基於使用者操作行為模擬及元件配置最佳化的介面設計：以二維圖形化使用者介面為例

Interface Design through the Simulation of User’s Operations and the Optimization of Component Layout: The 2D Graphical User Interface as an Example

周庭瑱(Ting-Chen Chou)

指導教授 ： 盧俊銘

國立清華大學/工學院/工業工程與工程管理學系/碩士(2021年)

摘要

介面的易用性是近年來受重視的議題，以往多採用專家或使用者的評估，這些方法雖被廣泛應用，但人為介入都是必要的，因此可能面臨使用者反應未盡自然、難以反映全體使用者行為、專業人員培訓困難等挑戰，且涉及一定程度的金錢與時間成本；此外，由於評估通常並非系統開發者所擅長，其與設計變更也因此常演變為分別進行的獨立作業，難免造成理解差異而無法有效解決評估時發現的問題或是影響了時效。因此，本研究旨在提出一套模擬使用者行為並運用演算法將介面佈局最佳化的方法，藉以減少人為介入帶來的限制、有效提昇介面的易用性。
本研究以熟手使用者與二維圖形化介面互動的績效作為探討標的，基於模擬使用者的操作行為，採用多目標遺傳演算法進行介面佈局最佳化，最佳化目標式包含縮短操作時間、限制按鍵幾何形狀與提昇按鍵一致性。操作時間的估計是以按鍵層級模型拆解使用者在介面上操作的行為，再透過霍夫曼模型推估按鍵間的移動時間，同時考量不同操作策略之間的差異；按鍵幾何形狀之目標則是在於避免按鍵過於狹長而導致搜索或點擊不易，因此針對長寬比過大的元件賦予懲罰值以限制此情形；按鍵一致性的目標則是為維持介面的整齊性並增加搜索效率，因此對功能相同卻大小不同的按鍵給予懲罰值加以限制。
基於前述概念，本系統能根據介面組成元件特性模擬使用者的操作行為、直接產生最佳化的介面佈局，為建立使用者模擬的模型並驗證最佳化的介面是否能如預期地提昇操作績效，本研究設計兩階段的實驗，各招募30名裸視或配戴隱形眼鏡矯正後視力達0.8以上且互不重複的參與者，第一階段實驗包含男性13名、女性17名，平均年齡為22.57 ± 1.17歲，第二階段實驗包含男性14名、女性16名，平均年齡為22.63 ± 1.59歲。第一階段實驗是為了建立模擬使用者行為的基礎，參與者需以滑鼠操作電腦，與根據一般性設計原則產生的二維圖形化介面互動，藉以蒐集操作二維平面介面時之各項行為（如點擊、確認等）所花費的時間，並從移動行為推估霍夫曼模型的斜率與截距；此外，即便於相同介面上執行相同任務，不同操作策略也會影響介面操作績效，為使策略區分能屏除人為的介入、提昇客觀性，本研究提出四種以數值分析為基礎的方法區分行為策略，以全面性地衡量使用者操作介面的績效；基於績效預測模型與策略區分方法，便可發展介面佈局最佳化的演算法並產生介面設計。第二階段為驗證實驗，要求參與者操作與第一階段相同的介面以及透過演算法產生的最佳化介面，過程中皆蒐集使用者的操作時間與瞳孔直徑以評估介面操作績效與客觀心智負荷，再根據比較結果確認本系統的成效。
結果發現，本研究所提出的策略區分方法中以「依序記錄被點擊的按鍵類別」的結果與人為主觀判斷結果最為接近，但四種方法都未能與人為判斷有完全一致的分類結果，因此採用人為主觀判斷的分類結果推估整體操作績效；基於按鍵層級模型拆解操作行為，再依據霍夫曼模型推估移動所需的時間，也成功建立了績效預估模型，雖然高估個別行為的操作時間，但就所有策略平均操作時間或被採用次數最多的策略之操作時間而言，預估時間與真實時間皆無顯著誤差，故預測的準確度在可接受範圍內。以上述使用者操作行為的模擬為基礎將介面佈局最佳化，可自動產生符合幾何形狀限制與按鍵一致性的介面，且改善後介面之預估操作時間較原始介面的預估操作時間短。經由第二階段實驗的驗證也發現，以本研究提出的演算法所產生的介面佈局確實能提昇使用者的操作績效，並且不會造成客觀心智負荷上升之負面效果。此外，從第二階段實驗的訪談亦可得知參與者對於介面改善時的具體期望：首先，參與者會考量「會被連續點擊的按鍵應擺放在相鄰位置」以減少移動距離、降低操作時間；其次，當「容易點擊」與「縮短移動距離」兩個目標衝突時，參與者會主觀地選擇其中一種；最後，個人所習慣或偏好的操作模式也是影響自認理想介面的因素。整體而言，改善後介面有如預期地縮短移動距離、將接續點擊的按鍵擺放在附近，因此大致上是符合參與者期待的。
總結而言，本研究透過使用者操作行為的模擬以及最佳化演算法改善二維圖形化介面佈局，除了克服過往在使用者測試或專家評估所遭遇的限制，也經實驗驗證確實能提昇操作效率。在資源或時間有限的狀況下，可在介面開發的前期應用此方法、具體整合評估與設計變更，以便快速地擬定具易用性的介面佈局。

