本論文的研究目的在於實作一個低成本的眼動儀，主要硬體是以紅外線無線攝影機來提供眼睛的紅外線影像，可見光攝影機提供景觀影像，並利用其無線傳輸的功能來裝置成頭戴式的可移動眼動儀。眼動儀的軟體部分主要是參考StarBurst演算法完成程式的編寫，除了眼動儀裝置的建構之外，本論文利用自製眼動儀來實作了三個人機介面的應用(包含Eye scrolling、Eye typing以及Eye controlled game)以及找尋AOI的心理認知實驗。 本論文也針對視線軌跡的部分提出分析之演算法，考量資料的準確性，用來分析的實驗數據主要是以iView X Hi-Speed所產生的眼動紀錄為主。在視線軌跡的比對方面提出了兩個做法，第一個是遞迴交集演算法，第二個是修改的Dynamic Time Wrapping演算法，以此兩個方法來對視線軌跡作比對。另外，為了找出在時間序列上的視線共同區域，在本論文的做法是先將2維空間的座標序列根據切割尺寸來轉換成1維的座標序列，然後再以Longest Common Sequences(LCSS)的方式作分析，以期望能以資訊科學的角度來分析視覺軌跡所呈現的心理資訊。

在現今的基於內容的圖像檢索的研究中，會將人的主觀認知考慮進去。因為傳統的圖像檢索中採取低階特徵來找出圖片上可能的重要區域的方法和人的感覺還是有著相當大的語意上的鴻溝。然而藉由考慮人對圖片的主觀認知，可以讓人找到對它而言圖片上重要的部分，再去做圖像檢索，找出使用者想要的圖片。這樣的作法是比較自然且直觀的。還能達到個人化的效果，因為每個人對同一張圖片上覺得重要的物體可能不盡相同。在本論文中的圖像檢索系統採用眼動軌跡當作人的主觀認知來輔助檢索。因為在心理學的研究中有提到，人在看圖片的時候會有較多的凝視點落在他覺得重要的區域上。所以藉由這個理論，本論文利用使用者看圖片的眼動軌跡即時的調整圖片上物體的重要性。最後將重要性高的數個物體去做圖像檢索，找出含有這些對這個使用者是重要的物體的圖片。經由實驗證實，眼動軌跡輔助圖像檢索的確可以減少不重要的物體對圖像檢索的干擾，繼而可以提升圖像檢索系統的效能。 / Recently, researches in Content-Based Image Retrieval (CBIR) focuses on incorporation of knowledge about human perception in the systems’ design and implementation process. This enables the design of more natural and intuitive image retrieval techniques in order to overcome some of the challenges faced by modern CBIR system such as the difficulty to extract important regions of an image. By researches of psychology, user’s eye tracking reflects his interest. So, in my CBIR system, user’s eye movements were used online to adjust the importance for objects in query image. Thus in my system, only those images with important objects will be retrieved. One experiment was performed: record the eye movement of participants on query images. Then compare my approach with a classic CBIR system according to performance. The results reveal that higher retrieval performance of my image retrieval system because of decreasing the influence of not importance objects to image retrieval system.

近年來隨著電腦科技的進步，發展出了各種人機介面的應用。不同於傳統的滑鼠與鍵盤，新的人機介面往往在其他不同的領域能夠得到更好的發揮，眼動（eye movement）即是一個例子，利用眼球移動在螢幕上的眼動軌跡來操控電腦。另一方面，螢幕所呈現的內容也會影響使用者的眼動行為，例如閱讀文章時不同的文章內容就會影響使用者在看文章時的眼動行為，因此我們就可以從眼動軌跡去推測出使用者對於螢幕的內容（如文章或圖片）的反應。本論文的目的就是要找出閱讀文章時的眼動軌跡和閱讀者對該文章的理解程度之間的關連性，最後發展出一個系統，利用閱讀文章時的眼動軌跡就能去推測閱讀者對於文章的理解程度，而無需後續的閱讀測驗。 / 一般人通常對自身的眼動軌跡的進行方式不甚了解，認為閱讀文章的眼動應該是順著字句的進行而移動。其實仔細觀察細小的眼動軌跡會發現比想像中複雜的多，例如當我們閱讀到文意較為困難的片段時，通常都會放慢速度，或是反覆閱讀，又例如看到自己有興趣的主題時會凝視在某些區塊較久的時間等等行為，這些都是在各種不同的情境之下做出相對的反應。本論文提出了資料探勘的方法來分析複雜的眼動軌跡，能針對每個人不同的閱讀模式找出不同的眼動規則。此方法能適應各種使用者的閱讀習性，對閱讀理解程度的預測能達到更精確的效果。 / With the advance of the computer technology, there are many kinds of HCI (Human Computer interface) applications been developed in recent years. Different from mouse and keyboard, new interface will bring more benefits in different area, and eye-movement is an example. It extracts the trace of eye movement to control the computer. Furthermore, the screen content will also affect the eye-movement. For example, when the user reads the articles, the content will affect the eye-movement, so we can speculate the reaction of the user after seen the screen content (such as article or picture) by eye-movement. The goal of the paper is to find out the relationship between eye-movement and the degree of comprehension, and develop a system which can automatically measure reading comprehension without any follow-up comprehension test. / Most people don’t realize how there eye move. They think eye movements should be carried out along the words. In fact, eye movement will find more complex than imagined, for example, when we read the context of the article is more difficult, we often slow down or read over and over again. In another example, people will gaze at some of the subjects that they are interest in for a longer period of time. These reactions correspond to a variety of different situations. This paper presents a data mining approach to analyze complex eye movement. It can find out the different rules of eye movement for users who have different reading strategy. This method will be able to adapt to a variety of users reading habits, and make more accurate prediction with the degree of the comprehension.

