Method, device and equipment for predicting product design parameters based on user needs

Abstract

This application discloses a method, apparatus, and device for predicting product design parameters based on user needs, relating to the field of natural language processing technology. The method comprises: extracting a first user need factor from user need data; inputting the first user need factor into a prediction model, and predicting the product design parameters corresponding to the first user need factor. This solution addresses the existing problem of excessive reliance on human intervention in the product design process.

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to a method, device and apparatus for predicting product design parameters based on user needs. Background Art

[0002] Innovation plays a vital role in the overall process of social development, and product innovation is inseparable from product design. From the perspective of the product's entire life cycle, product design is an important part of new product development, and the concept design stage is the top priority of the product design process.

[0003] The development of product concept design has gone through several stages. In traditional design methods, the success or failure of a design proposal relies heavily on the designer's knowledge and experience, leading to considerable uncertainty in product design quality. With the rise of productivity, better understanding user needs and translating them into product design parameters has become crucial for capturing market share and winning the competition. Consequently, product design, quality control, and quality improvement tools such as Quality Flow Diagram (QFD) have emerged. Since then, demand-driven product design has become the mainstream approach in the design field.

[0004] Currently, the internet is increasingly integrated into social life, information technology continues to improve, and e-commerce and crowdsourcing platforms are booming. Against this backdrop, the availability of user needs through various channels is increasing, while computers' ability to store and compute data has greatly increased. Consequently, a vast amount of user needs is constantly accumulating in the form of data. This makes efficiently and accurately analyzing this user need data and developing a universal design theory a significant challenge.

[0005] Currently, within the field of design science both domestically and internationally, several design theories have been developed to continuously optimize and standardize the product concept design process, such as the classic systematic design theory, TRIZ theory, and universal design theory. While these theories each have their unique advantages, they also face their own shortcomings and limitations when faced with the real challenges of massive demand data.

[0006] In summary, in today's digital society, the fields of product design and demand analysis face new opportunities and challenges. Finding an efficient and accurate method for analyzing user needs, leveraging machine learning algorithms, big data management and analysis techniques, deep learning models, and other technologies to develop a new set of intelligent conceptual design theories, is crucial and far-reaching. Summary of the Invention

[0007] The purpose of this application is to provide a method, device and equipment for predicting product design parameters based on user needs, thereby solving the problem of excessive reliance on human participation in the product design process in the prior art.

[0008] To achieve the above objectives, the present application provides a method for predicting product design parameters based on user needs, comprising:

[0009] Extracting a first user demand element from the user demand data;

[0010] The first user demand factor is input into a prediction model to predict product design parameters corresponding to the first user demand factor.

[0011] Optionally, extracting a first user demand element from the user demand data includes:

[0012] Preprocessing the user demand data;

[0013] Perform data labeling on pre-processed user demand data;

[0014] The BERT model is used to extract the first user demand element from the user demand data after data annotation.

[0015] Optionally, preprocessing the user demand data includes:

[0016] Delete invalid user demand data;

[0017] Perform word segmentation on the user demand data that has not been deleted.

[0018] Optionally, the training process of the BERT model includes:

[0019] Obtain a first training data set;

[0020] The first training data set is input into a pre-trained BERT model, and the pre-trained BERT model is iteratively trained to obtain the BERT model.

[0021] Optionally, the training process of the prediction model includes:

[0022] Obtaining a second user demand factor and product parameter training data set;

[0023] Analyzing the second user demand factor to obtain a user demand data set;

[0024] Based on a random forest algorithm, extracting a preset number of feature parameters related to the user demand dataset from the product design parameter training dataset, and calculating a first weight of each of the feature parameters;

[0025] According to the characteristic parameters and the first weight, the trained prediction model is trained using a logistic regression algorithm to obtain the prediction model.

[0026] Optionally, obtaining the second user demand factor includes:

[0027] Obtain user demand training data set;

[0028] Preprocessing the user demand training data set;

[0029] Performing data labeling on the preprocessed user demand training data set;

[0030] The labeled user demand training dataset is input into the trained BERT model to extract the second user demand element.

