- 學位論文
應用圖神經網路與情感語意分析之興趣點推薦方法
A Point-of-Interest Recommendation Method Based on Graph Neural Networks and Sentiment-Semantic Analysis
摘要
在現今資訊與行動裝置普及的時代，社群網路中記錄了大量的使用者到訪興趣點(Points of Interest)的打卡與評論資料。而這些打卡資料可能會影響其他人旅遊或休閒娛樂的選擇與決策。為了使使用者能快速找到需要的資料，POI推薦成為了社群網路不可或缺的系統之一。現有許多研究提出興趣點(POI)推薦方法，根據使用者的喜好與POI特徵來預測使用者未來可能感興趣之POI。POI推薦的重要性在於能夠提升使用者體驗，提供個人化的建議以幫助使用者快速找到興趣點。因此，要如何精確分析使用者喜好與POI特徵是值得探討的議題。為了有效使用評論資料中所隱含的使用者情緒，本研究提出應用圖神經網路與情感語意分析之興趣點推薦方法。首先透過圖神經網路分析使用者的社交特徵、使用者-POI的交互特徵以及POI的地理特徵，接著使用RoBERTa語言模型分析文字評論中隱含的情感語意特徵。最後，整合上述特徵，預測使用者未來可能會到訪之POI。實驗結果顯示，本研究所提方法之推薦結果優於其他相關之POI推薦方法。本研究方法能更精確建立使用者和POI的潛在特徵和情感特徵，有效提升了POI推薦的準確性。
並列摘要
In the current era of widespread information and mobile devices, social networks have amassed extensive check-in and review data of users visiting Points of Interest (POIs). These check-in data potentially influence others' choices and decisions regarding travel or leisure activities. To enable users to swiftly locate desired information, POI recommendation has become an indispensable system within social networks. Numerous studies have proposed POI recommendation methods, predicting POIs that users may find interesting in the future based on user preferences and POI characteristics. The significance of POI recommendation lies in its ability to enhance user experience by providing personalized suggestions, thereby assisting users in efficiently identifying points of interest. Consequently, the question of how to accurately analyze user preferences and POI features is a topic worthy of investigation. To effectively utilize the implicit user sentiment in review data, this study proposes a POI recommendation method that incorporates Graph Neural Networks (GNN) and sentiment semantic analysis. Initially, the method employs GNN to analyze users' social features, user-POI interaction features, and POI geographical features. Subsequently, it utilizes the RoBERTa language model to analyze the implicit sentiment semantic features in textual reviews. Finally, it integrates the aforementioned features to predict POIs that users may visit in the future. Experimental results demonstrate that the proposed method outperforms other relevant POI recommendation approaches. This research method can more accurately establish the latent features and sentiment characteristics of users and POIs, effectively enhancing the accuracy of POI recommendations.