閱讀是吸收知識的途徑，不同的閱讀模式所帶來的閱讀成效也會不同。如何透過機器學習的方式，從凝視點找出閱讀行為的關聯性，將是本研究的目標。實驗選擇低成本眼動儀紀錄讀者閱讀過程中的眼動資料，採用dispersion-based演算法找出凝視點，以計算凝視點特徵，包含凝視時間、凝視距離、凝視位置以及凝視方向。 本研究將閱讀模式分成五種類別，包含快讀、慢讀、精讀、跳讀與關鍵字識別，透過不同文章的呈現，引導30位測試者遵循其內容進行閱讀，藉此收集不同行為模式的眼動資料。實驗流程中所有的眼動資料會隨機被分成為兩份，依序建立不同維度的訓練資料，由交叉驗證的分類結果找出理想之特徵與維度。以每次挑選6位測試者的眼動數據為測試資料進行5次分類驗證，其平均正確率為78.24%、74.19%、93.75%、87.96%以及96.20%，均達到不錯的分類結果。 / Reading is one of the paths to acquire knowledge. The efficiency is different when different reading patterns are involved. It is the objective of this research to classify reading patterns from fixation data using machine learning techniques. In our experiment, a low-cost eye tracker is employed to record the eye movements during the reading process. A dispersion-based algorithm is implemented to identify fixation from the recorded data. Features pertaining to fixation including duration, path length, landing position and fixation direction are extracted for classification purposes. Five categories of reading pattern are defined and investigated in this study, namely, speed reading, slow reading, in-depth reading, skim-and-skip, and keyword spotting. We have recruited thirty subjects to participate in our experiment. The participants are instructed to read different articles using specific styles designated by the experimenter in order to assign label to the collected data. Feature selection is achieved by analyzing the predictive results of cross-validation from the training data obtained from all subjects. The average classification accuracies in five-fold cross-validation are 78.24%, 74.19%, 93.75%, 87.96% and 96.20% using the eye movements of the six randomly selected subjects as test data.

高互動的閱讀模式能夠引發學習者更高的學習動機與意願，同時也帶給學習者更直覺的閱讀學習體驗。過去使用眼動進行讀者與文本互動的閱讀模式研究仍缺乏有效的實驗驗證其對於提升閱讀理解成效的助益，因此本研究使用眼動追蹤儀器，結合網頁技術與閱讀標註概念，發展一套基於眼動技術發展之視覺互動文本閱讀模式，同時探討此一閱讀模式在支援英語文本閱讀之學習成效、學習時間、認知負荷、科技接受度與閱讀歷程序列等面向上，是否優於滑鼠操作之閱讀標註互動文本閱讀模式。 　　研究結果顯示：(1)採用眼動操作的實驗組學習者閱讀問答式說明文文體結構具閱讀標註補充之互動式閱讀文本進行閱讀的學習成效顯著優於採用滑鼠操作的控制組學習者；(2)眼動操作的實驗組學習者在閱讀問答式說明文文體結構具閱讀標註補充之互動式閱讀文本進行閱讀的學習時間、標註觸發次數、標註觸發時間、閱讀序列長度與單位學習時間標註觸發次數均顯著優於滑鼠操作的控制組學習者；(3)採用眼動技術操作的場地獨立型學習者對於閱讀問答式說明文文體具有顯著學習成效；(4)採用基於眼動技術發展之閱讀標註互動文本閱讀模式具有引導閱讀者規律進行前後文參照回視之特性；(5)文體結構差異對於採用眼動技術操作與滑鼠操作具閱讀標註補充之互動式閱讀文本之學習成效產生影響。 　　最後，根據研究結果，本研究提出眼動技術實際應用於閱讀標註互動文本閱讀的教學建議，並針對未來研究方向提出建議。

