利用時間和語義訊息結合趨勢感知模組之推播通知推薦系統

Abstract

推送通知服務是一種高效、經濟且用戶友好的行銷工具，能夠即時向用戶發送消息，提升用戶參與度並為網站帶入流量。然而，在推送通知推薦服務中，用戶通常以被動方式接收通知，不太主動尋找或參與內容。 因此在推播通知的場景之下要實現精準推薦的目標，我們需要解決幾個挑戰，包括同時考慮用戶的時間和上下文偏好、處理用戶點擊行為的動態特性，以及處理用戶和推播內容之間有限的交互資料。為了應對這些挑戰，我們提出了利用時間和語義訊息結合趨勢感知模組之推播通知推薦系統 (Temporal and Contextual Trend-aware Transformer Push Notification Recommender; Push4Rec)，這是一種專門為新聞文章設計的推播通知推薦模型。Push4Rec 採用多個關鍵學習器，集成時間和上下文訊息提取器以有效地提取相關且可操作的資訊。 它具有時間興趣學習器、上下文興趣學習器和趨勢感知學習器，使模型能夠分別衡量點擊行為慣性、提取用戶時間和上下文偏好，並確定近期趨勢的影響。 此外，我們還開發了融合功能和門控網路，可以靈活且全面地提取用戶點擊偏好。我們使用了合作公司的真實推播通知數據集對Push4Rec進行評估，實驗結果明確展示了每個學習器的獨特貢獻以及它們後續融合的重要性。 通過利用不同學習器的優勢，我們提出的模型優於其他的基準點擊率模型，在所有評估指標中提供了最優異的結果。 我們的研究結果顯示，Push4Rec不僅能夠有效地捕捉和綜合各種用戶行為和偏好，還證明了其在適應新聞推薦動態環境中快速變化的用戶-推播內容交互方面的有效性。 因此，我們相信 Push4Rec 此新穎方法在推送通知服務方面樹立了新標準，推動了個人化推薦系統領域的發展。

Push notification services provide an efficient, cost-effective, and user-friendly marketing tool that delivers real-time messages to users, enhancing user engagement and increasing website traffic. Nevertheless, in the context of push notification recommendation services, users tend to interact passively, receiving notifications without actively seeking or engaging with the content. Consequently, for precise recommendations, Click-Through Rate (CTR) prediction for push notifications requires addressing challenges such as user temporal and contextual preferences, the dynamic nature of user click behavior, and limited interactions between users and items. In response to these challenges, we propose Push4Rec, a novel push notification recommendation model designed explicitly for news articles. Push4Rec can handle issues such as time-sensitive interactions, sparse user-item interactions, rapidly evolving item candidates, and privacy preservation. Push4Rec employs multiple critical learners, integrating temporal and contextual information extractors to effectively distill relevant and actionable information. It features a temporal interest learner, a contextual interest learner, and a trend-aware learner, enabling the model to gauge click behavior inertia, extract user temporal and contextual preferences, and ascertain the impact of recent trends, respectively. Moreover, we have developed a fusion function and a gating network to allow for a flexible and comprehensive extraction of user click preferences. We assessed Push4Rec using a real-world push notification dataset from our partnering company, with the experimental results illuminating the unique contributions of each learner and the significance of their subsequent fusion. By capitalizing on the strengths of different learners, our proposed model has outperformed established benchmark click-through rate models, delivering state-of-the-art results across all evaluation metrics. Our findings reveal that Push4Rec is not only efficient in capturing and synthesizing various user behaviors and preferences, but it also proves itself as an effective tool in adapting to fast-changing user-item interactions in the dynamic environment of news recommendation. Thus, we believe that Push4Rec, with its novel approach, sets a new standard in push notification services, driving forward the field of personalized recommendation systems.