[0031] Optionally, analyzing the second user demand factor to obtain a user demand data set includes:

[0032] Calculate the TF-IDF representing the importance of the user demand factor using the TF-IDF algorithm;

[0033] Generating text topics corresponding to the user demand factors using a latent Dirichlet allocation (LDA) algorithm;

[0034] Dividing the text topics into levels;

[0035] The user demand dataset is obtained according to the hierarchically divided text topics and the TF-IDF.

[0036] Optionally, the method further includes:

[0037] Constructing a user demand map based on the user demand data set;

[0038] The user demand map is displayed.

[0039] Optionally, based on a random forest algorithm, a preset number of feature parameters related to the user demand dataset are extracted from the product design parameter training dataset, and a first weight of each feature parameter is calculated, including:

[0040] Calculating a second weight of each product design parameter in the product design parameter training set based on a random forest algorithm;

[0041] Deleting the product design parameters with the second smallest weight among the product design parameters according to a preset ratio;

[0042] cyclically executing the random forest function to calculate the third weights of the product design parameters that have not been deleted, and deleting the product design parameters with smaller third weights according to the preset ratio;

[0043] When the number of the undeleted product design parameters reaches the preset number, the undeleted product design parameters are determined to be the feature parameters, and the weight of the undeleted product design parameters is the first weight.

[0044] The present application also provides an apparatus for predicting product design parameters based on user needs, including:

[0045] An extraction module, configured to extract a first user demand element from the user demand data;

[0046] The prediction module is used to input the first user demand factor into the prediction model to predict the product design parameters corresponding to the first user demand factor.

[0047] An embodiment of the present application also provides a device for predicting product design parameters based on user needs, including: a processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the method for predicting product design parameters based on user needs as described above is implemented.

[0048] An embodiment of the present application further provides a readable storage medium having a program stored thereon, and when the program is executed by a processor, the method for predicting product design parameters based on user needs as described above is implemented.

[0049] The above technical solution of the present application has at least the following beneficial effects:

[0050] The method for predicting product design parameters based on user demand in an embodiment of the present application first extracts a first user demand element from user demand data; second, inputs the first user demand element into a prediction model to predict the product design parameters corresponding to the first user demand element. This allows for the precise extraction and analysis of user demand information from user demand data, mapping user demand to product design parameter requirements, automating product concept design, improving product concept design efficiency, and reducing the product design process's reliance on human factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of a flow chart of a method for predicting product design parameters based on user needs according to an embodiment of the present application;

[0052] Figure 2 This is an example diagram of analyzing user demand factors in an embodiment of the present application;

[0053] Figure 3 This is a schematic diagram of the structure of an apparatus for predicting product design parameters based on user needs according to an embodiment of the present application. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0055] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects and are not used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments of this application can be implemented in an order other than those illustrated or described herein. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0056] The following, in conjunction with the accompanying drawings, describes in detail the method, device and equipment for predicting product design parameters based on user needs provided by the embodiments of the present application through specific embodiments and their application scenarios.

[0057] like Figure 1 FIG. 1 is a flow chart of a method for predicting product design parameters based on user needs according to an embodiment of the present application, the method comprising:

[0058] Step 101: extracting a first user demand element from user demand data;

[0059] Here, it should be noted that the user demand data is presented in the form of text in natural language. Specifically, the user demand data can be obtained from the web, or from online user demand data submitted by users of crowdsourcing websites. If the data volume is small or the data volume is poor, it can be replaced by the evaluation text of users after purchasing products on e-commerce websites.

[0060] Step 102: Input the first user demand factor into a prediction model to predict product design parameters corresponding to the first user demand factor.

[0061] In this step, the prediction model is a pre-trained model, specifically a model trained based on a logistic regression algorithm.

[0062] The method for predicting product design parameters based on user demand in an embodiment of the present application first extracts a first user demand element from user demand data; second, inputs the first user demand element into a prediction model to predict the product design parameters corresponding to the first user demand element. This allows for the precise extraction and analysis of user demand information from user demand data, mapping user demand to product design parameter requirements, automating product concept design, improving product concept design efficiency, and reducing the product design process's reliance on human factors.

