近年來由於適地性社群網路的興盛，越來越多使用者在這些平台與好友分享打卡資訊跟生活點滴。其中興趣點推薦系統是適地性社群網路的核心服務之一。目前這類研究主要分析使用者序列性的打卡行為，從資料中探勘使用者移動行位的偏好，有些學者會考慮基於內容的推薦方法，從豐富的興趣點資訊中擷取出興趣點的表達項，可是上述提出的方法均沒有考量到時間與事件這兩項因素，我們認為這兩種因素都會影響使用者拜訪興趣點的意願。例如一個平常不運動的使用者會因為一個演唱會的契機而在體育館打卡，如果系統只考慮使用者對於興趣點的固有偏好，則會忽略了這種情境。
本研究的目的是建立一個結合即時活動偵測的興趣點推薦系統。我們提出了一個即時性的興趣點嵌入模型，除了跟一般的興趣點嵌入模型一樣考量興趣點的固定的資訊以外，我們的模型會額外分析地點上的打卡文章，擷取出即時資訊的精華。一方面運用卷積神經網路從興趣點的固定資訊學習固定的特徵向量表達。另一方面，我們會運用多模態嵌入方法，將地點、時間及文字的資訊同時嵌入到同一個向量空間，其好處是讓關鍵詞的表達項具備語意、空間及時間上的相關性，利於從巨量的興趣點動態文章中分析不同時間上的變化，從而偵測事件的發生，並擷取當中的隱含的更動資訊。最後，我們把即時性的興趣點嵌入模型結合矩陣分解，建置一套適時性的推薦系統。
我們把提出的系統實驗於Twitter的實際資料，內容包含一些使用者在期間內拜訪紐約市的動態文章，並加入Foursquare提供的興趣點資訊輔助找出興趣點的特徵向量。實驗結果證明加入即時活動資訊於興趣度推薦系統後，其系統推薦效果會明顯提高。

Recent years have witnessed the rapid growth of population in location-based social networks (LBSNs) where they are allowed to check-in and share daily lives with others. Point-of-Interest (POI) recommendation is one of the core services in LBSNs. Previous works focus on modeling the sequential patterns of check-ins to learn users’ preference. Some consider content-based method which learns POI representations given the introduction of them. However, those models do not consider factors such as time and events. For instance, a user who hates sports but go visiting a stadium because of a concert. Such kind of situation cannot be reflected in solely considering users’ preference for POIs.
In this study, we propose a real-time POI embedding model that, instead of capturing intrinsic information, is capable to mine real-time information of places into the latent representations according to the correspond geo-tagged posts. On one hand, we used a CNN on mining textual information of POIs to their intrinsic representation. On the other hand, a multimodal embedding model of location, time and text is applied to keep monitoring posts on POIs and extracts a set of features for representing events or burst information that may attract users. Furthermore, we combine that real-time POI embedding with matrix factorization method to form a more comprehensive POI recommendation.
To verify the effectiveness of our proposed method, we conduct experiments on Twitter dataset with geo-tagged tweets in NYC as location-based recommendation and enrich POI information with Foursquare. Experimental results show that POI recommendation system with taking real-time information into consideration can strongly improve the performance than the one without.

[1] S.Zhao, T.Zhao, I.King, andM. R.Lyu, “GT-SEER: Geo-Temporal SEquential Embedding Rank for Point-of-interest Recommendation,” in WWW, 2016, pp. 153–162.
[2] C.Cheng, H.Yang, M. R.Lyu, andI.King, “Where You Like to Go Next : Successive Point-of-Interest Recommendation,” in IJCAI, 2013, pp. 2605–2611.
[3] J.Zhang, C.-Y.Chow, andY.Li, “LORE: Exploiting Sequential Influence for Location Recommendations,” in SIGSPATIAL, 2014, pp. 103–112.
[4] B.Liu andH.Xiong, “Point-of-Interest Recommendation in Location Based Social Networks with Topic and Location Awareness,” in SDM, 2013, pp. 396–404.
[5] F.Wang, Y.Qu, L.Zheng, C. T.Lu, andP. S.Yu, “Deep and Broad Learning on Content-Aware POI Recommendation,” in CIC, 2017, vol. 2017–Janua, pp. 369–378.
[6] D.Bokde, S.Girase, andD.Mukhopadhyay, “Role of Matrix Factorization Model in Collaborative Filtering Algorithm: A Survey,” IJAFRC, vol. 1, no. 6, 2014.
