Innovative assisted design of accessible products based on multi-dimensional perceptual information association rule extraction

Due to the physical deterioration of the elderly with age and the physical impairment of the disabled, this group of people not only need to be taken care of, but also need appropriate assistive products and equipment to help them to improve their ability to live and quality of life [1–3]. With the rapid development of China’s economy, the variety of products for the elderly and disabled has been enriched and developed, and the demand for such products is still expanding.

In the process of product design, it is often necessary to comprehensively consider elements such as product function, material properties, experience needs, and appearance to ensure that the products are safe, practical, comfortable, and aesthetically pleasing [4–5]. This process is a comprehensive test for designers, such as product function positioning, demand decomposition, aesthetic ability, etc., and the differences in the method and process of designers’ decomposition of demands into elements cause product differences [6–8]. This problem is particularly prominent in the design of barrier-free and age-friendly products. It is mainly due to the diversity of the conditions of the elderly and the disabled, such as the deterioration of the physical functions of the elderly caused by the decline in visual and auditory perception, slowing down of motor reactions, etc., and these conditions can occur singly or simultaneously in a number of cases, resulting in a variety of product requirements, and the cross-cutting and complexity of the decomposition of the requirements into elements increases [9–12]. This has led to the concept of accessible design, which is a design used to reduce or eliminate barriers to the user, and it is a reflection of the human-centered design philosophy that will help facilitate the lives of most disadvantaged groups [13–15]. Therefore, at this stage, the development of accessible and age-friendly products can meet the general needs of the elderly and provide the most basic human services, but not “things follow people’s needs”, which will not be recognized by people with disabilities [16–17].

Aiming at the problem of conflicting demand data information and technical cost in accessible product design, this paper designs a mapping model based on association rule perceptual information and product solutions. In order to get a more scientific and efficient judgment method, this paper improves the traditional fuzzy sorting method, considering that the traditional sorting results can not reasonably explain the correlation between product characteristics and customer needs, by combining with the association rule mining algorithm, calculating the fuzzy weighted support degree to carry out the fuzzy set sorting. Finally, the model is used for multidimensional perceptual information accessibility in product design applications and usability testing.

2

Mapping of perceptual information to product solutions based on association rules

2.1

Product perceptual engineering design process

Originally from Japan, the theory of perceptual engineering (KE) is a theory and technology that quantifies the user’s perceptual factors for a product into rational factors and applies them to product design, expressing the user’s emotional needs for the product in the implicit cognition.

The product design process based on perceptual engineering is mainly divided into the determination of design elements, the acquisition of perceptual intention, and the construction and transformation of the relationship between the two and other steps, the specific steps and applications will be explained below.

2.1.1

Determination of design elements

Through perceptual engineering, designers can anticipate consumers’ perceptual perception of products, obtain perceptual design evaluations of products, and then guide their design to create products that fulfill consumers’ spiritual needs. Therefore, selecting the target is important for the effective design of the desired products.

At present, scholars’ research in this field has involved a variety of directions, such as home products, electronic products, transportation, cultural and creative products, medical products, mechanical products, product packaging and other areas of research. For the product shape, color, material and other aspects, there is a perceptual intention to explore. In this research, product design elements can be broken down into a number of basic components and numbered one by one in order to obtain product form parameters. In addition to product form, intangible features such as product configuration and service can also be analyzed.

Determining the design elements is a very important step. Design elements are the basic elements of product perceptual design, including color, material, shape, sound, etc., which have an important impact on the perceptual value of the product. When designing, the selected perceptual elements can be integrated and harmonized with the overall design of the product to obtain visual, tactile and auditory emotional resonance, and to achieve the perceptual goals of comfort and ease of use. Determining the design elements is one of the important contents in the research of product design based on perceptual engineering, which needs to take into account the existing research results, user needs, the application of perceptual tools, product positioning and brand image as well as creative inspiration and other aspects.

2.1.2

Perceptual Intent Acquisition

Designers use perceptual engineering for product design, which is based on the consumer’s emotional intention. Therefore, we need to obtain the consumer’s emotional intention of the target object as a database support. Common measurement methods include the questionnaire survey method, line of sight tracking method, text mining analysis method, and others. 1)

Questionnaire method

Generally use a 7-order Likert scale method to design the questionnaire, the method requires the respondent to a set of statements related to the measurement of the subject to express their views, in a 7-order evaluation scale to make a subjective choice of the target, scale 1 indicates that strongly disagree, scale 4 indicates that there is no opinion, scale 7 indicates that very much agree with the results of the survey to record the target of the respondents on the statement of the statement of the agreement or disagreement degree, the results of the questionnaire can be obtained after organizing the target object perceptual evaluation data, from which the most representative perceptual intentions are filtered out. The results of the survey record the degree of agreement or disagreement of the respondents with the statement of the target, and the results of the questionnaire can be organized to obtain the perceptual evaluation data of the target object, from which the most representative perceptual intention can be screened out.

2)

Line-of-sight tracking method

In the study of perceptual engineering, this measurement method of sight tracking is often used, the main measuring instrument is eye-tracking instrument, which can record the observation process of the subject to the target object, obtain the eye movement data, get the sight trajectory map and hot spot map, find out the subject’s first focus on the target object and the part of the object that has the highest attention level, explore the relationship between eye movement and human psychological activities, and form the consumer’s emotional demand for the product. It can effectively help designers determine the design focus according to consumers’ needs.