[0063] As an optional implementation, step 101, extracting a first user demand element from the user demand data, includes:

[0064] Preprocessing the user demand data;

[0065] This step can reduce the interference of useless information on the entire dental process by preprocessing the user demand data and prepare for the labeling of the user demand data.

[0066] Perform data labeling on pre-processed user demand data;

[0067] This step uses the "BIO" annotation method to annotate user demand data. B, I, and O are the abbreviations of begin, intermediate, and others, respectively, marking the beginning, middle, and other parts of the annotated content. The annotated demand elements are all from the user's perspective, and other useless information is not annotated. The data annotation examples are shown in the following table:

[0068] text Annotation text Annotation text Annotation have O pair B noise B speed B open I sound I freeze I Door I very I achievement O ， O Small I able O Pick B ， O ， O thing I No B null B Taste I film I between I very I ring I enough I square I stop I big I Convenience I interest I 。 O 。 O 。 O

[0069] Table 1

[0070] The BERT model is used to extract the first user demand element from the user demand data after data annotation.

[0071] It should be noted here that the BERT model is a pre-trained model. The full name of BERT is Bidirectional Encoder Representation from Transformers. The BERT model in this step is a trained model. The training process of BERT will be explained in detail later.

[0072] It's also worth noting that BERT uses a bidirectionally encoded Transformer architecture, consisting of multiple layers of Transformer encoding blocks. BERT offers both simple and complex models: BERTBASE, which includes 12 layers of Transformer encoding blocks, and BERTLARGE, which includes 24 layers of Transformer encoding blocks. The sentence input from the encoder first passes through a self-attention layer, which helps the encoder focus on other words in the input sentence when encoding each word, while reducing model complexity. The output of the self-attention layer is passed to a feed-forward neural network. The feed-forward neural network corresponding to each word in each position is exactly the same, or in other words, each layer window is a one-dimensional convolutional neural network for each word.

[0073] In this optional implementation, by preprocessing the user demand data, annotating the preprocessed user demand data, and finally using the trained BERT model to extract the first user demand data from the annotated user demand data, it is possible to accurately extract and analyze user demand information from text-based user demand data, thereby reducing dependence on human factors.

[0074] Furthermore, after extracting the first user demand factor, the method further includes:

[0075] Evaluate the BERT model based on preset indicators and obtain evaluation results;

[0076] The preset indicators include at least one of the following:

[0077] Precision rate;

[0078] recall rate;

[0079] F1 value.

[0080] Here, evaluating the BERT model allows users to understand the quality of the BERT model's results in extracting user demand factors, thereby further determining the accuracy of the predicted product design parameters.

[0081] As an optional implementation, preprocessing the user demand data includes:

[0082] Delete invalid user demand data;

[0083] Here, it should be noted that invalid user demand data includes non-user demand data, missing values, etc. By deleting these data, the accuracy of data processing can be improved and the interference of useless information on the entire research process can be reduced.

[0084] Perform word segmentation on the user demand data that has not been deleted.

[0085] In this step, the specific process of word segmentation includes:

[0086] First, extract the column storing user requirements from the data. Each element in the column is a user requirement.

[0087] Secondly, add an ID to the header of each user demand data; the purpose of adding the ID is to facilitate the identification of the mapping relationship between user demand data and product design parameters;

[0088] Again, each piece of user demand data is segmented by word or single word;

[0089] Finally, the user requirements after word segmentation are wrapped word by word, that is, each word or character occupies one line in the document.

[0090] The main purpose of word segmentation of user demand data in this step is to prepare for the next step of data labeling, so as to facilitate the labeling of user demand data.

[0091] As an optional implementation, the training process of the BERT model includes:

[0092] Obtain a first training data set;

[0093] Here, it should be noted that the first training data set is user demand data preprocessed according to the aforementioned method.

[0094] The first training data set is input into a pre-trained BERT model, and the pre-trained BERT model is iteratively trained to obtain the BERT model.

[0095] As mentioned above, the BERT model is a pre-trained model. The BERT model provides a raw pre-trained model. By inputting personalized data and fine-tuning the model parameters, satisfactory results can be achieved. Because manually labeling large amounts of data is labor-intensive, and small sample sizes cannot cover a large amount of data, this optional implementation uses iterative training to train the model. Specifically, the first training dataset is repeatedly fed into the BERT model to train it. This reduces the manual data processing workload and improves training efficiency.

