Exploiting User and Venue Characteristics for Fine-Grained Tweet Geolocation

Abstract

Which venue is a tweet posted from? We call this a fine-grained geolocation problem. Given an observed tweet, the task is to infer its discrete posting venue, e.g., a specific restaurant. This recovers the venue context and differs from prior work, which geolocats tweets to location coordinates or cities/neighborhoods. First, we conduct empirical analysis to uncover venue and user characteristics for improving geolocation. For venues, we observe spatial homophily , in which venues near each other have more similar tweet content (i.e., text representations) compared to venues further apart. For users, we observe that they are spatially focused and more likely to visit venues near their previous visits. We also find that a substantial proportion of users post one or more geocoded tweet(s), thus providing their location history data. We then propose geolocation models that exploit spatial homophily and spatial focus characteristics plus posting time information. Our models rank candidate venues of test tweets such that the actual posting venue is ranked high. To better tune model parameters, we introduce a learning-to-rank framework. Our best model significantly outperforms state-of-the-art baselines. Furthermore, we show that tweets without any location-indicative words can be geolocated meaningfully as well.

A Survey of Location Prediction on Twitter

“…Similar to the idea of local words, they prefer geospecific n-grams, i.e., those whose tweets are mostly located in a small eclipse on the map. Alternatively, Chong and Lim [77] apply a learning to rank method which encodes tweet content by a smoothed probability estimation that a word occurs at a venue. In their following work [78], word importance for different locations is distinguished.…”

“…In experiments, they find that such a multi-indicator approach is more robust than single-indicator approaches, which is error-prone due to ambiguity. Chong and Lim [77] provide another angle to utilize the context information and observe that both venues' active time and users' visiting place histories could help on tweet location prediction. They investigate venues' active time and estimate the probability that a location is popular given a time by a smoothed kernel density estimation method.…”

Internet of Entities (IoE): A Blockchain-based Distributed Paradigm for Data Exchange between Wireless-based Devices

“…Considering that the trackers can be devices that operate, specifically, in such a context, their sensors data can be integrated to those related to a group of entities in order to create functionalities aimed to specific groups of users. This is an approach that leads towards two interesting advantages: it is able to uncover implicit characteristics of the involved entities by following non canonical criteria [17,60]; each group of entities can be anonymously characterized on the basis of the sensors data of the entities that belong to it.…”

USTAR: Online Multimodal Embedding for Modeling User-Guided Spatiotemporal Activity

“…First, most of the previous studies on learning spatiotemporal embeddings neglect Non-GeoTagged Social Media (NGTSM) records, which is a large proportion of records compared to the GTSM records in social media 2 https://mktgathpu.wordpress.com/the-social-media-past-present-future 3 https://www.internetlivestats.com/twitter-statistics/ platforms. For instance, less than 5% of postings in Twitter are geotagged [10,11]. This percentage is expected to have a downward trend over the next few years as users become increasingly concerned about their privacy, which forces social media platforms to tighten their privacy agreements 4 .…”

“…Second, social media records generally come with the user indices, by which different users can be uniquely understood in an anonymous manner (i.e., a user index is a numerical identity given to each user as depicted in Table I and it does not reveal the actual identity of the user). Studies [10,12] report that there are spatially motivated user behaviors (e.g., spatially close users produce similar textual contents and users tend to visit venues that are near to each other), which are useful to understand the dynamics of spatiotemporal units. However, such user behaviors have not been exploited to learn representations for the spatiotemporal units.…”