3)

Text mining analysis

The online reviews of products on the Internet nowadays contain a large number of consumer evaluations of products, including some subjective emotional evaluations. Taking the online reviews of the products as the data source, the online reviews are captured by the crawler software, and then screened by computer methods, such as the LTP language processing platform, to filter out the irrelevant text, and dig out the analysis of the consumers’ sentimental feelings towards the products.

2.2

Multidimensional perceptual information acquisition

The product market is paying more and more attention to developing products in a way that enhances user satisfaction and improves the core competitiveness of products by responding quickly to users’ individualized needs and changes in market dynamics. Users will have different emotional experiences when interacting with and using products, which is often called the perceptual stage of user perception. In this stage, the user obtains a variety of different feelings of perceptual information through their sensory experiences. Based on the perceptual information can clearly explore the user’s fuzzy emotional appeal, for designers to better grasp the user’s perceptual needs and improve the accuracy of the solution to explore the space to provide favorable help.

At this stage, the collection of user perceptual information is mainly based on manual survey methods such as questionnaires, expert interviews, field visits, etc., with the advantages of ease of use, convenience, flexibility and strong communication. However, there are some problems: (1) the small sample data volume leads to unreliable conclusions; (2) the process of data collection and transcoding field surveys requires a lot of material and financial resources, the utilization rate is low, and it is highly subjective: (3) individual cognition has a timeliness, which leads to the inability to collect the required data in an all-rounded, multi-dimensional and detailed way, and the information obtained has deviations. In view of the limitations of traditional manual data collection, the basic data in the paper utilizes a web crawler to construct a credible multidimensional perceptual information acquisition process. The basic process of multidimensional perceptual information acquisition is shown in Figure 1.1) Initiate a request, through the HTTP library to the target site to initiate a request; 2) Get the response content, if the server responds, it will get the required page information: 4) Get the analysis content in html format, use the webpage analysis library to parse the information; 3) Save the data, the acquired data such as text, pictures, and so on, can be saved to the database.

2.2.1

Data acquisition

The multi-dimensional perceptual information of the user is mainly obtained from the explicit and implicit evaluation of the user’s feelings about the product, and the information data is collected from the six information dimensions of the user’s basic attributes, interest attributes, social attributes, behavioral attributes, psychological attributes, and user value, and the user’s preference and interests are explored through the multi-analysis of the explicit and implicit feedback information in the six horizontal and vertical dimensions, so as to construct a user model, i.e., user labeling. That is user labeling. In this paper, we use an automobile as the experimental research object and carry out experimental validation.

Figure 2 shows the data information of an accessibility product website. The product image and multidimensional information experimental data are derived from the website’s user and product information extracted by a web crawler from January 2018 to December 2020, and 15,708 pieces of user information were obtained by crawling. Its information collected 25 user attribute labels according to the six attribute dimensions of the user, which contains: user age, gender, location, disability level, education level, occupation, income and consumption level, comfort, purchase purpose, purchase price, purchase brand, appearance evaluation, the most satisfied aspects, the least satisfied aspects, recursive coverage relationship as shown in Figure 2.

2.2.2

Data pre-processing

Because of the incomplete and inconsistent information content of the acquired data, in order to ensure the validity of the experimental data, the pre-processing operation is performed on the user information. The main operational steps are divided into four steps: data cleansing, data integration, data generalization, and data conversion. Data cleaning is the process of re-examining and verifying data, aiming at deleting redundant information, correcting erroneous information, and ensuring data consistency; data integration refers to the organic concentration of data from different sources, formats, and natures; data approximation refers to the streamlining of data volume to the maximum extent possible; and data transformation refers to the transformation of different levels of data into a unified scope.

2.3

Theoretical Framework and Quantitative Modeling of Perceptual Imagery Discourse

2.3.1

Lexical similarity theory

The theoretical foundation of lexical similarity is built on the definition of similarity that D has a broad meaning [18]. The number of commonalities and differences between objects A and B will determine the degree of similarity between them. If the more commonalities the objects A and B carry, the greater their degree of similarity. Conversely, if the differences between the objects are greater, the less similar they are. If objects A and B are identical, the similarity reaches its maximum value at this point, which is the maximum limit of similarity. When objects A and B are independent of each other, their similarity is minimized, which is the lower limit of similarity between them.

According to DekangLin’s generalized theory, let two objects are A and B respectively, then the computational company formula of its similarity definition can be written in the following form: 1 $s i m (A, B) = f (I (c o m m o n (A, B)), I (d e s c r i p t i o n (A, B)))$

Where sim(A, B) - the similarity of objects A and B;

I(common(A, B)) - the amount of information on the same factors of objects A and B

I(description(A,B)) - the amount of information that completely characterizes objects A and B.

The geometric model was chosen for this study. In the geometric model, lexical objects are represented by points in the geometric model space, and the spatial distance between the points is used to measure the degree of similarity between them. The concept of semantic differentiation means that a feature space is constructed with a set of adjectives with opposite lexical meanings. When a word is described by these different adjectives, the coordinates of the word in the feature space are formed.