[0096] For example, if the first training data parameter set includes "I need frozen meat, and it must have low power consumption", the demand factors extracted by the BERT model after iterative training are "frozen meat" and "low power consumption".

[0097] It should also be noted that during the iterative training of the BERT model, the BERT model can be further evaluated based on preset indicators to obtain evaluation results.

[0098] If the evaluation result does not meet the preset conditions, iteratively train the BERT model using the second training dataset until the evaluation result meets the preset conditions;

[0099] The preset indicators include at least one of the following:

[0100] Precision rate;

[0101] recall rate;

[0102] F1 value.

[0103] During the BERT model training process, by evaluating the BERT model, the operator can understand the current status of BERT, thereby providing guidance for the subsequent training process and improving training efficiency.

[0104] As an optional implementation, the training process of the prediction model includes:

[0105] Obtaining a second user demand factor and product parameter training data set;

[0106] In this step, the second user demand factor can be a user demand factor extracted using the extraction method of the above-mentioned embodiment of the present application; the source of the product parameter training data set is the standardized description of product design parameters on the e-commerce website. For example, for refrigerator products, its product design parameters include: brand, size, weight, refrigeration type, total volume, operating volume, energy efficiency level, defrost mode, fixed frequency / variable frequency, refrigerant, door opening structure, etc.

[0107] Analyzing the second user demand factor to obtain a user demand data set;

[0108] Here, it should be noted that since the second user demand input obtained by the above method is disorganized and has no rules to follow, and the product design parameters are constrained by various conditions (such as cost, industry standards and specifications), it is not realistic to convert all demands one by one into product design parameters. Therefore, the purpose of this step is to process user demands and find the rules in the demand elements (the importance of the demand, the category of the demand) in order to further utilize the demand data.

[0109] Based on a random forest algorithm, extracting a preset number of feature parameters related to the user demand dataset from the product design parameter training dataset, and calculating a first weight of each of the feature parameters;

[0110] Random forest is an algorithm that integrates multiple trees through the idea of ​​ensemble learning. The basic unit of random forest is the decision tree. Given that random forest has a wide range of uses and performs well in machine learning tasks, and as an integration of tree models, it can also be used for feature selection. In the embodiment of the present application, the random forest algorithm is used to realize the selection of multiple features in product design parameters. The basis for selection is the importance of the feature to the degree of satisfaction of user needs.

[0111] Here, it should be noted that after this step, each type of user demand will obtain several most relevant feature parameters corresponding to it.

[0112] According to the characteristic parameters and the first weight, the trained prediction model is trained using a logistic regression algorithm to obtain the prediction model.

[0113] Here, it should be noted that the probability distribution formula of the logistic regression model is as follows:

[0114]

[0115] The above formula means to find the probability of y=1 under the given conditions of x and θ, where x is the input and θ is the specified parameter.

[0116] As an optional implementation method, obtaining the second user demand factor includes:

[0117] Obtain user demand training data set;

[0118] Here, it should be noted that the user demand training data set can be user demand data collected in advance from the web end;

[0119] Preprocessing the user demand training data set;

[0120] Here, it should be noted that the preprocessing in this step is similar to the aforementioned preprocessing process and will not be repeated here.

[0121] Performing data labeling on the preprocessed user demand training data set;

[0122] Similarly, this step can be marked using the "BIO" marking method.

[0123] The labeled user demand training dataset is input into the trained BERT model to extract the second user demand element.

[0124] Here, it should be noted that each step of this optional implementation method can be performed in the manner of the aforementioned processing process. In order to avoid repetition, it will not be described here.

[0125] As an optional implementation, analyzing the second user demand factor to obtain a user demand data set includes:

[0126] Calculating the TF-IDF that represents the importance of the user demand factor using a term frequency-inverse document frequency (TF-IDF) algorithm;

[0127] It should be noted that the TF-IDF algorithm is a commonly used weighting technique in information retrieval and text mining. It is a reliable statistical method whose indicators can be used to measure the relative importance of a word or phrase in a document in a corpus. Therefore, in the task of evaluating the importance of requirements, the embodiments of this application use the TF-IDF algorithm to measure the relative importance of user requirements.