關鍵字

易用性 ； 操作時間預估 ； 策略區分 ； 霍夫曼模型 ； 費茨定律

並列摘要

Usability of interfaces has received much attention in recent years. Generally speaking, usability is evaluated through either user testing or expert review. These frequently-used methods however involve much human intervention, which may lead to unnatural behavior being observed, not representing the target population, and difficulty in expert development. Besides, they are cost- and time-consuming. Further, the separated processes of design and evaluation may lead to discrepancy in understanding between the two teams, and hence fail to solve the problem efficiently and in time. Therefore, this study aimed to propose a method that can simulate the user’s behavior of interface interaction and consequently optimize the interface layout, which helps reduce the restrictions caused by human intervention as well as enhancing the usability of interface.
This study focused on the performance of experienced users while operating a computer program with a two-dimensional graphical interface. Interface layout could be optimized by using the multi-objective genetic algorithm based on the simulation of user’s operations. The objectives included minimizing operation time, unifying the geometry of buttons, and avoiding inconsistent shape of the same type of button. Following these concepts, an interface with good usability while complying with general design principles could be generated.
The user’s operations could be simulated according to the characteristics of interface components, so as to generate the optimized layout. To establish the simulation model and verify whether the performance can be optimized, a two-stage experiment was proposed. At each stage, 30 participants with a 16/20 vision (naked or corrected) were recruited. The first stage of the experiment included 13 males and 17 females, with an average age of 22.57 ± 1.17 years old. The second stage of the experiment included 14 males and 16 females, with an average age of 22.63 ± 1.59 years old. The first stage of the experiment was to build the simulation model. Participants were asked to operate a desktop PC with a mouse to interact with the two-dimensional graphical interface. The collected behavioral data can be used to estimate the operation time as well as verifying methods for strategy identification. In the second stage, participants were asked to interact with both the original interface presented at the first stage and the optimized interface generated by the algorithm. During the experiment, operation time and pupil diameter were collected to evaluate the user’s performance and mental workload.
It turned out that “button types of sequential clicks” was the best one among four methods to identify strategies. However, none of the four methods yielded consistent classification of strategies with human judgment. So, strategies were eventually identified by human judgement. The operation behavior was decomposed based on the keystroke-level model, and the time required for movement was estimated according to the Hoffmann’s model. The data collected in the first stage of experiment showed that the accuracy of the prediction seems to be acceptable. Subsequently, the optimized interface layout was generated along with the predicted operation time, which was shorter than the estimated operation time of the original interface. Through the verification in the second stage of the experiment, it was also found that the interface layout generated by the algorithm proposed in this study could indeed improve the user’s performance, while not leading to a higher mental workload. Furthermore, from the interviews in the second stage of the experiment, it was found that the optimized layout was generally in line with the expectations of participants.
To sum up, this study improved the two-dimensional graphical interface layout through the simulation of user behavior and optimization algorithms. Its effectiveness had also been verified with experimental data. In the case of limited resources, this method can be applied in the early stage of interface development to integrate evaluation and redesign, so as to rapidly generate interface layout with good usability.

並列關鍵字

usability ； prediction of operation time ； identification of strategies ； Hoffmann’s model ； Fitts’ law

參考文獻

中文：

1. 王譯辰（2018）。中高齡與高齡族群之腳眼協調策略分析：以足部定位為例。清華大學工業工程與工程管理學系學位論文。

英文：

1. Acosta-Vargas, P., Salvador-Ullauri, L. A., & Luján-Mora, S. (2019). A heuristic method to evaluate web accessibility for users with low vision. IEEE Access, 7, 125634-125648.

2. Aiello, G., La Scalia, G., & Enea, M. (2012). A multi objective genetic algorithm for the facility layout problem based upon slicing structure encoding. Expert Systems with Applications, 39(12), 10352-10358.