to remove implicit noise; S32: Demand item mining; The low-quality data itself has time attributes and a certain input order, and the hierarchical clustering algorithm is used for cluster analysis; S33: Attention analysis; The attention level of the demand item is calculated by weighted calculation based on the number of texts, forwardings, likes, and comments on the demand item. S34: Multi-agent demand expression; The requirement template is constructed in the form of a triple, expressed as B1 = <requirement number, requirement subject, requirement attention>.

4. The complex product multi-agent demand mining and importance analysis method according to claim 3, characterized in that: Step S32 specifically includes: S321: Text vectorization; text vectorization is performed using word frequency-inverse text frequency; S322: text clustering; input the feature value vector of the first text and establish the first cluster; input the feature value vector of the next text and use the angle cosine value to calculate the distance between each feature word; if the maximum cosine similarity of the two feature value vectors is greater than the threshold, the second input text is classified into the first cluster, otherwise a new cluster is established, and the vector is the center vector of the new cluster; update the center vector of the cluster, and end a cluster; S323: Clustering loop: input the feature value vectors of the entire text one by one in chronological order, and repeat the above process until the end to obtain the final clustering result; S324: naming the clusters; counting the most frequently appearing feature words in the clusters, and matching the characters with the demand word library; the feature words with the highest matching degree are the demand words of the clusters, and the clusters become demand items.

5. The complex product multi-agent demand mining and importance analysis method according to claim 4, characterized in that: Step S4 The specific steps include: S41: data preprocessing; Delete abnormal and redundant data; screen high-value data features; discretize numerical feature items; S42: define the rule antecedent, consequent and their associated rules; Before mining association rules, it is necessary to define the antecedents and consequents of the algorithm and their association rules; introduce the five elements of total quality management, which include people, machines, materials, methods, and environment, divide the design objects, resources, environment, activities and other factors into antecedents and consequents, and determine the constraints of association rules; S43: Association rule mining based on FP-Growth; Input the transaction set of the complex product use process, minimum support, minimum confidence, use the FP-Growth algorithm to mine the strong association rules that meet the constraints, and output the association rules between the design object and various resources and activities in the manufacturing, operation, and maintenance stages; S44: Rule-based RAMS attribute mapping; Define RAMS-related design attributes, build mapping rules between association rules and design attributes, and map the association rules of manufacturing, operation, and maintenance stages to the design stage; S45: Multi-subject demand expression; The requirement template is constructed in the form of a triple, expressed as B2 = <requirement number, requirement subject, requirement description>.

6. The complex product multi-agent demand mining and importance analysis method according to claim 5, characterized in that: Step S5 specifically includes the following steps: S51: document preprocessing; Use statistical machine learning models and combine them with professional dictionaries in the target product demand field to learn and master the rules of word segmentation; for some demand information stored in tables, perform standardization checks on the table format; for standardized documents with hierarchical structures, use regular expressions to identify chapter information and secondary title information in high-quality data, and use text similarity calculations to match the document structure with the target product family metastructure tree; S52: demand word extraction; S53: Multi-agent demand expression; The requirement template is constructed in the form of a triple, expressed as B3 = <requirement subject, requirement item, requirement value>.

7. The complex product multi-agent demand mining and importance analysis method according to claim 6, characterized in that: Step S6 specifically includes the following steps: S61: Construct standardized demand expression template; The standardized requirement expression template is: MR=<N,A,B,C,D> , N = requirement number, A = subject, A∈{A1, A2, …, A6}, B = requirement item, C = additional content, C∈{C1, C2, C3}, D = requirement weight; Among them, the requirement number N is the unique identifier of a single requirement item; subject A refers to the requirement subject that puts forward the requirement; B is the requirement item obtained by requirement mining and fusion; additional content C is the valuable content other than the requirement item and the requirement subject after the requirement data of different qualities are processed by the three multi-subject requirement mining methods mentioned above, including one or more of the requirement attention C1, requirement description C2 and requirement value C3; requirement weight D refers to the relative importance of the requirements after weight calculation, which is used to support designers in making subsequent design decisions and resource allocation; S62: Requirement item fusion based on semantic similarity; For repeated or similar requirements, the requirement items are converted into word vectors, and the cosine similarity is calculated to define the semantic similarity between the requirement items. When the semantic similarity is not less than the threshold β, the two requirement items are combined into one, and then the requirement values ​​are merged. When the semantic similarity is less than the threshold β, the two requirement items are retained. S63: Demand value fusion; For the demands of multiple entities, if the demand items and demand values ​​are the same, they will be directly merged into one item, and the demand entities will be multiple demand entities; if the demand items are the same but the demand values ​​are different, the one with a higher overall weight of the multiple entity demands will be selected to cover the other.