[0128] The main logic of the TF-IDF algorithm is: if the number of times a word appears in a document increases, its importance increases; if the frequency of the word in the entire corpus increases, its importance decreases. The TF-IDF algorithm works as follows:

[0129] Term Frequency (TF) indicates the frequency with which a term appears in a text.

[0130]

[0131] Right now:

[0132] Where n ij For entry n i The number of times a term appears in document j. The denominator represents the number of times all terms appear.

[0133] The Inverse Document Frequency (IDF) is calculated by dividing the total number of documents by the number of documents containing the term and taking the logarithm of the quotient. If the fewer documents containing a term t, the higher the IDF, this indicates that term t has good class differentiation capabilities.

[0134]

[0135] Right now:

[0136] TF-IDF is the product of TF and IDF, namely:

[0137] TF-IDF=TF*IDF

[0138] In the task of evaluating the importance of user needs, the TF-IDF algorithm is used to measure the relative importance of user needs. This algorithm actually targets the desired needs described in the KANO model. These are needs that change proportionally with changes in the indicator. While users generally don't express desirable and essential needs, inverse needs can be transformed into desired needs through semantic inversion. Therefore, this embodiment of the application only targets users' desired needs.

[0139] Generating text topics corresponding to the user demand factors using Latent Dirichlet Allocation (LDA) algorithm;

[0140] This step specifically uses the extracted user demand elements as input and generates user demand topics using the LDA algorithm. The LDA algorithm is an unsupervised text topic model used to generate topics within a document. LDA considers a document to be a probability distribution of topics, and each topic is a probability distribution of words.

[0141] Here, it should be noted that the Dirichlet distribution is the conjugate distribution of the multinomial distribution, and its probability density function is:

[0142]

[0143] in,

[0144] Dividing the text topics into levels;

[0145] It's important to note that in Quality Function Deployment (QFD), user needs are summarized, consolidated, and categorized, with user needs divided into two or three levels depending on the specific scenario. In QFD, this step is implemented using affinity diagrams (KJ method). Therefore, this step involves categorizing user needs into several broad categories through induction and summarization. The needs of each category are then summarized, dividing user needs into two tiers, thereby achieving a hierarchical analysis of user needs.

[0146] The user demand dataset is obtained according to the hierarchically divided text topics and the TF-IDF.

[0147] It should be noted that in the QFD product design process, there are two methods for analyzing user needs: using questionnaires and the Analytic Hierarchy Process (AHP) to measure the importance of user needs, and using affinity diagrams to classify user needs. Similarly, these two methods are also used in the embodiments of this application to analyze user needs. The TF-IDF algorithm is mainly used to evaluate the importance of user needs, and the LDA algorithm is mainly used in the hierarchical analysis of user needs.

[0148] like Figure 2 As shown, the second user demand factors include not being too noisy, low power consumption, being able to accommodate amplifiers, large volume, low noise, first-level energy efficiency, power saving, large capacity, and being quiet; after analysis using this optional implementation method, the user demand data set obtained includes: user demand data with the theme of "low noise", including user demand factors such as "not being too noisy, low noise, and being quiet"; user demand data with the theme of "low power consumption", including user demand factors such as "low power consumption, first-level energy efficiency, and being power-saving"; user demand data with the theme of "large capacity", including user demand factors such as "being able to accommodate amplifiers, large volume, and being large"; it can be seen that by using the analysis method of this optional implementation method, the disorganized and irregular second user demand factors can be summarized into a user demand data set with clear hierarchy and importance classification.

[0149] Furthermore, as an optional implementation, the method further includes:

[0150] Constructing a user demand map based on the user demand data set;

[0151] The user demand map is displayed.

[0152] That is to say, after analyzing the importance and hierarchy of user needs, a user needs map is constructed with reference to semantic similarity to visually display user needs.

