電影偏好結構多樣性及開放性之量表建立與驗證

The Development and Validation of Preference Diversity and Openness to Novelty Scale for Movie Preference Structure

張媺媺(Mei-Mei Chang)

指導教授 ： 唐牧群

國立臺灣大學/文學院/圖書資訊學研究所/碩士(2015年)

https://doi.org/10.6342/NTU.2015.02873

摘要

推薦系統除了既有的準確度之外，推薦結果的多樣性與新穎性也漸被學界重視，本研究從消費者偏好結構與心理特質上探討不同使用者對多樣性與新穎性推薦接受程度的差異，稱之為偏好多樣性與偏好開放性，以作為未來推薦系統發展的參考。本研究旨在發展電影偏好結構多樣性及開放性之量表，偏好多樣性欲反映的是消費者其偏好的廣泛、多元程度，偏好開放性則是消費者接納陌生事物的開放程度，以編製出適合衡量消費者偏好的測量工具，並驗證量表與電影觀眾真實偏好間的相關程度。
量表編製係從相關文獻中的量表、問卷之題項，與焦點團體訪談，據以編製量表題庫。正式偏好多樣性及偏好開放性量表皆各為5題題項，以李克特七點量尺測量。量表驗證則蒐集電影資料並設計電影問卷以調查消費者真實觀賞電影行為與電影偏好。最後，得有效問卷293份並進行相關分析。
研究結果顯示：偏好多樣性及偏好開放性量表在信度方面，初始及正式量表Cronbach’s α值均達0.7以上，具穩定性及內部一致性；在效度方面，經過表面效度、內容效度檢驗，並以因素分析對於量表的建構效度予以驗證，在外在效度上量表與消費者觀影行為間具有相關性，故兩量表在信、效度屬良好，具一致性、可靠性及有效性。本研究亦發現兩量表間呈現正向相關，並與消費者真實觀影行為有關聯，偏好多樣性以使用者曾經觀看過電影的類別、平均相似度和社會網絡分析方式中電影網絡的成分數為指標進行測量，偏好開放性則分析使用者所觀看的電影中屬於小眾電影以及對新電影的接受度等。結果顯示，偏好多樣性高的人，所觀看的電影彼此之間是相似度低、多樣性高的，而偏好開放性高的人，會觀看較多的小眾電影，對於新電影其不感興趣的電影比例較低。

關鍵字

偏好多樣性 ； 偏好開放性 ； 電影推薦系統 ； 偏好結構 ； 量表建立

並列摘要

There is an increasing attention in the Recommender Systems field that diversity and novelty are key qualities of recommendation results beyond accuracy.This study explores the differences between different users acceptdiversity and novelty from the perspective of personal traits and preference structure,namely, “preference diversity” and “openness to novelty.”This study aimed to develop preference diversity and openness to novelty scale for movie preference structure, construct reliable and valid instruments, and validate the correlation between the scales and movie viewers’ preferences. Preference diversity denotes how widely scattered one’s interests are. Openness to novelty represents the extent to which an individual is willing to venture out of familiarity.
Items were created based on the literature review and the results of focus group.Initial scales were then carefully examined and revisedbased on experts’ opinions.Through exploratory factor analysis and item analysis, each formal scale comprised 5 items scored on a 7-point Likert scale.Besides, film data were collected andquestionnairewas established to collectmovie-viewing behavior and movie preference.293 valid surveys were collected and correlation analyses were then performedbetween scales and movie questionnaire.
Major finding are as following:Cronbach’sαvalues were above 0.7 on both initial and formal scales, which showed stability and internal consistency.Face validity, content validity,construct validity, and external validitywere tested and it was found that the two scales hadsatisfactory consistency, reliability, and validity.Results showed that there was a significant positive correlation betweenthe scales and the number of consumers’ corresponding movie-viewing behaviors.Preference diversity was measured by movie genresthat the users had seen, their average similarity, and number of components calculated usingsocial network analysis,whereasopenness to noveltywas measured by non-mainstream films that the users had watchedand their acceptance ofunfamiliar movies.Results also showed that the higher one’s preference diversity, the more diversity was the movies s/he had watched, as measured by average dissimilarity.Furthermore, those who have higher openness to novelty tend to watch more non-mainstream films and interestedin greater number of unfamiliar movies.

並列關鍵字

preference diversity ； openness to novelty ； movie recommender system ； preference structure ； scale development

參考文獻

葉妤貞、黃芳銘、林珊如（2007）。大學生線上遊戲休閒涉入量表之發展。運動休閒管理學報，4（2），頁 16-32。

Chamorro-Premuzic, T.,& Furnham, A.(2005). Personality and intellectual competence. Mahwah, N.J. ; London : Lawrence Erlbaum Associates.

（Hanneman, R. A., & Riddle, M, 2005）

Adomavicius, G., & Tuzhilin, A. (2005). Toward the next generation of recommender systems: A survey of the state-of-the-art and possible extensions. Knowledge and Data Engineering, IEEE Transactions on, 17(6), 734-749.

Amir, O., & Levav, J. (2008). Choice construction versus preference construction: The instability of preferences learned in context. Journal of Marketing Research, 45(2), 145-158.