In this study the object is the natural language of product perceptual imagery, multiple features are used to describe an object, and an interval scale is used to quantify the features. The interval scale in this paper will use real numbers to represent quantitative information. Assuming that n feature is chosen, the m object can be represented as a matrix of m×n. Make this m×n as the following equation: 2 $A = [\begin{matrix} a_{11} & \dots & a_{1 n} \\ ⋮ & ⋱ & ⋮ \\ a_{m 1} & \dots & a_{m n} \end{matrix}]$ where A - the lexical geometric model.

a - points in the space of the geometric model, the

m - vocabulary designation.

n - number of features describing the vocabulary.

The eigenvalues of the geometric model used in this study are the dimensions of the multivariate scaling analysis, n value is the number of dimensions, m is the number of words after screening, and the objects in matrix A are the words describing the perceived imagery of the product. Since 85 words are used, m is 85, and after the above processing, the semantic lexical similarity model used in the study is constructed as a geometric spatial model of lexical similarity of m×n (the number of dimensions that will be used in the multivariate scaling process).

After completing the required Chinese natural evaluation language screening, the vocabulary base for quantitative model construction is obtained. At the same time, the semantic quantitative model construction is carried out after completing the construction of the theoretical framework required for quantitative computation with respect to the characteristics of the vocabulary problem of product-perceived imagery in the study.

2.3.2

Spatial Distance Description of Imagery Discourse

In the space of a reasonable geometric model of imagery semantic vocabulary, it is necessary to choose a reasonable distance calculation formula, i.e., vocabulary similarity expression, for a specific problem demand. Although there are many kinds of similarity calculation formulas for geometric modeling, the traditional similarity calculation is based on the concept of macroscopic lexical implication similarity, i.e., the similarity of the lexicon’s expressive effect in the face of all the objects it can describe. However, in this study, the vocabulary needs to express only one object, which is the user’s subjective imagery of the product appearance and styling. Therefore, the generalized similarity formula is not used in calculating lexical similarity.

Assuming that there is n object, the degree of similarity is determined by comparing two by two, resulting in a preliminary expression matrix for the degree of similarity of the thing. Assuming that S_ij is the degree of proximity between the i rd object and the j th object, and using Δ to represent the matrix of proximity between the n objects, the matrix can be expressed as follows: 3 $Δ = [\begin{matrix} s_{11} & \dots & s_{1 n} \\ ⋮ & ⋱ & ⋮ \\ s_{n 1} & \dots & s_{n n} \end{matrix}]$ where s - point in space;

n - vocabulary code.

The similarity calculation method uses pearson correlation coefficient, let X and Y be the eigenvalue vectors corresponding to two objects i_x, i_y in matrix A, then the expression formula of similarity is as equation (4): 4 $s i m (i_{x}, i_{y}) = \frac{cov (X, Y)}{σ_{x} σ_{y}} = \frac{E [(X - μ_{x}) (Y - μ_{y})]}{σ_{x} σ_{y}}$

Where sim(i_x, i_y ) - similarity between lexical objects i_x, i_y. The magnitude of similarity between the objects is defined based on the cosine data of the angle between the two spatial vectors of object i_x, i_y in the geometric model. At the same time, after the calculation of the similarity of all the objects in each correspondence, all the data are normalized in order to preserve the properties of the data for the expression of similarity, as well as to facilitate subsequent research and calculation.

2.3.3

Spatial Distance Analysis of Imagery Discourse

The multidimensional scaling method transforms the “distance” between objects into a multidimensional spatial form, so it is widely used in similarity problems. The multidimensional scaling of matrix Δ is used to further analyze the potential connection of the imagery words, and attempt to determine the coordinates of each word in the feature space of lower dimensions. Specifically, the multidimensional scaling method is used to construct the distance matrix D_ij = (d_ij) between the objects through the similarity matrix C_ij = (c_ij), and the data transformation details are shown in Equation (5): 5 $d_{i j} = \sqrt{(c_{i i} + c_{j j} - 2 c_{i j})}$

In order to perform a k-dimensional fit, this will be done by constructing n*k a matrix of points X and trying to make X the corresponding distance matrix D_ij = (d_ij) have information that faithfully reflects that in D_ij = (d_ij) the matrix. The D_ij = (d_ij) non-diagonal elements are first selected and arranged in order from smallest to largest, and then ordered: 6 $S^{2} (X) = \frac{\min \sum_{i p j} (d_{i j}^{*} - d_{i j})}{\sum_{i p j} d_{i j}^{2}}$

During the operation, $d_{i j}^{*}$ will be adjusted so that S²(X) reaches the minima. At this point, the $d_{i j}^{*}$ of the case where $S^{2} (\hat{X})$ reaches a minima is referred to as the least squares regression of $d_{i j}^{*}$ . It is now assumed that there exists a X₀ such that with k fixed: 7 $S^{2} (X) = \min_{X_{n * k}} S (X) = S_{k}$

At this point, X₀ is the best fit for k, and S_k is called the pressure index.

After the above process, the multidimensional scaling method quantifies the degree of similarity between the objects in the form of “distance”, and each object is reflected in its resulting coordinate system in the form of coordinate positioning. The “distance” between each coordinate position reflects the degree of similarity between the original objects. At the same time, the multidimensional scaling method utilizes multiple iterations of approximation to continually adjust the degree of fit between the distances between objects in the coordinate system and the similarity of the actual objects being examined until the accuracy reaches a satisfactory level.

