1. Zheda Road 38 Hangzhou China 310027
2. Zheda Road 38 hangzhou China 310027
Dynamically localizing users in online social networks by their posted microblogs has been an interesting research topic and attracted great attention. It is challenging since people seldom post location-related microblogs due to privacy concern. A promising approach is to leverage microblogs posted by the target user's social friends to increase inference accuracy. However, it is difficult to do so because: firstly, microblogs from user and his friends are usually not synchronized; and secondly not all microblogs from friends are helpful for localization. To address these issues, in this paper, we propose GotU, a protocol that leveraGes sOcial Ties for efficient User localization in online social networks. GotU is based on a simple observation that users' statistical locations (e.g., residential cities) can greatly improve the dynamic localization accuracy and they do not need synchronized microblogs from social friends for inference. Specifically, GotU is composed of two steps: i) statistical localization and ii) fine-grained localization. Statistical localization focuses on identifying the most likely cities that the target users would stay. To efficiently utilize social ties information, we propose the notation of co-location friend and its detection algorithm, which is able to filter out the friends geographically far from the target users. With this, we design an efficient algorithm to increase inference accuracy of statistical localization. In fine-grained localization, we estimate the dynamic locations of the target users by extracting and matching the point of interest (POI) names in the target users' microblogs and design an algorithm to eliminate POI ambiguity based on statistical location. Experiments on real world data collected from Twitter demonstrates that our approach can achieve much higher accuracy than previous studies.
Copyright 2020 Science China Press Co., Ltd. 《中国科学》杂志社有限责任公司 版权所有