2012年9月，NCC終於初步解禁置入性行銷，並公布有關於電視節目置入的暫行條款。事實上，美日等先進國家之置入性行銷早已成熟，韓國政府也鼓勵韓國企業進行置入，以資助影視產業的發展，而台灣雖然過去已有業者進行置入性行銷，但囿於法令限制，在置入性行銷的發展仍是相當遲緩的，也因此，台灣在影視產業一直得不到到足夠的資金去拍攝優質影片。 幸好，目前政府已逐步開放置入性行銷，不過目前國內廠商對於置入性行銷收費標準仍然相當混亂，因此，本研究期盼能透過釐清各置入點之效果以提供國內廠商收費的參考指標。過去研究置入性行銷效果多以問卷方式調查，然而使用問卷方式調查會有事後回憶誤差等問題，本研究採用眼動儀來直接測量受測者之收視過程，紀錄受測者之眼球掃視軌跡以及凝視個數、凝視時間，輔以問卷測量消費者之回憶度、辨識度、品牌態度以及購買意願，以更嚴謹的方式找出最佳置入點。 本研究參考過往文獻，找出涉入度以及置入手法兩個自變項，確認涉入度與置入手法對於消費者在觀看戲劇時是否有不同的感受，怎麼樣的情形置入效果會最佳。 研究以實驗法進行，自變項為涉入度以及置入手法，因此本研究將受測者分為高涉入度顯性、高涉入度隱性、低涉入度顯性、低涉入度隱性共四組。研究結果發現，顯性置入與隱性置入在受測者回憶度上有顯著的差別，且在隱性置入的情況，低涉入度的產品相較於高涉入度的產品更容易被忽略。 研究也發現，受測者在觀賞戲劇時，影響置入效果最大的因素是置入產品與劇情的相關性，其餘的影響因素還有大小、顏色、位置、競爭等要素， 建議未來廠商在進行置入時可依據本身產品特性並考慮各種影響因素，以達到最佳置入效果。 / Because of our government’s restriction, in the past few years, product placement has always been a debated issue. Due to these restrictions, our TV and movie industries are always lack of funding to produce high quality content. Fortunately, on 2012 August, NCC finally agreed gradually deregulation. However, our development of product placement still falls behind other countries and the charging standard is still in a mess. In order to provide domestic firms a charging reference, we started this research, hoping to understand the effect of every location. Instead of using questionnaire as our only measurement, we also use eye tracking to help us understand the effect more clearly. According to the references, we found out product placement and involvement as our independent variables and divided our subjects to four groups. Those are high involvement explicit, low involvement explicit, high involvement implicit and low involvement implicit. The research outcome shows that different location would bring to significantly different degree of recall. People who were exposure to the explicit drama clip would have higher degree of recall than those who were exposure in the implicit drama clip. And we also found out that if our product was a low involvement product, the product would not be noticed especially in the implicit drama clip. We also found that when people saw a drama, the most important factor to affect their memory was the product’s connection with the plot. Other factors were color, size, placement and competition. We suggested that when domestic firms want to make a product placement, they could consider their own product’s attribute and all these factors together to make the best product placement effect.

目前偵測一般人情緒的方式大部分在研究人的行為，例如：臉部表情，以及分析人體的各項生理數值，例如：心跳、體溫以及呼吸頻率。然而這些研究只單純探討人的外在行為或生理訊號在不同情緒下的變化，而人的眼睛包含外在行為跟生理訊號，本研究將探討不同情緒下眼睛有什麼特別的反應。 我們先制訂一套實驗流程，在流程中我們以不一樣的情緒圖片給予受測者刺激，然後記錄受測者的眼動反應，並且讓受測者回報自己的情緒狀態。本研究也記錄受測者在情緒刺激下的眼動反應，並將眼動之反應轉換成序列資料，再針對不同情緒下的序列建立隱藏馬可夫模型（Hidden Markov Models：HMM）。希望藉著情緒模型，從眼動行為中偵測受刺激者處於何種情緒狀態。 本研究發現人在看圖時會依據對圖片內容的好惡，產生有意義的眼動反應。我們利用相對應的眼動反應建立情緒辨識系統，在辨識三種情緒時，辨識率能夠達到六成。