2.3.4

Representative Imagery Discourse Extraction

Each object is represented as a coordinate locus in the coordinate system in which it is generated, and in order to effectively differentiate the data, cluster analysis is used to analyze the similarity results above. Although there are many algorithms available for cluster analysis such as k-mean clustering, self-organizing maps, fuzzy clustering, etc., none of them provide a direct way to determine the number of clusters. In order to determine the appropriate number of clusters, hierarchical clustering and the sum-of-squares-of-discrepancies ward method were used). Firstly, homogeneous clusters were constructed using the k-means clustering method, and the perceptual words closest to the center of each cluster were selected as representative words. Secondly, the distance of each set of imagery vocabulary pairs to the geometric center of its cluster group is calculated and ranked separately, and the imagery vocabulary pairs closest to the center position in each cluster are selected as the representative affective imagery dimensions, and the number of representative imagery vocabulary pairs is equal to the number of clusters.

The process of semantic quantization and the theory of model construction are described above, and the specific construction of the quantization model is completed by the MATLAB software m language.

2.4

QFD Core - Quality House Modeling

House of Quality (HoQ) is the core of QFD, which integrates customer needs with product technical characteristics, and calculates the degree of correlation between customer needs and technical characteristics by building a matrix to determine customer satisfaction and improve product technical characteristics. The quality house structure model is shown in Figure 3.

According to the figure, the quality house is divided into left wall, roof, ceiling, room and floor and the details of the five sections are:

The left wall part contains customer demand related information, including categories and weights of customer demand. The corresponding customer needs and the importance of each need can be filtered out by conducting a survey and research on the customers.

The ceiling section is the technical characteristic parameters of the product, which are related to the customer’s needs, and their relevance can be described both verbally and quantitatively.

The roof section is constructed on top of the ceiling and represents the correlation between the technical characteristics of the product. Due to the differences in technology and design methods during the product development process, the technical characteristics of the product may contain each other, which needs to be guaranteed in order to ensure the scientific rationality of the calculations.

The room section represents the correlation matrix between customer needs and the technical characteristics of the product, through the establishment of the correlation matrix can be obtained for each technical characteristic of the final importance of the size of the important information for the expansion of the OFD after the study.

The floor part is the importance degree of product technical characteristics calculated by integrating the above quality house information, and the weights of technical characteristics are assigned from the perspective of maximizing customer satisfaction, which can provide reference for further product improvement work.

2.5

Prioritization analysis based on fuzzy weighted association rule mining

2.5.1

Final Prioritization Analysis of TCs in a Fuzzy Environment

The study of prioritization of fuzzy sets in fuzzy environments includes two methods: order relation and order function. Sequence function is a method of mapping fuzzy sets onto the numerical axes by mapping function to perform ordering; while sequence relation is a method of determining the order of fuzzy sets by constructing the association relationship between fuzzy sets and determining the order of fuzzy sets according to the magnitude of their affiliation degree [19]. For the final prioritization analysis of TCs in this paper, the ordinal relationship calculation method needs to take two and two fuzzy numbers as a comparison, which is more complicated to calculate and will bring extra judgment error; the pre-function method takes the Hemming distance as a distance indicator, and the minimum set is selected as a reference first, and only the distance between each fuzzy set and the minimum fuzzy set can be calculated to get the sorting result.

2.5.2

Improved fuzzy weighted association rule mining algorithm

1)

Association rule algorithm Apriori algorithm

Apriori algorithm is a representative algorithm for association rule mining, and its core idea is to discover the intrinsic connection between things. The association rule is a relational expression like X→YX→Y, and X and Y are two sets of terms that are independent of each other [20]. The degree of association between itemsets can be expressed in terms of support and confidence. Support can represent the probability that the itemset (X, Y) appears in the database, i.e., the size of the probability that the two events occur at the same time, and confidence represents the probability that the itemset Y also appears in the itemset X, which is often expressed as (Y|X).

By setting the minimum support threshold, the database is scanned to mine all the sets of items that satisfy the minimum support threshold, which is called frequent itemset. For example, suppose a set {A, B} is a frequent itemset, which means that the number of times A and B appear in a record at the same time is greater than or equal to the minimum support min_support, then the number of times its subsets {A} and {B} appear must be more than min_support, and thus its subsets are all frequent itemsets.

2)

Final Prioritization Analysis of TCs

Let the set of all transactions in the constructed comment database be denoted as T = {t₁,t₂,t₃…t_n}, and the set of items I = {i₁,i₂,i₃…i_n}, W = {w₁,w₂,w₃…w_n} denote the weights. The following definitions are proposed: Definition 1:

Item weight (IW) denotes a non-negative real number, and for each item the weight is denoted as i_jw.

