基於內容並結合趨勢感知的新聞推薦系統

An Efficient Trend-aware Item-based News Recommendation System

林懷宇(Huai-Yu Lin)

指導教授 ： 林守德

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202000302

摘要

本研究旨在為新聞建立一前後文推薦系統。目標是根據使用者當前閱讀的新聞來預測接下來會想繼續閱讀的新聞。為了解決這個問題，我們首先觀察了真實世界的新聞資料集，並且觀測到在新聞推薦上，時間扮演著重要的因素。大多數的新聞會在發佈後的一小時內來到點擊的高峰，並且迅速的在二十四小時後降溫，也因此冷啟動造成的問題在新聞推薦中影響甚巨，同時這個特性也造成以協同過濾為基礎的推薦系統難以在新聞有良好的表現。為了解決此問題，我們採用了以內容為基礎的模型作為推薦系統的基底。接著，使用 GRU 來進行序列資料的預測，來推估每篇新聞未來的受歡迎程度。最後，我們考慮新聞的各種特徵，如：候選新聞的受歡迎程度、發布時間、與當前新聞的相似度，將這些特徵輸入深度學習的模型並對推薦分數做預測，以預測使用者下一篇會點擊的新聞。我們在線下及線上的實驗，都顯示出我們的模型可以抓到新聞受歡迎程度的變化趨勢，並且有更好的推薦表現。

關鍵字

前後篇新聞推薦 ； 趨勢感知 ； 內容基底

並列摘要

In this thesis, we aim to design a content-based filtering recommendation system that is trend-aware and efficient enough to be performed online. The purpose is to predict which news a user will read after his or her last reading news. To solve this problem, we first observed the real world data and found that most news would be popular in 1 hour after being published, while in the other hand, it would be non-popular just after 24 hours. Hence, the cold-start problem is critical in news recommendation. To solve cold start problem, firstly, we use content-based model as our foundation. Second, we use a GRU model to perform time series forecasting so that we can monitor news popularity efficiently. Finally, by considering different features of news, such as freshness, popularity, similarity to previous news, the model will rerank the ranking score and choose items with the highest scores as recommendation items. We experiment our model on offline and online task and shows that by considering popularity the recommendation system can perform better on both offline and online task.

並列關鍵字

next news recommendation ； trend-aware ； content-based

參考文獻

Quoc Le and Tomas Mikolov. Distributed representations of sentences and documents. In International conference on machine learning, pages 1188–1196, 2014.

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pretraining of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805, 2018.

Minmin Chen. Efficient vector representation for documents through corruption. arXiv preprint arXiv:1707.02377, 2017.

Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme. Bpr: Bayesian personalized ranking from implicit feedback. In Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence, pages 452–461. AUAI Press, 2009.

Jason Weston, Hector Yee, and Ron J Weiss. Learning to rank recommendations with the k-order statistic loss. In Proceedings of the 7th ACM conference on Recommender systems, pages 245–248. ACM, 2013.