應用情緒感知於數位機上盒之研究

Study of Emotional Perception on Digital Set-Top Box

傅郁翔(Yu-Hsiang Fu)

指導教授 ： 李仁貴

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2013年)

https://doi.org/10.6841/NTUT.2013.00755

摘要

本文主要整合數位機上盒與情緒感知介面以做為收視調查的平台，讓數位媒體業者能更直接與快速的了解到觀眾對於媒體內容的反應，以提升收視調查的時效。數位機上盒提供了許多的媒體內容，而媒體內容的表現為媒體業者關心的焦點，因此本文攝取與分析觀眾在觀看影片時的生理信號並進行情緒辨識分類，用以解析觀眾對於媒體內容的情緒反應。
文中先以挑選好的影片來誘發四位觀眾的高興、愉悅、厭惡、恐懼與一般等五種情緒，並攝取觀眾在觀看影片時的心電訊號。而心電訊號經過分析處理、取得特徵與正規化之後，將七種特徵參數透過KNN(K-Nearest Neighbor algorithm)的分類器來進行情緒的分類，以辨識出觀眾觀看影片的情緒狀態。
結果中顯示，利用影片刺激四位觀眾後，利用KNN 演算法將七種特徵值進行情緒的分類，當四位觀眾一起辨識五種情緒狀態時，其辨識率為76.09%。另外，文中也針對單一觀眾進行分析，當以自身情緒資料來分辨情緒，其辨識率最高為45.06%；而以非自身之情緒資料來分辨情緒，其辨識率最高僅為27.59%。由此顯示個別差異性會於情緒辨識有極大影響，需要持續增加觀眾的個數，以降低個別化的差異性。

關鍵字

數位機上盒 ； 心率變異度分析 ； 情緒反應

並列摘要

Nowadays, with digital Set-Top Box, variety of media contents could be provided with ease and the performance of these contents becomes the focus of concern to media industry. Under the circumstances, the aim of this study is to integrate digital Set-Top Box with emotion perception interface as an audience measurement platform to enhance the measurement efficiency and quickly learn the reaction when audience viewing these contents. This study adopts and analyses the biophysical signals when audience watching video to make emotion recognition classification, which could use for recognizing audiences’ emotion reaction to media content.
The use of pre-selected video would induce four audiences' affective response included laughing, pleasure, disgust, fear, and normal. Meanwhile, Electrocardiogram of audiences through whole process will be recorded as well. After analyzation, acquisition features, and normalization, the seven biophysical signals from Electrocardiogram would be classified by KNN(K-Nearest Neighbor algorithm) classifier to recognize audience’s affective response.
As a result, when recognized by four audience together, the accuracy of using video to induce audiences' affective response and using KNN to classify seven biophysical signals are 76.09%. Moreover, when recognized by single audience, the accuracy of using self-testing data and using training data are 45.06% and 27.59%.
Hence, individual difference will cause huge effect to the emotion recognition. To reduce deviation and increase accuracy, more experiment data would be extremely necessary.

並列關鍵字

digital Set-Top Box ； Heart Rate Variability ； emotional reactions

參考文獻

[5] 李美華、黃詩芸，台灣無線數位電視之競爭策略與節目規劃，傳播與管理研究，第九卷，第一期，2009，第63-92 頁。

[9] 彭玉賢，數位時代來臨-收視率調查面臨新變革，公共電視研究發展部，2006。

[18] 陶振超，情緒在媒介訊息處理中扮演的角色：認知取徑媒體研究之觀點，中華傳播學會年會論文，2011。

[1] L. C. Yeh, et al., “An innovative application over communications-asaservice: Network-based multicast IPTV audience measurement,” in Proc. of 13th Asia-Pacific Network Operations and Management Symposium (APNOMS), Sept. 2011, pp. 1-7.

[2] F. Alvarez, et al., “Audience measurement modeling for convergent broadcasting and IPTV networks,” IEEE Transactions on Broadcasting, vol. 55, issue 2, Jun. 2009, pp. 502-515.