Definition 2:

Transaction weight (ITW) denotes the direct product of the weights of itemset i in a transaction: 8 $I T W (X) = \prod_{k = 1}^{| x |} (\forall i_{k} \in x) t_{i} [i_{k} [w]]$

Definition 3:

Weighted support (WS) denotes the ratio of the sum of the item set transaction weights of an item set to the total number of transactions: 9 $W S (X) = \frac{\sum_{i = 1}^{n} \prod_{k = 1}^{| X |} (\forall i_{k} \in x)}{n}$

In the fuzzy setting, it is assumed that the fuzzy dataset is also a transaction set, while the linguistic value set LI = {I₁, I₂, I₃,…, I_n}, where l₁is a fuzzy set representation and L = {low, medium, high} represents the degree of affiliation of the items with the fuzzy set. The following definition is then proposed:

Definition 4:

Fuzzy Transaction Weight (FITW) is the direct product of all fuzzy item values of a fuzzy item set and their weights: 10 $F I T W (X, A) = \prod_{k = 1}^{| l |} (\forall i_{k} \oplus x) t_{i} [I_{k} [w]]$

Definition 5:

Fuzzy Weighted Support (FWS) is the ratio of the sum of the weights of all fuzzy transactions to the total number of transactions, denoted as: 11 $F W S (X) = \frac{\sum_{i = 1}^{n} \prod_{k = 1}^{| k |} (\forall i_{k} \in X)}{n}$

The kano classification of each TC is determined based on the result L of the obtained set of strongly correlated items, i.e., the Kano classification of the customer demand CR with the highest correlation corresponding to each TC, and the full range of TCs is classified into basic, onedimensional and charismatic attributes. The corresponding Kano coefficient is Q₁ = 0.5, Q₂ = 1, Q₃ = 2. By Eq: 12 $W_{F_{i}} = {[μ (_{K_{i}}^{)} + (1 - μ) (_{K_{i}}^{)}]}^{c Q}$

Calculating the final importance of the TCs and sorting the TCs by size finally yields a prioritized set of product technical characteristics: TCs = {TC₁, TC_i, TC_j,….TC_n | TC_i ≥ TC_j, i > j}.

Here, we construct the item set in the transaction based on the correlation relationship between the technical characteristics TCs and each customer requirement CRs, and its correlation coefficient is used as the weight of each technical characteristic w_ij, i = 1,2,…,I, j = 1,2,…,J, which indicates the weight of the j rd technical characteristic in the i th customer requirement.

3

Multidimensional perceptual information accessible product design applications

3.1

Selection of word pairs for product perceptual imagery

3.1.1

Collection and Screening of Initial Imagery Word Pairs

In order to be able to better express the user’s perceptual needs, unlike the original collection of a single imagery vocabulary, this paper focuses on the collection of a number of pairs of imagery words, which not only reduces the later again for the preparation of antonyms, the proposed words are also more accessible to the user’s understanding, because when the user is able to use a set of imagery words to evaluate the existing product, it means that the understanding of the product among users on this imagery is difference and that the users are able to understand the imagery words well.

Based on this, in the initial stage of this paper, when collecting perceptual imagery, we focus on selecting pairs of words with opposite meanings that appear more frequently from magazines, newspapers, Taobao, Jingdong and other e-commerce platforms, and refer to the research results of related fields to collect relevant imagery words, and obtain 66 pairs of imagery words, and then delete the industry terminology that can only describe the function of the product and the professional terminology of designers to be excluded, and finally retain 56 pairs of imagery words. Finally, 56 pairs of imagery words have been retained. The results are shown in Table 1.

Table 1.

The image of the selected description of the digital camera is right

Number	Perceptual evaluation	Number	Perceptual evaluation	Number	Perceptual evaluation
V1	Exquisite; -- rough;	V23	Portable; - inconvenient;	V45	Monotonous;-- Change;
V2	Fluent; -- blunt;	V24	Abundant; Poor;	V46	Slender; -- rough;
V3	Coordinated; Sudden;	V25	Innovative; Imitation;	V47	Rigorous; -- casual;
V4	Light; -- heavy;	V26	Affinity; -- cold;	V48	Avant-garde; --conservative;
V5	Positive; Negative;	V27	Mysterious; -- honest;	V49	High quality; --overdone;
V6	Childish; -- mature;	V28	Wild; - civilized;	V50	Male; -- women;
V7	Technology; Cold;	V29	Frontier; -- outdated;	V51	Stiff; - soft;
V8	Public; -- low-key;	V30	Balanced; A dissonance;	V52	Simple; - complex;
V9	Lively; -- mechanical;	V31	fancy Pure;	V53	Bright; -- dim;
V10	Negligent; - considerate;	V32	Practical; -- decorative;	V54	Exaggerated; - reserved;
V11	Interesting; -- boring;	V33	masculine -- feminine;	V55	Small; -- atmospheric;
V12	Generous; -- reserved;	V34	Smooth; Change;	V56	Plain; - gorgeous;
V13	Tough; Soft;	V35	Convenient; - inconvenient;	V57	Clumsy; - clever;
V14	Personality; -- public;	V36	Cold; - gracious;	V58	Fashionable; -- plain;
V15	Vulgar; - elegant;	V37	Imposing; - cold acid;	V59	Economical; -- luxury;
V16	Natural; Artificial;	V38	Durable; -- vulnerable;	V60	Tight; -- loose;
V17	Beautiful; -- ugly;	V39	The rules; -- rebellious;	V61	In the office; Low gear;
V18	Thick; -- guaranteed;	V40	Easy to understand; -- difficult to understand;	V62	Angular; - round;
V19	Pleasing; -- disturbed;	V41	Accurate; - distorted;	V63	Cheap; Expensive;
V20	Handsome; - soil gas;	V42	Comfortable; -- miserable;	V64	Professional; - amateur;
V21	Safe; -- dangerous;	V43	Lasting; Short;	V65	Ordinary; -- novel;
V22	Easy to use; --Hard to use;	V44	Simple; -- complicated;	V66	Harmonious; -- the chaos;

