採用線上評論之旅遊景點推薦

Abstract

線上旅遊評論記錄了實用的旅遊資訊和個人經驗，是旅客在出發到一個陌生景點遊玩前的重要參考資料，且對於選擇自助旅行的背包客們來說，景點推薦比整個行程的推薦更為實用。當處理線上旅遊評論時，我們便面臨了兩個問題，一是兩位評論作者到同一旅行景點時所觀察與在意的部份可能不相同，因此旅遊推薦系統只參考評分資料並不能完全代表使用者的喜好，二是旅遊評論使用的字詞太過生活化且模糊，導致使用字詞作為向量維度會失去旅遊特徵間的隱藏關係。 本研究提出的旅遊景點推薦系統採用線上旅遊評論作為資料，使用了旅遊本體將字詞對應到相關的概念及類別作為觀點向量，並且結合計算使用者間觀點向量的相似度與評論有用度評分作為使用者評論品質的指標來產生預測分數。最後我們利用Epinions.com做為評估的資料來源，實驗結果顯示本研究所提出的方法比傳統方法能更準確預測景點推薦分數。

Travel destination recommendation is more functional to a backpacker than a whole trip plan recommendation. For travelers who have never been to the target destination, online opinions, which reveal useful information and practical experiences, are important materials in information search stage. However, merely simple rating cannot represent user preference in travel opinion domain because the viewpoints of the two reviewers are in different angles. Also, employment of traditional term vectors results in missing semantic similarity relationship by reason of vague travel features. In this research, we propose a travel recommender system based on online travel reviews to deal with vague travel features and travel preference determination. Employment of travel ontology maps terms to concept and category viewpoint vectors. Opinion quality value combining user viewpoint vector similarity and helpfulness rating interaction contributes to recommendation prediction scores. We utilize a dataset collected from Epinions.com to evaluate our approach. The experiment results show that our proposed method outperform other traditional methods.