針對情感商品的推薦機制－以流行音樂為例

情感商品，如音樂、電影等，與一般單純為了使用功能的功能商品有很大的不同。因為情感商品的評價與個人感受有關，情感商品在網路上通常會存在比較多主觀的評論；商品的效用也更與商品本身內容及通常能帶給使用者什麼感覺與情緒來的有關。傳統上，對於網路評論，我們通常只關注評論中所述及的商品屬性，主要在找正負傾向規則，而不會去企圖找出像是「聽了讓人感到很遺憾」這種引發人類情緒的情感商品規則。本研究以流行音樂這個情感商品為例，提出一個針對情感商品的推薦機制。首先我們先建立能了解網路評論狀況的情感標籤分類器，用於隨時了解某商品目前網路評論的情感傾向；另外也建立一個同時考慮到音樂歌詞及音質特性的音樂內容分類器，用於從音樂的內容特徵來得到某音樂商品可能音樂情感傾向。經過資料的收集、分析與訓練，網路評論分類器與音樂內容分類器的精準率、召回率與F1均達令人滿意程度，進而本研究以實驗分析在用戶悲傷情緒下應推薦的音樂來說明情感商品的推薦規則建立過程。

關鍵字

情感分析；流行音樂；意見挖掘；網路評論；推薦規則

並列摘要

Purpose-This study aims to propose a mechanism based on web reviews opinion mining and product contents (e.g., audio and lyrics in our case) for hedonic product recommendation. Design/methodology/approach - The classifiers, web review SVM classifiers and music content SVM classifiers, were proposed and a prototype was also built. Finally, we designed an experiment for exemplifying the process of determining the recommended product when the user is in a particular mood. Findings-The acceptable precision, recall, F1 ratio were obtained for the two classifiers. The experiment indicated the recommendation rule while users are in sad mood. Research limitations/implications-We only take as an example of pop music. Other hedonic products (e.g., dancing) might be more complicated to analyze their contents owing to video. Practical implications-Following our proposed mechanism, the suppliers of hedonic products would know how to recommend proper contents to users to invoke their desirable feelings. Originality/value-The proposed mechanism is brand new. As we know, there is no such a recommended mechanism for hedonic product in literature.