PEU-RNN: 基於遞歸神經網路構築之連續興趣點推薦系統

Abstract

近年來由於智慧型手機的興盛,人們的生活習慣正逐漸改變,使用者常常在社群 網路上分享自己所到訪的景點或是餐廳,藉此與朋友進行互動,這類的行為模式儼 然已成為生活中不可或缺的一部分。然而,隨著社群網路打卡資料的累積,興趣點 (Point-of-interest, POI) 推薦系統的研究也跟著流行起來,透過歷史打卡紀錄來分析使 用者的興趣偏好,進而提供使用者更好的到訪體驗,或者是讓商家預測消費族群,得 到更好的廣告效益。在本論文中,我們採用遞歸神經網路 (Recurrent Neural Netowrk) 來進行連續興趣點推薦,將連續打卡偏好、使用者偏好與地理因素整合以構成我們的 推薦系統。我們的實驗資料來源為 Gowalla ,是著名的適地性社群網路(Location-based social networks, LBSNs)服務。從實驗結果顯示,我們的推薦系統比現有的連續興趣點 推薦系統更加準確。

With the rapid growth of mobile devices, living habits of human are gradually changing. Users share their own visiting experiences to point-of-interests (POI) on locationbased social network (LBSNs) by checking in, which becomes a new way to interact with friends and also an important part of our life. Additionally, the great increasing of social check-in data makes the POI recommendation research popular. By analyzing users’ behavior and history check-in preferences, we can not only improve users’ visiting experience and also predict the future visitors. In this paper, we propose the PEU-RNN model, a successive point-of-interest recommendation with RNN, which can incorperates sequential prefereces and user preferences. Result on Gowalla dataset shows that our methods are more accurate that the state-of-the-art successive point-of-interest recommendation methods.