本研究是探討中文雙字詞的語意透明度是否影響詞彙在心理辭典中的表徵方式。實驗一、二操弄詞彙的語意透明度、字頻及詞頻，以詞彙判斷作業進行。結果發現顯著的語意透明度效果，語意不透明的詞（成語詞）之反應時間和錯誤率均較語意透明的詞（組合詞）少；詞頻效果也達顯著，高頻詞的反應時間和錯誤率均較低頻詞少。而字頻與語意透明度間有交互作用存在，組合詞有顯著的字頻效果，成語詞則不受字頻影響。實驗三、四則利用眼動追蹤技術，探討不同語意透明度、字頻、詞頻的詞彙在語句中的閱讀處理歷程。結果發現，在有關凝視時間的指標上，均出現顯著的語意透明度效果和詞頻效果。成語詞的字頻效果除了出現於第一次凝視時間指標外，在其餘指標上均無字頻效果。組合詞的字頻效果則出現於第一次凝視時間及總凝視時間上，而第一次經過的其餘眼動指標也有存在字頻效果的趨勢。本篇研究以Taft annd Zhu(1995)的多層次交互激發表徵模式為架構，提出一中文成語詞與組合詞的詞彙表徵模型。此模式顯示組合詞在詞素層次具有詞素表徵，在詞彙層次具有全詞表徵。而成語詞在詞素層次無對應的詞素表徵，是以全詞方式表徵於詞彙層次中。 關鍵字：眼動、語意透明度、心理辭典 / The present study aims to investigate the semantic transparency of Chinese di-syllabic words whether influences the representation in mental lexicon. Two sets of experimental tasks were adopted. In the first set (experiment 1 and 2), the target words were varied in terms of semantic transparency, character frequency and word frequency. Subjects were asked to deicide whether the target word presented is a word or not. A two-way ANOVA indicated main effects for semantic transparency that idiomatic words were responded faster than compositional words, for word frequency that high frequency words were responded faster than low frequency words. There were also significant interactions between character frequency and semantic transparency. In the compositional condition, the results clearly showed a significant character frequency effect, but in the idiomatic condition, the results showed no character frequency effect. In the second set (experiment 3 and 4), eye movements of subjects were monitored as they read sentences containing target words that varied on terms of semantic transparency, character frequency and word frequency. There were significant semantic transparency effect and word frequency effect in the fixation duration measures. Except the first fixation duration, there was no character frequency effect with idiomatic words in other measures. The results showed a significant character frequency effect with compositional words in the first fixation duration (FFD) and total viewing duration(TVD), and showed a marginal significant character frequency effect in other first pass measures. The present study proposed a model of mental lexicon based on the Interactive-activation Model (Taft & Zhu, 1995). In our model, the compositional words have morpheme and whole word representation, and the idiomatic words are processed as single units, have whole word representation but no morpheme representation. Key words：eye movement、semantic transparency、mental lexicon

本研究主要目的是想要探討當人們面對不同重要性的決策時，情緒對決策的影響是否也會隨之改變，研究者利用眼動及生理測量去量測在不同的實驗情境下，受測者眼動型態以及生理反應。本研究兩個實驗皆為受試者間設計，獨變項分別為情緒(快樂與悲傷)與決策重要程度(高與低)。在實驗一中，研究者給予受試者聆聽不同情緒的音樂，並且藉由指導語操弄決策重要性。另外，受試者被告知在每個嘗試次中必須要挑選出一台想要購買的筆電，以完成決策。根據實驗一的結果，研究者發現無論在眼動指標或者是生理指標皆有情緒與決策重要性的交互作用。然而因為實驗設計的因素，受試者訊息處理的方式並沒有辦法很清楚的被呈現，並且也沒有發現任何早期訊息的篩選過程。在實驗二中，為更進一步檢驗情緒與決策重要性交互作用的情況，實驗中增加了額外的捷思線索(Heuristic cue)。因為增加線索，在實驗二中，產品的訊息便增加了衝突性以及更能夠區別出不同訊息的特性(有用的與無用的訊息)。實驗二的結果呈現出早期訊息篩選的歷程，並且也顯現出決策的動態過程。早期的情緒與決策重要性的交互作用顯現出，快樂與悲傷的受試者在面對高重要性的決策時，的確會有早期的訊息篩選歷程。而在晚期的歷程中，則顯現出快樂的受試者在面對高重要性決策時，會花比較多的時間再評估對他們而言有用的訊息。另外，面對高重要性決策的悲傷受試者，會比面對低重要性決策的悲傷受試者更快完成決策。文中結果將根據Russo 與Leclerc’s(1994)的階段分析與Isen(1993)跟Forgas(1991)所提出的假設進行更詳盡的討論。 / In the present study, the researchers examined how the emotions influenced the decision process when the participants faced decisions with different importance. The eye movement patterns and psychophysiological effects were observed in four different conditions (high importance/positive; high importance/negative; low importance/positive; low importance/negative). In experiment 1, emotion was manipulated by listening music, and decision importance was manipulated by instruction. SCR, HR (heart rate), and eye movement measurement were recorded. The results showed the interaction of emotion and cognition, both in eye movement and psychophysiologcal effect. However, due to the experiment design, it was hard to tell whether the information was processed analytically or heuristically and no early information selective process revealed. In experiment 2, an additional cue was provided in order to examine the interaction between the emotion and decision importance with more differentiated and conflicting alternatives. The results of experiment 2 revealed the early information selective process and showed dynamic process in decision making. The early emotion and cognition interaction showed the happy and sad emotions with high decision importance did process the information selectively. While in the later interaction, it showed the happy emotions with high decision importance spent more time to evaluate the useful information. Moreover, the sad emotions with high decision importance completed the task faster than those with low decision importance. The results were discussed from Russo and Leclerc’s (1994) stage analysis and hypotheses proposed by Isen (1993) and Forgas’s (1991).

