Article Preview
TopIntroduction
A recommendation system (RS) is an online information service system that suggests appropriate items to users (Kulkarni and Rodd 2020). Generally, RSs employ personalization to improve the effectiveness and efficiency of recommendations. They have been widely used in current online platforms of multiple areas, such as e-commerce (e.g., Amazon), entertainment (e.g., YouTube), and social media (e.g., Weibo)(Liu and Du 2020). In the recommendation process, scholars mainly focus on developing algorithms to learn users’ preferences and then to find relevant items to users (Kulkarni and Rodd 2020, Zhang and Song 2021). Especially with the development of artificial intelligence (e.g., deep learning techniques), enhanced information retrieval and users' preferences learning have made RS more effective, so that it can provide more relevant items to users (Zhang et al. 2019, Wu et al. 2021). However, some scholars have found that the overpersonalization resulting from the interest-relevancy recommendation may lead to the failure of recommendation (Cheng et al. 2017, Niu et al. 2018). Therefore, content-diversity is developed to solve this problem (Hou, Pan and Liu 2018, Wu et al. 2020, Szpektor, Maarek and Pelleg 2013). The trade-off between interest-relevancy and content-diversity has also become a classical problem in which attracted many scholars have studied (Javed et al. 2021, Niu et al. 2018, Hou et al. 2018, Panteli and Boutsinas 2021, Smyth and Mcclave 2001, Bag, Ghadge and Tiwari 2019).
With the continuous development of the mobile network, many RSs have integrated mobile techniques with the traditional recommendation methods to improve recommendation effectiveness based on spatial contexts(Wedel and Kannan 2016). And the new system is called the location-based context-aware RS (CARS)(Savage et al. 2012). News feed apps in China, such as Toutiao and Baidu News Feed have all adopted the new system to enhance their service. As for studies on location-based CARS, the context-relevancy method is obviously mostly concerned. It refers to utilizing contexts to improve the quality of the RS, which involves context acquisition, integration, modeling, reasoning, and dissemination (Suhaim and Berri 2021, Nawara and Kashef 2021, Gao et al. 2019, Sojahrood and Taleai 2021). Among them, context integration has attracted much attention from scholars.
Although algorithms that contain contextual information have been developed to meet users’ contextual demands (Setiowati, Adji and Ardiyanto 2018), it is contended that users tend to avoid location-congruent contents because of the privacy concern, sometimes, which may reduce user satisfaction(Kulkarni and Rodd 2020, Mou, Cui and Kurcz 2020). However, few studies have investigated when context-relevancy will cause negative consequences and validate the role of privacy concern. This question is necessary to be explored because CARS can improve user satisfaction by identifying the condition of users' perception of privacy threat before applying the privacy protection mechanism(Cheng et al. 2017). In addition, given that the privacy threat may be caused by the overpersonalization and oversensitivity of context-relevancy(Bradbury, Jhumka and Leeke 2018), the content-diversity method may also be a potential solution to the privacy issue, which was nonetheless less discussed before (Chen et al. 2021).