8. The complex product multi-agent demand mining and importance analysis method according to claim 7, characterized in that: Step S7 specifically includes the following steps: S71: Relative weight calculation among multiple subjects based on improved Kano; Combined with hesitant fuzzy sets, an improved Kano model is constructed to calculate the relative weights among multiple agents; S72: Calculation of technical requirement weights based on information entropy and fuzzy DEMATEL; Combine information entropy and fuzzy DEMATEL method to analyze the mutual influence relationship between technical requirements, obtain the importance of the requirements, and calculate the comprehensive requirement weight; S73: User demand weight calculation based on emotion-cloud model; Based on the emotion-cloud model, the user's emotional preferences are quantified to achieve the weight calculation of user needs. The process includes: S731: Sentiment analysis; Analyze subjective texts with emotional preferences through sentiment dictionaries, subdivide the emotional intensity from positive to negative into seven levels, and correspond emotional degree adverbs to quantifiers; calculate sentiment scores by considering the number of emotional words in the text, the total number, and the frequency of degree adverbs; S732: Conversion between language information and cloud; Calculate adjustment coefficients based on a given set of language terms, build a multi-level cloud model, and specify upper and lower bounds of the interval; S733: Calculate user demand weights; use the constructed cloud model to process the data, calculate the subjective weight of the demand according to the middle value of the interval number of the cloud model, calculate the objective weight according to the interval width, and calculate the user demand satisfaction by combining subjective and objective calculations; calculate and normalize the demand attention to obtain the user demand weight; S74: Multi-agent demand weight fusion based on combined weighting; The construction of the improved Kano model to calculate the relative weights between multiple agents includes the following steps: S711: Preparation for improving the Kano model; combining the Kano model with hesitant fuzzy sets to establish an improved Kano questionnaire; combining the five types of demand, namely basic type, expected type, exciting type, irrelevant type and reverse type, to develop a Kano evaluation table; S712: Construct a possibility matrix; obtain survey matrices P and N based on the results of the improved Kano questionnaire, and calculate the interaction matrix S; calculate the possibility vector Y of each demand category based on the Kano evaluation table; combine to obtain the possibility matrix T of multiple demand subjects for demand; S713: constructing a similarity matrix between demand subjects; calculating the similarity of the possibility vectors between demand subjects by means of cosine similarity; combining the similarities between all demand subjects to obtain a similarity matrix; S714: Calculate the relative weights between demand entities; sum the similarity matrix to obtain the relative weights of other demand entities to demand entity A j The overall similarity of other demand entities indicates that demand entity A j The support degree of the demand subject A is obtained by normalization j The credibility of the demand subject is calculated to obtain the relative weight θ Aj ; The technology requirement weight calculation based on information entropy and fuzzy DEMATEL combines information entropy and fuzzy DEMATEL methods to analyze the mutual influence relationship between technology requirements, obtain the importance of the requirements, and calculate the comprehensive requirement weight by combining information entropy; the comprehensive requirement weight calculation by combining information entropy includes the following steps: S721: Calculate the initial weight of the technology requirement based on the information entropy; calculate the information entropy value E of the technology requirement based on the possibility vector Y obtained in step S712, and further calculate the initial weight ω of the technology requirement v ; S722: Initial evaluation of the impact relationship between requirements; using a 5-level scale language term ID, namely, huge impact {E: (0.7, 0.9, 1.0)}, high impact {H: (0.5, 0.7, 0.9)}, medium impact {M: (0.3, 0.5, 0.7)}, low impact {E: (0.1, 0.3, 0.5)}, no impact {E: (0.0, 0.1, 0.3)}; by analyzing the internal impact relationship between technical requirements, construct an initial evaluation matrix IM; S723: construct a comprehensive influence matrix; combine the triangular fuzzy number to process the initial evaluation matrix IM, obtain the fuzzy direct influence matrix M, normalize it to obtain the normalized influence matrix N, then calculate the fuzzy comprehensive influence matrix, and obtain the clear comprehensive influence matrix T after defuzzification; S724: Calculate influencing factors; calculate the influence degree D based on the comprehensive influence matrix v and the degree of influence C v ; Further calculation to obtain the centrality M v and the degree of cause R v ; Further calculate the impact importance of demand η v ; S725: Draw an impact relationship network diagram of the technical requirements; divide the requirements into a cause group and an effect group according to the above calculation results; draw an impact relationship network diagram of the technical requirements, with arrows pointing from the cause group requirements to the effect group requirements, and mark the initial weights corresponding to the technical requirements; S726: Construct a comprehensive weight function; consider the direct impact of demand on the initial weight, as well as the promoting or inhibiting effect of the indirect impact of demand on the demand weight, construct a weight function of technical demand based on the calculation results, and obtain the final technical demand weight.

9. The complex product multi-agent demand mining and importance analysis method according to claim 8, characterized in that: The requirements mining and importance analysis results collation include the following steps: Returning to step S63, according to the calculated overall weight of the multi-subject demands, for the case where the demand items are the same but the demand values ​​are different, the multi-subject demand with a higher overall weight is selected to cover the other to obtain the final demand mining result.

Images