| 摘要: | 隨著網路的普及，我們可以輕易的在網路上找尋資料，例如，想看的影集、書籍，等等，然而資訊爆炸使得找尋想要的資料成為一件困難的事情。例如截然不同的兩位使用者，搜尋相同的關鍵字會得到相同的結果，因此如何根據使用者的興趣以及過去的紀錄進行推薦，成了一件重要的事情。協同過濾是一個成熟且廣泛應用於推薦系統中的技術，然而協同過濾同時也存在一些致命的問題，例如:冷啟動。本篇論文主要透過將基於存量(memory-based)的協同過濾中的基於使用者(user-based)協同過濾與基於專案(item-based)的協同過濾以注意力機制結合，讓我們的ACCF模型可以同時考慮兩者，使得在做推薦預測時，能有更多的資訊可以參考，並且在做推薦時透過注意力機制自動調整兩個模型所占的比重，藉此降低冷啟動帶來的負面影響。此外，考慮到傳統協同過濾無法處理使用者興趣演變的問題，因此，本篇論文的模型ACCF將使用者興趣演變納入考量。實驗結果顯示，ACCF的推薦性能優於其他的推薦演算法。此外，我們也在幾個真實的資料集上進行實驗，證明ACCF比起其他以協同過濾為基礎的推薦系統，擁有較佳的表現。;With the popularization of the Internet, we can easily find information on the Internet, such as movies, books and so on. However, the information explosion makes it difficult to find the information we want, accurately. For example, when two different users search for the same keywords, they will get the same results. Therefore, how to make recommendations based on users′ interests and users’ past behaviors becomes an important thing. Collaborative filtering (CF) is a mature and widely used technology in recommendation system. However, collaborative filtering also has some fatal problems, such as cold start. In this paper, we combine User-based CF with item-based CF with attention mechanism, so that our ACCF model can consider both of them simultaneously, so that when making recommendation prediction, more information can be referred to. Besides, the attention mechanism can automatically adjust the weight of the two models, thus reducing the negative impact of cold start. In addition, considering that traditional collaborative filtering cannot deal with the evolution of users′ interests, the model ACCF in this paper takes the evolution of users′ interests into consideration. The experimental results show that ACCF′s recommendation performance is better than other recommendation algorithms. In addition, we have also conducted experiments on several real datasets, proving that ACCF performs better than other collaborative filtering-based recommendation systems. |
