以使用者電影興趣檔與未知電影評價驗證偏好多樣性與偏好開放性

Abstract

傳統上認為準確度是影響推薦系統使用者滿意與否的主要因素，然而準確度指標的單調性、過於顯而易見等問題，促使研究者發展非準確度指標以改善使用者滿意度，卻發現使用者對於具有多樣性、新穎性的推薦結果偏好程度不一致的情況。同時，較少研究將使用者本身的特性納入評估過程。本研究試圖了解此種偏好不一致的情況，是否與個人特質－偏好多樣性與偏好開放性有關。 本研究以電影為推薦項目，透過張媺媺（2015）建構之偏好屬性量表測量偏好多樣性與偏好開放性，並以實驗性的電影資料平台蒐集觀影經驗並建構使用者興趣檔。再以使用者興趣檔與未知電影評價，探索受試者的觀影行為是否可以反映出個人的偏好屬性，以有效樣本293人進行相關分析，並驗證偏好屬性量表的效度。 研究結果驗證使用者的偏好多樣性與偏好開放性可以透過其行為上的表現－觀影經驗中的使用者興趣檔與未知電影評價中推斷。針對偏好多樣性，本研究以電影清單內平均相似度、電影類別、電影類別分布平均程度等方式測量電影多樣性。對於偏好開放性，本研究以感興趣的電影數量、感興趣的未知電影新穎程度、未知電影接受程度、已知電影與未知電影的平均相似度等方式測量電影新穎性。結果顯示出偏好多樣性程度越高者，其觀影經驗的多樣性越高；偏好開放性程度越高者，對已知或未知電影的接受程度越高、感興趣的未知電影新穎性越高。

It has been pointed out that the accuracy-based measures do not fully reflect the values the users derived from the recommender system, most noticeable of which are “non-obviousness” and diversity of the recommendations. To effectively improve user’s satisfaction, the key to a proper balance between accuracy, diversity, and novelty might lie in how willing the users are to explore diverse or novel items. Therefore, recommendation strategies should be applied adaptively according to the individual’s preference characteristics, namely, “preference diversity” and “openness to novelty”. Built upon previous research on the psychological scale of “preference diversity” and “openness to novelty”, a user study was conducted to test how users’ “preference diversity” and “openness to novelty” scores correlated with their past movie profile and judgment of previously unseen and unknown movies. A total of 293 participants were recruited to take part in the study, in which they were to judge a total of 220 movies so their judgment of movie seen, known but unseen, and movies previously unknown could be elicited; this is then followed by their filling out of the “preference diversity” and “openness to novelty” scales. Results showed that users’ “preference diversity” score was significantly correlated with the diversity of movies seen, as measured by average similarity, and movie genre entropy, which validated the diversity scale. As for “openness to novelty,” it was found that users with higher “openness to novelty” scores were also most likely to show interests in a higher percentage of known but not yet seen movies, unknown movies, and all movies. The correlation between the “openness to novelty” scores with the “novelty” of movies was also tested. The novelty of movies was measure both generically, by their popularity, and individually, by gauging the similarity between each individual’s judgment of seen and unknown movies. Both popularity-based and similarity-based novelty were found to be significantly correlated with “openness to novelty”, which indicated that individuals with high openness to novelty, in general, appreciated more obscure movies and movies dissimilar to those they have seen before.