[0153] In this optional implementation, the purpose of constructing a user needs map for products in a specific field is to transform the colloquial, unstructured, and massive user needs into a clearly hierarchical, clearly prioritized, and intuitive visual map. This helps designers better understand user needs and achieve higher user satisfaction with product design. Taking into account importance and hierarchical relationships, the demand elements are visualized in the form of a needs map, thereby more intuitively assisting designers in understanding the most important and core user needs in the product field, as well as the demand points, hot spots, and pain points that require attention. The user needs analysis is completed through the application of the TF-IDF and LDA algorithms. The needs analysis results are then visualized in the form of a user needs map.

[0154] As an optional implementation, based on a random forest algorithm, a preset number of feature parameters related to the user demand dataset are extracted from the product design parameter training dataset, and a first weight of each feature parameter is calculated, including:

[0155] (A) calculating a second weight of each product design parameter in the product design parameter training set based on a random forest algorithm;

[0156] Here, it should be noted that the workflow of random forest is as follows:

[0157] Construct a forest of decision trees in a random manner, where each decision tree in the forest is independent and unrelated to each other.

[0158] After obtaining the random forest, a new sample is input, and each decision tree classifies the sample (for classification algorithms);

[0159] The classification results of each decision tree are obtained, and the category with the most classification results is the predicted category of this sample.

[0160] Random forests have a wide range of applications, working with both discrete-valued attributes, such as the ID3 algorithm, and continuous-valued attributes, such as the C4.5 algorithm. Furthermore, random forests can be used for unsupervised clustering and outlier detection, with excellent performance.

[0161] During product design, a product may have hundreds or even thousands of features, and more for more complex products. We need to select those features that have the greatest impact on user satisfaction to reduce the number of features used in model building. There are many methods for calculating correlations, such as principal component analysis and the Pearson correlation coefficient. This method uses random forests, which have a better averaging effect, for correlation calculation.

[0162] The idea behind calculating feature relevance using random forests is to compare the contribution of each feature to each decision tree in the random forest, take the average, and then compare the contributions of each feature. The Gini impurity index or the out-of-bag (OOB) error rate is often used to evaluate feature contribution. The formula for the Gini impurity index is as follows:

[0163]

[0164] (B) deleting, according to a preset ratio, the product design parameters having the second smallest weight among the product design parameters;

[0165] (C) looping and executing the random forest function to calculate the third weights of the product design parameters that have not been deleted, and deleting the product design parameters with smaller third weights according to the preset ratio;

[0166] (D) When the number of the undeleted product design parameters reaches the preset number, determining the undeleted product design parameters as the feature parameters, and the weight of the undeleted product design parameters as the first weight.

[0167] The specific implementation process of the above steps (B) to (D) can be:

[0168] Sort the feature variables (product design parameters) in the random forest in descending order according to the feature importance (second weight) calculated in the feature importance step;

[0169] Determine the removal ratio, remove relatively unimportant indicators from the current feature variables, and then obtain a new feature set;

[0170] Use the obtained new feature set to build a new random forest, calculate the importance of each feature in the feature set and sort them in descending order.

[0171] Repeat the above steps until the required number (preset number) of features remain.

[0172] like Figure 3 As shown, the embodiment of the present application further provides a device for predicting product design parameters based on user needs, including:

[0173] Extraction module 301, used to extract the first user demand element from the user demand data;

[0174] The prediction module 302 is configured to input the first user demand factor into a prediction model to predict the product design parameters corresponding to the first user demand factor.

[0175] In the device for predicting product design parameters based on user needs, an extraction module 301 extracts a first user need element from user need data. A prediction module 302 then inputs the first user need element into a prediction model to predict the product design parameters corresponding to the first user need element. This allows for the precise extraction and analysis of user need information from user need data, mapping user needs to product design parameter requirements, and automating product concept design, improving its efficiency and reducing the reliance on human factors in the product design process.

[0176] The extraction module 301 includes:

[0177] A first processing submodule, configured to pre-process the user demand data;

[0178] The second processing submodule is used to perform data labeling on the pre-processed user demand data;

[0179] The extraction submodule is used to extract the first user demand element from the user demand data after data annotation by using the BERT model.

[0180] The first processing submodule includes:

[0181] A first processing unit, configured to delete invalid user demand data;

[0182] The second processing unit is used to perform word segmentation processing on the user demand data that has not been deleted.