3.1.2

Screening of representative imagery word pairs

Due to the similarity of meaning between the obtained pairs of perceptual imagery words, it is necessary to select representative imagery words that can represent the perceptual needs of the category among the words with similar meanings. For this purpose, a questionnaire was designed to collect similarity ratings of image words as a way to achieve word clustering. The specific operation is as follows: the obtained 18 words are compared two by two, the respondents evaluate their similarity according to whether they are similar or not, the score of 0 means very dissimilar, the score of 1 means relatively dissimilar, the score of 2 means not clear, the score of 3 means relatively similar, the score of 4 means very similar, and the score of the word-to-self comparison is 5. The specific details of the questionnaire are shown in Appendix A, and the final similarity matrix was used for cluster analysis. There are 19 subjects in the experiment, all of them belong to the research scholars of perceptual engineering, and they have a deeper understanding of perceptual imagery, the final 19 similarity matrices were averaged and normalized, and the distance matrix was obtained by subtracting the similarity matrix from the unit matrix of the same dimension, which was used as the input for the multidimensional scale analysis, and the results are shown in Table 2.

Table 2.

The selected dimensions and corresponding stress values and RSQ values

Dimension	Stress	RSQ
1	0.4519	0.3827
2	0.2286	0.7015
3	0.1624	0.8081
4	0.1129	0.8698
5	0.0843	0.9102
6	0.0684	0.9277

Stress and RSQ were used to indicate the fit between the spatial structure and the actual data and the degree of explanation of the structural space to the variability of the input data, respectively, where the smaller the value of Stress, the better, generally less than 0.1 is considered to be better, while greater than 0.2 is unacceptable. And the larger the RSQ the better, generally considered greater than 0.6 is acceptable. Therefore, the final choice of six dimensions with good performance in all indicators turned out to be the most reasonable, in which the coordinates of the dimensions of each perceptual imagery word pair are shown in Table 3.

Table 3.

The six dimensional coordinates of each image

N	Sensual image	Dimension 1	Dimension 2	Dimension 3	Dimension 4	Dimension 5	Dimension 6
1	Atmospheric - compact	1.048	0.57	-0.508	0.333	1.098	-0.094
2	Portable - inconvenient	1.212	-1.974	0.756	-1.402	2.619	0.838
3	Durable -- vulnerable	0.787	1.153	2.57	-2.75	-0.334	0.172
4	Beautiful -- ugly	-0.527	-1.308	0.49	2.294	-0.656	1.486
5	Personal -- common	-1.947	-1.301	0.78	-1.914	-0.317	0.224
6	Stylish -- pristine	0.05	-0.079	-1.311	-0.783	-2.126	-0.489
7	Simple -- complex	0.837	1.662	-0.925	-0.646	-0.155	-2.221
8	Tech - cold	0.147	0.306	-1.644	0.689	-1.212	-0.78
9	Cheap -- expensive	-1.695	-0.884	-0.958	1.332	0.855	-1.627
10	Delicate -- rough	0.96	-2.108	-0.336	0.149	0.645	-0.035
11	Upscale -- low	0.429	2.149	-2.33	-1.463	3.356	-1.613
12	Professional - amateur	-1.529	-2.27	0.472	2.709	-1.557	1.497
13	Easy to use -- difficult to use	-1.764	-0.718	0.278	-1.075	1.397	2.159
14	Angular - round	-2.443	-2.237	-0.105	-2.899	1.264	3.244
15	Masculine, feminine	3.44	0.594	-2.564	-0.296	-0.824	0.782
16	Innovative -- imitation	0.092	-3.029	1.079	-0.317	-0.435	-0.449
17	Practical -- decorated	-0.122	0.81	-0.39	-3.13	-2.092	1.594
18	Bright - dim	1.346	1.891	-0.006	0.173	1.624	-2.44

Cluster analysis was performed using the coordinate values of each imagery to obtain a dendrogram of Ward’s connections, as shown in Figure 4.

Based on the grouping lines five major categories were finalized, so the perceptual imagery can be classified into five groups, and the final division results are shown in Table 4.

Table 4.

The clustering results of 18 emotional images

Group	The perceptual image contained within the group
1	beautiful	fashion	exquisite
2	portable	succinct	easy-to-use
3	durable	cheap
4	individualistic	technological	upmarket	professional	innovative
5	atmospheric	angular	masculine

Based on the clustering results obtained, one representative perceptual image can be selected from each category in turn for further perceptual prediction modeling.

The words in the first category are mainly used to evaluate the aesthetic characteristics of digital cameras, and “fashionable” was chosen to express the users’ needs because it can be used as a comprehensive evaluation of the aesthetics of users in a specific period of time.

The second category of words generally expresses the usability needs of the camera, and “easy to use” was chosen to express the perceptual needs of the users.▯▯

The third category of words reflects the practicality of the camera, for which “durability” is chosen as the representative perceptual image, and “cheap” is inevitably unsuitable as an optimization goal.

