Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1 Introduction

People enjoy shopping on the Internet with their computers or smart phones. Some shopping sites provide functions to customize products, such as shoes, wrist watches, shirts, and sports uniforms. Users create their favorite design by selecting a color, shape, and design elements. However, it can be difficult for users to select an appropriate design for each part of the product. Moreover, in the process of creating the products, users may not notice some good design elements.

Therefore, we use an Interactive Evolutionary Computation (IEC) method to customize products and create a product that satisfies the user. The IEC method creates products based on user subjective evaluations and Evolutionary Computation (EC) technique [1]. Previously proposed systems that apply IEC include various image processing filter designs [2], music generation [3,4,5], and image retrieval systems [6]. However, IEC has a problem that user evaluation loads of candidate solutions are large.

To address this problem, IEC researchers have discovered that simple evaluation processes are more effective [1]. Previously, we proposed a tournament-style evaluation to evaluate candidate solutions by paired comparison [7]. We confirmed that this tournament-style evaluation is more effective for reducing user evaluation loads than a 10-stage evaluation method for music, animations, and images. However, with tournament-style evaluations, it is difficult for users to find relationships among candidate solutions between the current and previous paired comparison evaluations. Therefore, a user may not be motivated to use the system because paired comparisons where relationships are unknown must be repeated. To address this problem, it is effective to ensure that the previous competition’s winning candidate solution matches new solution.

Then, we propose a Winner-based Paired Comparison (WPC) method whereby a winning candidate solution reaches the next competition. The WPC method performs paired comparisons between the winner candidate solution of the previous competition and new solution. Therefore, a user can find the relationship between competitions and will be motivated to continue the evaluations. Additionally, the number of evaluations per one generation is \((n-1)\) when the number of candidate solutions is n. We create a wrist watch design system with WPC method and verify the effectiveness of the proposed system with real users. Many people are interested in wrist watch design, and design is an important factor in the fashion of each user. The wrist watch design of our system consists of four parts: hand, dial, dial edge, and strap. We employ a Genetic Algorithm (GA) as the evolutionary algorithm in the IEC method. We examine the effectiveness of the proposed system for real users relative to satisfaction level for the generated design, usability, and evaluation time.

Fig. 1.
figure 1

Outline of the proposed system

Full size image

2 Proposed System

2.1 Winner Based Paired Comparison Method

Figure 1 shows the schematic of the proposed system. First, the system randomly generates initial candidate solutions and presents them to the user. The user evaluates the presented candidate solutions with the WPC method. When the user has evaluated all candidate solutions, the system generates new solutions with a GA operation and presents new candidates to the user. The system repeats these operations and creates objects that user can satisfy.

Figure 2 shows candidate solution evaluation method of WPC. First, WPC assigns an evaluation value of 1 to all generated candidate solutions. Then, the system generates a pair of candidate solutions randomly and presents it to the user for evaluation. When the user selects their favorite candidate, the WPC method adds the evaluation value of the losing candidate solution to the evaluation value of the winning candidate solution. In Fig. 2, as A wins the first round, WPC adds the evaluation value of B to the evaluation value of A. Then, the evaluation value of A is 2. A also wins in the second round; thus, the WPC adds the evaluation value of C to evaluation value of A. Then, evaluation value of A is 3. Since D wins in the third round, the WPC adds the evaluation value of A to the evaluation value of D. Thus, the evaluation value of D is 4. Finally, the evaluation values of each candidate solution are \(A=3, B=1, C=1\), and \(D=4\). The proposed system uses these evaluation values for the GA operations.

Fig. 2.
figure 2

Candidate solution evaluation method of WPC

Full size image
Fig. 3.
figure 3

Evaluation interfaces of the proposed system

Full size image
Fig. 4.
figure 4

Gene coding of wrist watch design

Full size image

2.2 Evaluation Interface

Figure 3 shows evaluation interfaces of the proposed system. We create the proposed system as an Android application. When the user starts the application, the application shows the start interface (Fig. 3(a)). If the user taps the start interface, the application shows the evaluation interface (Fig. 3(b)). Here, the user evaluates two designs. The winning candidate solution is passed to the next competition. The user can confirm the number of the current evaluation by the number in the lower-right of the interface. When the user completes the evaluations, the application shows the created best design (Fig. 3(c)). If the user taps the “Good bye!!” button, the application is closed.

2.3 Wrist Watch Design

Figure 4 shows the gene coding of the wrist watch design. The wrist watch design consists of four parts: hand, dial, dial edge, and strap. Each part has eight or sixteen designs, which are expressed by 3 or 4 bits. Then, the system can create 16,384 designs because the gene length is 14 bits. Figure 5 shows the design elements of the wrist watch design. We determined the bit patterns of each design part by considering the similarity between the appearance of each design part and bit pattern.

Fig. 5.
figure 5

Design parts of wrist watch design

Full size image

3 Evaluation Experiments

3.1 Outline of the Experiment

Here, we investigate the effectiveness of the proposed system for real users using the wrist watch design system. Twenty subjects (twelve men and eight women) in their twenties participate in the evaluation experiment. The men subjects used the proposed system based on the concept that customers will select the wrist watch design that they want to wear to enjoy fashion. The women subjects used the proposed system based on the concept that they want men of a friend to wear.

Table 1 shows the experimental parameters. We set the number of generations and candidate solutions to 10 and 8; thus the subjects evaluated seven paired comparisons per generation and 70 paired comparisons for all generations. The mutation operation flips each gene locus without the current elite candidate solution according to a predetermined mutation rate. We set the mutation rate to 20%. After the experiment, the subjects evaluated satisfaction level for the generated design and the usability of the proposed system in a five-stage evaluation.

Table 1. Experimental parameters
Full size table
Fig. 6.
figure 6

Satisfaction level for the generated design

Full size image
Fig. 7.
figure 7

Usability for the proposed system

Full size image
Fig. 8.
figure 8

Evaluation time for the proposed system

Full size image

3.2 Experimental Results

Figure 6 shows the satisfaction level for the generated designs. More than half of all subjects assigned a value of greater than 4 to the generated design. The average satisfaction level was 4.15. Therefore, we confirmed that the proposed system can create designs that satisfied the subjects.

Figure 7 shows the usability of the proposed system. More than half of all subjects assigned a value of greater than 4 for the usability. The average usability was 3.75. Therefore, we confirmed that the proposed system can be used easily and can create designs that satisfied the subjects. However, 35% of all subjects assigned a value of less than 3 for usability. They commented that they wanted response visual effects when using the application because they may not understand which design has been selected.

Figure 8 shows the evaluation time for the proposed system. The longest (shortest) evaluation time was three minutes and thirty-one seconds (one minute and four seconds). The average evaluation time was two minutes and thirteen seconds. More than 85% of all subjects finished evaluating the designs because the proposed system employs paired comparisons as a simple evaluation interface.

From the experimental results, we confirmed that the proposed system can create a wrist watch design that can satisfy the user without increasing user evaluation loads. Therefore, the proposed method is effective at creating wrist watch designs in a simple manner.

4 Conclusions

We have proposed a wrist watch design system with an IEC method that employs the WPC method. We performed an evaluation experiment with real users to verify the effectiveness of the proposed system. The experimental results showed that the proposed system can create wrist watch designs that satisfy user preference in a simple manner. In future, we will improve our system in consideration of the subjects’ comments and compare the effectiveness of the proposed system to other IEC algorithms.