Spotify用戶心理屬性對於不同音樂發掘工具效能的影響

Abstract

本研究旨在探討Spotify用戶在面對兩種音樂發掘工具（「標籤搜尋」與「每週新發現」）時，其心理偏好屬性對推薦系統效能的影響。研究以音樂涉入程度、音樂自我認同、偏好多樣性與偏好開放性四個心理構面為核心變項，輔以五大人格特質作為補充分析，採用實驗法搭配Spotify API真實聆聽資料與問卷調查，針對144位Spotify用戶進行配對樣本設計。使用者在兩種推薦情境下分別聆聽播放清單並進行評分，研究結合信度分析、因素分析、t檢定、逐步迴歸、線性混合模型與結構方程模型進行統計分析。 結果顯示：「每週新發現」在新穎性、多樣性與驚喜性上明顯優於「標籤搜尋」，但熟悉性與平均歌曲喜好評分則相對較低，兩者在整體滿意度上無顯著差異。熟悉性與播放清單多樣性為滿意度的重要預測因子，而非新穎性。此外，「音樂自我認同」與「每週新發現」具顯著交互作用，顯示音樂在自我建構中扮演關鍵角色之使用者，更能從探索性推薦中獲得正向體驗。而五大人格特質雖具心理解釋力，但在推薦評價上預測力相對較弱。 本研究強調心理屬性在音樂推薦中的調節角色，支持「探索與準確的權衡」理論，並提出「邊界探索」（boundary exploration）概念，作為推薦系統個人化設計的潛在方向。研究亦指出目前推薦演算法之黑箱特性與實驗場域侷限，建議未來可擴大樣本範圍，並進行自然使用情境下之追蹤研究，以提升推薦系統的精準性與使用者接受度。

This study investigates how Spotify users’ psychological preference traits influence their evaluation of two music discovery tools: a seed-based search (representing accuracy-driven recommendations) and Discover Weekly (representing exploration-oriented recommendations). Four music-related psychological constructs—music involvement, music self-identity, preference for diversity, and openness to novelty—were examined, with the Big Five personality traits included as supplementary variables. A within-subject experimental design was conducted with 144 Spotify users, combining survey data with Spotify API listening histories. Participants experienced both discovery tools and evaluated the generated playlists, and results were analyzed using reliability testing, factor analysis, paired t-tests, stepwise regression, linear mixed models, and structural equation modeling (SEM). Results showed that Discover Weekly outperformed seed-based search on perceived novelty, diversity, and serendipity, but scored lower on familiarity and average track ratings. No significant difference was found in overall satisfaction between the two tools. Familiarity and perceived diversity emerged as significant predictors of satisfaction, while novelty was not. Importantly, music self-identity had a positive moderating effect on the impact of Discover Weekly, suggesting that users who view music as central to their identity derive more value from exploratory recommendations. In contrast, the Big Five traits had weaker associations with recommendation outcomes. The study highlights the moderating role of psychological traits in music recommendation, supporting the theory of balancing familiarity and exploration. The notion of "boundary exploration" is introduced to describe recommendations that expand users’ preferences while maintaining stylistic coherence. Limitations include reliance on college-aged samples, and the experimental setting’s deviation from naturalistic use. The opaque nature of Spotify’s algorithms might introduce confounding variable to our assertation that music self-identity is conducive to openness to personalization. Future research should address these gaps through broader samples and in-situ usage studies to enhance the personalization and acceptance of recommendation systems.