以排行榜與線上評論為依據的行動遊戲推薦策略效能比較

推薦系統對於線上商品及服務的銷售與購買有極大的貢獻。其中，行動軟體因其種類、數量及相關資訊量龐大，使消費者在尋找其有興趣的標的時產生相當大的困難。因此，主流的行動軟體平台Google Play和Apple’s App Store均提供推薦機制協助用戶做出下載決定，唯這些平台的推薦並未考慮消費者個人的偏好。本研究利用Apple's App Store台灣內行動遊戲的銷售榜與線上評論資料，實作包括：「銷售榜」、「也購買」、「偏好分群」及「品質*偏好」四種非個人化、半個人化及個人化推薦策略，並透過實驗設計決定各推薦策略最佳參數，再使用實驗法以修正後準確率及召回率比較其效能。本研究的實驗結果顯示「偏好分群」>「也購買」>「品質*偏好」>「銷售榜」，最後，本研究討論歸納四種推薦策略的優點及缺點。

關鍵字

推薦系統；行動遊戲；軟體品質；線上評論；銷售榜

並列摘要

Recommender systems have greatly contributed to the sale and purchase of online products and services. Mobile applications, as one of the online products, cause consumer confusion due to their huge variety, massive quantity, and overloaded information. Thus, major mobile application platforms, Google Play and Apple’s App Store for example, provide their recommendation mechanisms to facilitate users’ download decisions. However, so far, those mechanisms offer merely general recommendations and their recommendations do not take into consideration consumers’ personal preferences. This study gathered the game rankings and online reviews from Apple’s App Store Taiwan. Accordingly, this study implemented four types of impersonalized and personalized recommendation strategies, including "download ranking", "also bought", "preference cluster" and "quality*preference". Furthermore, this study conducted experiments to determine the near-optimal parameter values for each strategy, and then compared the performance of the recommendation strategies using the metrics of accuracy and recall. The experimental results revealed that preference cluster better than also bought better than apps quality and quality*preference better than download ranking. Finally, this study discussed and concluded the advantages and disadvantages of the four recommendation strategies.