本文主要探討中文聲調在口語字彙觸接過程中所扮演的角色。實驗一藉由眼動實驗中的Visual World Paradigm作業，觀察中文聲調影響口語字彙辨識的時序歷程。受試者在聽到指導語和目標字之後，用滑鼠在螢幕上點選聽到的目標字，例如，螢幕上出現的字包含一個目標字：「摸」、一個競爭字(與目標字只有聲調相同：「挖」或是與目標字只有音段相同：「抹」)，以及兩個聲調與音段和目標字完全不同的無關字：「怒」、「菊」。為了觀察目標字、競爭字及無關字在口語字彙處理時的競爭，我們會計算各個字彙的凝視比例。實驗一中由於聲調與目標字相同的競爭字與目標字的第一個音段就開始產生差異，因此未觀察到聲調早期介入的影響。實驗二透過與實驗一相同的實驗程序及方法，操弄目標字和競爭字中聲調和前兩個音段(Cohort)的異同以探測更早期的聲調影響。螢幕呈現包含一個目標字「湯」、一個競爭字(前兩個音段和聲調皆與目標字相同：「胎」，或是只有前兩個音段相同但聲調與目標字不同：「泰」)，以及兩個聲調與音段和目標字完全不同的無關字「剖」、「痕」。結果顯示，聲調在語音訊息前兩個音段時就會產生影響，也就是聲調的影響在語音結束前即有作用。再者，本文發現聲調無法單獨且獨立地對於語音辨識產生影響，此看法與聲調表徵需以“toneme” node獨立地存在於the modified TRACE model的看法不盡相同 (Malins & Joanisse, 2010; Ye & Connine, 1999; Zhao, Guo, Zhou, & Shu, 2011)。 / The present study aims to examine the role of tonal information during Mandarin Chinese spoken character recognition. Two eye-tracking experiments were conducted with the visual world paradigm, which participants heard a Chinese monosyllabic character and used a mouse to click on the corresponding character in a visual array of 4 characters on the screen. Experiment 1 manipulated the relationship between the spoken target characters and written characters on the screen, including a target (e.g., /mɔ1/‘touch’), a tonal competitor (the tone was the same as target except segment: e.g., /wa1/‘dig’) or a segmental competitor (the segmental structure was the same with the target except tone: e.g., /mɔ3/ ‘wipe’), and two unrelated distractors (the segments and tone were different from target: e.g., /nu4/ ‘anger’, and /tɕy2/ ‘chrysanthemum’). The fixation proportions on target, competitors and the unrelated distractors were computed during the unfolding of the auditory target stimuli. The results showed tonal difference was detected before the end of auditory stream. However, no early involvement of tonal information was found, which may due to the tonal competitor and target shared no segment from the first phoneme. In order to examine the earlier tonal processing, Experiment 2 manipulated two types of cohort competitors sharing the initial two segments with the target (e.g., /tʰɑŋ1/ “soup”), a cohort-tone competitor, e.g., /tʰaj1/ “fetus” (both tone and initial two segments are the same with target) and a cohort-only competitor e.g., /tʰaj4/ “peaceful” (initial two segments is the same with the target but with different tone). Result showed that tone affected spoken character recognition while processing the two initial segments. In addition, tone could not affect spoken character processing independently, which might be inconsistent with the assumption that tone is a separate level of representation, called “toneme” node, in the modified TRACE model (Malins & Joanisse, 2010; Ye & Connine, 1999; Zhao et al., 2011).