Steam平台內，擁有超過3萬多個遊戲，這些眾多推出的遊戲，而難以選擇符合自身喜好的遊戲，因此消費者若要尋找符合興趣喜好的遊戲，則需花費更多的時間進行查找。若能開發有效的推薦系統，更能使消費者容易觸及到符合各自喜好的遊戲，吸引消費者進行消費。 本研究以steam平台資料集，結合傳統以使用者為基礎(user-based)推薦方式如:以雅卡爾相似度進行評分預測的協同過濾方法，以及以模型為基礎(model-based)推薦方式如:奇異值分解(SVD)、非負矩陣分解(NMF)兩方法，設計實作一電腦遊戲評量之擇優推薦系統。 藉由將原始資料集進行資料前處理，組合成使用者評分矩陣，以符合後續研究使用，再將評分矩陣對兩模型進行運算與訓練，並以擇優的方式將兩模型的預測進行不同程度的結合與實驗，產生混合的推薦結果。實驗一提出一擇優方式，在兩模型間選擇較好的組合，以此對使用者進行推薦。而在實驗二，改進了實驗一的擇優方式，對原先的擇優方式與判斷標準進行改良，發現實驗二的結果與實驗一整體結果相差不大，然實驗二所提出之方式卻更能讓使用者接觸到那些未曾接觸的遊戲。 ;here are more than 30,000 games on the Steam platform. These numerous games make it difficult for users to choose games that match their own preferences. Therefore, user need to spend more time to find games that match their preferences. If we can develop an effective recommendation system, user can easily reach games which users are interested in and also attract user to purchase. This study uses the steam platform dataset, combined with traditional user-based and model-based recommendation methods such as singular value decomposition (SVD) and non-negative matrix decomposition (NMF) method, designed and implement a selected recommender system based on computer game evaluation. For the use of research, we pre-processing the original data set, combining into user scoring matrix. Then the scoring matrix is used to train the two models, and the predictions of the two models are combined and selected optimized by different degrees of criteria. The last combined predictions produces mixed recommendation results. In the first experiment, we propose a method of choosing the best combination between the two models, so as to recommend the user. In the second experiment, we improved the method of first experiment, improved the original method and standard, and found that the result of second experiment was not much different from the overall result of first experiment, but the method proposed in second experiment allows users to access games that have not been touched.