[7] T.Bogers andA.Van DenBosch, “Collaborative and content-based filtering for item recommendation on social bookmarking websites,” in CEUR Workshop Proceedings, 2009, vol. 532, pp. 9–16.
[8] M.Ye, P.Yin, W.-C.Lee, andD.-L.Lee, “Exploiting geographical influence for collaborative point-of-interest recommendation,” in SIGIR, 2011, p. 325.
[9] Q.Yuan, G.Cong, Z.Ma, A.Sun, andN. M.-Thalmann, “Time-aware point-of-interest recommendation,” in SIGIR, 2013, p. 363.
[10] D.Lian, C.Zhao, X.Xie, G.Sun, E.Chen, andY.Rui, “GeoMF : Joint Geographical Modeling and Matrix Factorization for Point-of-Interest Recommendation,” KDD, pp. 831–840, 2014.
[11] B.Liu, Y.Fu, Z.Yao, andH.Xiong, “Learning geographical preferences for point-of-interest recommendation,” in KDD, 2013, p. 1043.
[12] Y.Liu, W.Wei, A.Sun, andC.Miao, “Exploiting Geographical Neighborhood Characteristics for Location Recommendation,” in CIKM, 2014, pp. 739–748.
[13] Y.Ding andX.Li, “Time weight collaborative filtering,” in CIKM, 2005, p. 485.
[14] Y.Koren, “Collaborative filtering with temporal dynamics,” in KDD, 2009, p. 447.
[15] T.Lee andY.Park, “A time-based approach to effective recommender systems using implicit feedback,” Expert Syst. Appl., vol. 34, pp. 3055–3062, 2008.
[16] M.Ye, K.Janowicz, C.Mülligann, andW.Lee, “What you are is When you are: The Temporal Dimension of Feature Types in Location-based Social Networks,” in GIS, 2011, p. 102.
[17] Z.Cheng, J.Caverlee, K.Lee, andD. Z.Sui, “Exploring Millions of Footprints in Location Sharing Services,” ICWSM, vol. 2010, no. Cholera, pp. 81–88, 2011.
[18] H.Gao, J.Tang, X.Hu, andH.Liu, “Exploring temporal effects for location recommendation on location-based social networks,” in RecSys, 2013, pp. 93–100.
[19] H.Yin, W.Wang, H.Wang, L.Chen, andX.Zhou, “Spatial-Aware Hierarchical Collaborative Deep Learning for POI Recommendation,” in TKDE, 2017, vol. 29, no. 11, pp. 2537–2551.
[20] B.Hawelka, I.Sitko, E.Beinat, S.Sobolevsky, P.Kazakopoulos, andC.Ratti, “Geo-located Twitter as the proxy for global mobility patterns,” Cartogr. Geogr. Inf. Sci., vol. 41, no. 3, pp. 260–271, Nov.2013.
[21] C.Zhang et al., “GeoBurst: Real-Time Local Event Detection in Geo-Tagged Tweet Streams,” in SIGIR, 2016, pp. 513–522.
[22] C.Zhang et al., “TrioVecEvent: Embedding-Based Online Local Event Detection in Geo-Tagged Tweet Streams,” in KDD, 2017, pp. 595–604.
[23] A.Ritter, S.Clark, Mausam, andO.Etzioni, “Named entity recognition in tweets: an experimental study,” in EMNLP, 2011, pp. 1524–1534.
[24] T.Mikolov, G.Corrado, K.Chen, andJ.Dean, “Efficient Estimation of Word Representations in Vector Space,” in ICLR, 2013, pp. 1–12.
[25] Y.Kim, “Convolutional Neural Networks for Sentence Classification,” EMNLP, pp. 1746–1751, Aug.2014.
[26] A.Morandi et al., “Rectified Linear Units Improve Restricted Boltzmann Machines,” in ICML, 2010, no. 3, pp. 807–814.
[27] R.Pan et al., “One-class collaborative filtering,” in ICDM, 2008, pp. 502–511.
[28] Y.Koren, R.Bell, andC.Volinsky, “Matrix factorization techniques for recommender systems,” Computer (Long. Beach. Calif)., vol. 42, no. 8, pp. 30–37, 2009.
[29] N.Srivastava, G.Hinton, A.Krizhevsky, I.Sutskever, andR.Salakhutdinov, “Dropout: A Simple Way to Prevent Neural Networks from Overfitting,” J. Mach. Learn. Res., vol. 15, pp. 1929–1958, 2014.