In the fourth category, the word “innovative” reflects more on the technological content of the product, while “innovative” can comprehensively reflect the technological content of the product, so “innovative” is finally chosen as the user’s perceptual demand for digital cameras.

In the fifth category, “masculine” was chosen as the representative imagery word, because “masculine” has more significant differences in the needs of different users, so it was chosen as the representative perceptual imagery. The final perceptual images obtained are: fashionable - rustic, masculine - feminine, durable - perishable, easy to use - difficult to use, innovative - imitative.

3.2

Topic word associations for user requirements

The use of Python language to realize the correlation process of user requirements, the output results, for example, the air conditioning “mute” theme, a total of 7 strong correlation, adjust the order of the strong correlation vocabulary can be relatively easy to understand the statement, such as [wind, almost, can not hear] - [sound], adjust the order of Adjusting the order of the strongly associated words will result in a relatively easy to understand statement such as [wind, almost, can’t hear] - [sound]. By similar methods, a total of 50 strong association rules are obtained. A low noise:

A1 [inaudible] - [sound], A2 [a little] - [sound], A3 [almost, inaudible] - [sound], A4 [a little, wind] - sound], A5 [a little, inaudible] - [sound], A6 [heating, inaudible] - [sound], A7 [wind almost inaudible] - [sound].

B Energy Saving:

B1 [Product, Energy Saving] I [Energy Saving], B2 [Grade 1, Energy Efficiency] - [Energy Saving], B3 [Product, Effectiveness, Energy Saving] - [Energy Saving], B4 [- Grade] - [Energy Efficiency], B5 [- Grade. power saving] - [energy efficiency], B6 [-grade, inverter] - [energy efficiency], B7 [-grade air conditioner] - [energy efficiency].

C Fast cooling heat:

C1 [Fast] - [Cooling], C2 [Fast] - [Cooling], C3 [High Efficiency] - [Cooling], C4 [Fast] - [Cooling], C5 [Fast, Heating] - [Cooling], C6 [fast, air conditioning] - [cooling].

D Mobile phone interconnection:

D1 [control] - [cell phone], D2 [smart, control] - [cell phone], D3 [function, control] - [cell phone], D4 [control, wifi] - [cell phone], D5 [function, smart, control] - [Cell Phone], D6 [Function, Intelligent, Control, wifi] - [Cell Phone].

E sensing wind power:

E1 [soft] - [wind], E2 [heating, wind] - [soft], E3 [strong] - [wind], E4 [fast, effect, strong] - [wind], E5 [soft] - [comfortable Comfort].

F safety protection:

F1 [leakage] - [protection], F2 [switch] - [protection], F3 [switch, leakage] - [protection], F4 [switch air conditioning] - [protection].

The 50 strong association rules are summarized into 9 user product requirements as shown in Table 5.

Table 5.

User requirements without barriers to the rules of association rule

User demand	Strong correlation rules	Support	Confidence	Ascension
A	A1	0.01335	0.68129	72.01292
	A2	0.00547	0.75222	53.60586
	A3	0.01044	0.68111	23.77713
	A4	0.00757	0.98421	20.06997
	A5	0.01104	0.60984	10.25404
	A6	0.00822	0.6191	58.89725
	A7	0.0124	0.80512	85.38274
B	B1	0.01206	0.92724	80.26716
	B2	0.01485	0.68187	17.1543
	B3	0.008	0.93135	31.00875
	B4	0.01248	0.91189	85.78589
	B5	0.00674	0.96733	65.98953
	B6	0.01443	0.79679	36.56515
	B7	0.00843	0.7566	12.25966
C	C1	0.00734	0.75605	20.03919
	C2	0.00984	0.65121	61.3382
	C3	0.00575	0.65783	35.63621
	C4	0.0138	0.89794	11.39264
	C5	0.01305	0.61532	55.00875
	C6	0.00788	0.72726	73.74591
D	D1	0.00868	0.86541	19.38871
	D2	0.0103	0.77379	2.27551
	D3	0.00749	0.74605	16.68737
	D4	0.0092	0.82716	49.27182
	D5	0.0076	0.71909	33.31272
	D6	0.01044	0.80562	8.9039
E	E1	0.0077	0.82442	77.42159
	E2	0.00549	0.90763	86.50152
	E3	0.01321	0.77485	5.50477
	E4	0.01126	0.67392	42.52892
	E5	0.01185	0.73315	31.39435
F	F1	0.00983	0.69089	60.9686
	F2	0.00789	0.70062	59.33258
	F3	0.01346	0.70204	16.03612
	F4	0.00911	0.61959	80.82295

Apriori association algorithm will produce invalid strong association rules, after analysis, it may be caused by the following three reasons: first, due to the fact that some of the user needs, such as dehumidification and air purification and other demand information, appear less frequently in the original corpus, resulting in the problem that the weight of the subject words is too small, and they will not appear and be excluded from the first 10 subject words; second, due to the fact that some Chinese words can contain multiple semantics, it is not very good in the process of word separation, which leads to the deviation of Chinese text word separation; third, due to the large number of Chinese near-synonyms, it is more difficult to obtain the reflection of words with similar meanings in the mathematical sense. Through massive data collection, analyzed by the combination of model and Apriori correlation algorithm, mining obtained the user needs of accessible products, by the theory of demand can be divided into the basic functional needs as well as high-level aesthetic needs, specific functional needs are: low noise, energy saving, fast cooling and heating, cell phone APP control, strong mode of the wind feeling soft, safety protection; aesthetic needs are: beautiful and generous, High-end intelligence and fashionable novelty.