[0183] The device further includes a first training module, which includes:

[0184] A first acquisition submodule is used to acquire a first training data set;

[0185] The first training submodule is used to input the first training data set into a pre-trained BERT model, iteratively train the pre-trained BERT model, and obtain the BERT model.

[0186] The device further includes a second training module, which includes:

[0187] A second acquisition submodule is used to obtain a second user demand factor and a product parameter training data set;

[0188] A third acquisition submodule is configured to analyze the second user demand factor to obtain a user demand data set;

[0189] a third processing submodule, configured to extract, from the product design parameter training dataset, a preset number of feature parameters related to the user demand dataset based on a random forest algorithm, and calculate a first weight of each of the feature parameters;

[0190] The second training submodule is used to train the trained prediction model using a logistic regression algorithm according to the feature parameters and the first weight to obtain the prediction model.

[0191] The second acquisition submodule includes:

[0192] A first acquisition unit is used to acquire a user demand training data set;

[0193] A third processing unit, configured to preprocess the user demand training data set;

[0194] A fourth processing unit, configured to perform data labeling on the preprocessed user demand training data set;

[0195] The extraction unit is used to input the user demand training data set after data annotation into the trained BERT model to extract the second user demand element.

[0196] The second acquisition submodule includes:

[0197] A first calculation unit is used to calculate the TF-IDF representing the importance of the user demand factor by using a term frequency inverse document frequency TF-IDF algorithm;

[0198] A generating unit, configured to generate a text topic corresponding to the user demand factor by using a latent Dirichlet allocation (LDA) algorithm;

[0199] a fifth processing unit, configured to hierarchically divide the text topics;

[0200] The second acquisition unit is configured to acquire the user demand dataset according to the hierarchically divided text topics and the TF-IDF.

[0201] Wherein, the device further includes:

[0202] A construction module, configured to construct a user demand map based on the user demand dataset;

[0203] A display module is used to display the user demand map.

[0204] Wherein, the third processing submodule includes:

[0205] a second calculation unit, configured to calculate a second weight of each product design parameter in the product design parameter training set based on a random forest algorithm;

[0206] a sixth processing unit, configured to delete, according to a preset ratio, product design parameters having a smaller second weight among the product design parameters;

[0207] a seventh processing unit, configured to cyclically execute the random forest function to calculate the third weights of the product design parameters that have not been deleted, and delete the product design parameters with smaller third weights according to the preset ratio;

[0208] The determining unit is configured to determine, when the number of the undeleted product design parameters reaches the preset number, that the undeleted product design parameters are the characteristic parameters, and the weights of the undeleted product design parameters are the first weights.

[0209] An embodiment of the present application also provides a device for predicting product design parameters based on user needs, including: a processor, a memory, and a program stored on the memory and executable on the processor. When the program is executed by the processor, the various processes of the method embodiment for predicting product design parameters based on user needs as described above are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0210] The present application also provides a readable storage medium having a program stored thereon. When the program is executed by a processor, the various processes of the method embodiment for predicting product design parameters based on user needs as described above are implemented, and the same technical effects are achieved. To avoid repetition, the description is omitted here. The readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0211] Finally, it should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0212] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting product design parameters based on user needs, characterized in that: include: Extracting a first user demand element from the user demand data; Inputting the first user demand factor into a prediction model to predict product design parameters corresponding to the first user demand factor; The prediction model is a model obtained by training based on a logistic regression algorithm, and the training process of the prediction model includes: Obtaining a second user demand factor and product design parameter training data set; Analyzing the second user demand factor to obtain a user demand data set; Based on a random forest algorithm, extracting a preset number of feature parameters related to the user demand dataset from the product design parameter training dataset, and calculating a first weight of each of the feature parameters; Training a pre-trained prediction model using a logistic regression algorithm according to the feature parameter and the first weight to obtain the prediction model; Analyzing the second user demand factor to obtain a user demand data set includes: Calculating the TF-IDF representing the importance of the second user demand factor using a TF-IDF algorithm; Generating a text topic corresponding to the second user demand factor using a latent Dirichlet allocation (LDA) algorithm; Dividing the text topics into levels; According to the hierarchically divided text topics and the TF-IDF, a user demand dataset with clear hierarchies and importance classification is obtained; Wherein, based on the random forest algorithm, a preset number of feature parameters related to the user demand dataset are extracted from the product design parameter training dataset, and a first weight of each of the feature parameters is calculated, including: Calculating a second weight of each product design parameter in the product design parameter training set based on a random forest algorithm; Deleting the product design parameters with the second smallest weight among the product design parameters according to a preset ratio; cyclically executing the random forest algorithm to calculate the third weights of the product design parameters that have not been deleted, and deleting the product design parameters with smaller third weights according to the preset ratio; When the number of the undeleted product design parameters reaches the preset number, the undeleted product design parameters are determined to be the feature parameters, and the weight of the undeleted product design parameters is the first weight.

