通常，基於深度學習的推薦系統有許多不同的混合推薦方法，這些方法結合了協同過濾和基於內容的過濾，而且它們大多使用內容特徵來獲取輸入資訊，例如使用者和項目資訊、評論文本、輔助訊息等，以提高推薦性能。 然而，在純協同過濾的情況下，我們沒有像混合式推薦系統那樣豐富的輸入資訊。儘管如此，我們相信從協同過濾中也可以獲得有用的輸入資訊。通過實現不同的協同過濾方法，可以從執行結果中提取類似於內容資訊的附加訊息作為輸入，來豐富推薦系統的輸入資訊。 因此，本文提出了一種同時結合基於模型和基於記憶的協同過濾的深度學習推薦系統，並從兩者的執行結果中提取有用的資訊作為輸入，將其應用到我們提出的模型中以增加輸入資訊。我們在兩個公開的 MovieLens 資料集上進行實驗，大量的實驗結果證明我們提出的模型比其他現有方法具有更好的性能。 ;Generally, recommendation systems based on deep learning have many different hybrid recommendation methods, which integrate collaborative filtering and content-based filtering. Most of them use content features to obtain input information, such as user and item information, review text, side information, etc., to improve recommendation performance. However, in the case of pure collaborative filtering, we do not have as rich input information as a hybrid-based recommendation system. Nevertheless, we believe that useful input information can also be obtained from collaborative filtering. By implementing different collaborative filtering methods, additional information similar to content information can be extracted from the execution results as input, enriching the input information of the recommendation system. Therefore, this paper proposes a deep learning recommendation system that combines model-based and memory-based collaborative filtering at the same time, and extracts useful information from the execution results of the two as input, and applies it to our proposed model to increase input information. We conducted experiments on two public MovieLens datasets, and a large number of experimental results prove that our proposed model has better performance than other existing methods.