3.3

Analysis of results

This study can provide effective and reasonable guidance for accessible product development and design decisions, and ensure reliability in the analysis of user requirements. According to the technological competitive benchmarks of R&D enterprises, the engineers formulated the product development strategy, and the final accessibility product development objectives are shown in Table 6. It can be seen that the importance degree of adding voice control to V8 is 0.919, which ranks first in terms of importance of all target items.

Table 6.

Unimpeded product development goals (homemade)

Engineering characteristics	Importance (intuition fuzzy number)	Importance (real number)	Target
V1	0.0111,0.854	0.0867	Flow resistance
V2	0.0131,0.8543	0.0801	Condenser heat
V3	0.0111,0.815	0.0889	Compressor refrigerating quantity
V4	0.0116,0.815	0.0885	Air flow
V5	0.011,0.815	0.0857	Wind resistance
V6	0.013,0.815	0.0883	Develop the air quantity distribution table
V7	0.0158,0.815	0.0884	Applied led screen
V8	0.019,0.876	0.0919	Increased voice control
V9	0.0126,0.815	0.0842	The fuselage is cylindrical
V10	0.0104,0.815	0.0845	The top of the machine is round
V11	0.0136,0.826	0.0869	The machine is low in circle
V12	0.0126,0.826	0.0876	The outlet USES a rounded rectangle

3.4

Usability testing

Usability is the degree to which a product can be used effectively, efficiently, and satisfactorily by a specific user in a specific scenario. The framework of usability is defined as five points: learnability, efficiency, memorability, errors and satisfaction. The usability of a product has a direct impact on the user’s satisfaction and the quality of the product design. In order to identify problems in the design of accessible products and to verify whether the design meets the user’s needs for use and the expectations of elderly passengers, the author conducts usability testing of the operator interface.

After obtaining the raw data of the test subject’s evaluation, the original score needs to be converted, in which the score value of odd-numbered items is calculated as “raw score-1”, and the score value of even-numbered items is calculated as “5-raw score”, and the total score range is 0-40 points, while the range of SUS is 0-100 points, and the score of SUS needs to be obtained by multiplying the score value by 2.5. Once the score has been calculated, you can view the corresponding score based on the curve grading range of the SUS score. 85 to 89 is A-, 90 to 95 is A, and 96 to 100 is A+.

By distributing the SUS scale to the eight test persons, after transforming the raw values obtained, the final rating results were obtained as shown in Table 7.

Table 7.

Usability test scores

	Q1	Q2	Q3	Q4	Q5	Q6	Q7	Q8	Q9	Q10	Score	Grade
Subjects1	4	3	4	4	3	3	3	4	3	4	84.2	A+
Subjects 2	3	4	4	4	3	3	4	4	3	3	84.2	A+
Subjects 3	4	3	4	4	3	3	3	3	4	3	83.5	A
Subjects 4	3	3	4	4	3	4	3	3	3	4	83.5	A
Subjects 5	4	4	3	4	3	4	4	3	3	3	84.2	A+
Subjects 6	3	4	3	4	4	3	3	4	3	3	83.5	A
Subjects 7	4	3	4	3	4	3	4	3	4	4	85.1	A+
Subjects 8	3	3	4	4	4	4	3	4	3	3	84.2	A+
Mean	3.5	3.375	3.75	3.875	3.375	3.375	3.375	3.5	3.25	3.375	84.05	A

The analysis of the overall assessment results reveals that the ratings of the tested persons are all at 83, i.e., grade A or above, indicating that the overall assessment results of the accessible product design analyzed in this paper are good.

4

Conclusion

From the perspective of perceptual engineering, this paper constructs a mapping model of perceptual information and product solutions based on association rules through the description and analysis of spatial distance of imagery discourse to assist in accomplishing the application of multidimensional perceptual information for accessible product design. After collecting and screening initial imagery word pairs and analyzing topic words with user requirements, the final accessibility product development goals and design solutions were derived. The SUS scale was distributed to the eight test subjects, and after transforming the raw values obtained, the final ratings of the innovative design of the accessible products in this paper were obtained. The analysis of the overall assessment results shows that the ratings of the tested persons are all 83 points, i.e., grade A or above, indicating that the overall assessment results of the accessible product design analyzed in this paper are good.

Langue:: Anglais

Périodicité:: 1 fois par an
Sujets de la revue:: Sciences de la vie, Sciences de la vie, autres, Mathématiques, Mathématiques appliquées, Mathématiques générales, Physique, Physique, autres

RSS Feed de la revue

Innovative assisted design of accessible products based on multi-dimensional perceptual information association rule extraction

Qi Han

Shu’nan Liu

Publié en ligne: 19 mars 2025

Reçu: 03 nov. 2024

Accepté: 18 févr. 2025

DOI: https://doi.org/10.2478/amns-2025-0453

Mots clésMultidimensional information processing, Lexical similarity, Fuzzy weighting, Association rules, Accessible products

© 2025 Qi Han, published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Mots clés
Multidimensional information processing, Lexical similarity, Fuzzy weighting, Association rules, Accessible products