2. The method according to claim 1, characterized in that From the user demand data, extract the first user demand element, including: Preprocessing the user demand data; Perform data labeling on pre-processed user demand data; The BERT model is used to extract the first user demand element from the user demand data after data annotation.

3. The method according to claim 2, characterized in that Preprocessing the user demand data includes: Delete invalid user demand data; Perform word segmentation on the user demand data that has not been deleted.

4. The method according to claim 2, characterized in that The training process of the BERT model includes: Obtain a first training data set; The first training data set is input into a pre-trained BERT model, and the pre-trained BERT model is iteratively trained to obtain the BERT model.

5. The method according to claim 1, wherein Obtain the second user demand factors, including: Obtain user demand training data set; Preprocessing the user demand training data set; Performing data labeling on the preprocessed user demand training data set; The labeled user demand training dataset is input into the trained BERT model to extract the second user demand element.

6. The method according to claim 1, wherein The method further comprises: Constructing a user demand map based on the user demand data set; The user demand map is displayed.

7. A device for predicting product design parameters based on user needs, characterized in that: include: An extraction module, configured to extract a first user demand element from the user demand data; A prediction module, configured to input the first user demand factor into a prediction model to predict product design parameters corresponding to the first user demand factor; wherein the prediction model is a model trained based on a logistic regression algorithm; The device further includes a second training module, which includes: A second acquisition submodule is used to obtain a second user demand factor and a product design parameter training data set; A third acquisition submodule is configured to analyze the second user demand factor to obtain a user demand data set; a third processing submodule, configured to extract, from the product design parameter training dataset, a preset number of feature parameters related to the user demand dataset based on a random forest algorithm, and calculate a first weight of each of the feature parameters; a second training submodule, configured to train the pre-trained prediction model using a logistic regression algorithm according to the feature parameters and the first weights to obtain the prediction model; Wherein, the third acquisition submodule includes: A first calculation unit is used to calculate the TF-IDF representing the importance of the second user demand factor by using a term frequency-inverse document frequency TF-IDF algorithm; a generating unit, configured to generate a text topic corresponding to the second user demand factor by using a latent Dirichlet allocation (LDA) algorithm; a fifth processing unit, configured to hierarchically divide the text topics; A second acquisition unit is configured to acquire the user demand dataset with clear hierarchies and importance classifications according to the hierarchically divided text topics and the TF-IDF; Wherein, the third processing submodule includes: a second calculation unit, configured to calculate a second weight of each product design parameter in the product design parameter training set based on a random forest algorithm; a sixth processing unit, configured to delete, according to a preset ratio, product design parameters having a smaller second weight among the product design parameters; a seventh processing unit, configured to cyclically execute the random forest algorithm to calculate the third weights of the product design parameters that have not been deleted, and delete the product design parameters with smaller third weights according to the preset ratio; The determining unit is configured to determine, when the number of the undeleted product design parameters reaches the preset number, that the undeleted product design parameters are the characteristic parameters, and the weights of the undeleted product design parameters are the first weights.

8. A device for predicting product design parameters based on user needs, characterized in that: include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein when the program is executed by the processor, the method for predicting product design parameters based on user needs as described in any one of claims 1 to 6 is implemented.

9. A readable storage medium, characterized in that The readable storage medium stores a program, and when the program is executed by the processor, the method for predicting product design parameters based on user needs according to any one of claims 1 to 6 is implemented.

Images