NeuroTPR: A neuro‐net toponym recognition model for extracting locations from social media messages

Social media messages, such as tweets, are frequently used by people during natural disasters to share real‐time information and to report incidents. Within these messages, geographic locations are often described. Accurate recognition and geolocation of these locations are critical for reaching those in need. This article focuses on the first part of this process, namely recognizing locations from social media messages. While general named entity recognition tools are often used to recognize locations, their performance is limited due to the various language irregularities associated with social media text, such as informal sentence structures, inconsistent letter cases, name abbreviations, and misspellings. We present NeuroTPR, which is a Neuro‐net ToPonym Recognition model designed specifically with these linguistic irregularities in mind. Our approach extends a general bidirectional recurrent neural network model with a number of features designed to address the task of location recognition in social media messages. We also propose an automatic workflow for generating annotated data sets from Wikipedia articles for training toponym recognition models. We demonstrate NeuroTPR by applying it to three test data sets, including a Twitter data set from Hurricane Harvey, and comparing its performance with those of six baseline models.     

A spatial‐temporal‐semantic approach for detecting local events using geo‐social media data

Social media networks allow users to post what they are involved in with location information in a real‐time manner. It is therefore possible to collect large amounts of information related to local events from existing social networks. Mining this abundant information can feed users and organizations with situational awareness to make responsive plans for ongoing events. Despite the fact that a number of studies have been conducted to detect local events using social media data, the event content is not efficiently summarized and/or the correlation between abnormal neighboring regions is not investigated. This article presents a spatial‐temporal‐semantic approach to local event detection using geo‐social media data. Geographical regularities are first measured to extract spatio‐temporal outliers, of which the corresponding tweet content is automatically summarized using the topic modeling method. The correlation between outliers is subsequently examined by investigating their spatial adjacency and semantic similarity. A case study on the 2014 Toronto International Film Festival (TIFF) is conducted using Twitter data to evaluate our approach. This reveals that up to 87% of the events detected are correctly identified compared with the official TIFF schedule. This work is beneficial for authorities to keep track of urban dynamics and helps build smart cities by providing new ways of detecting what is happening in them.     

Assessing relevance of tweets for risk communication

ABSTRACT

Although Twitter is used for emergency management activities, the relevance of tweets during a hazard event is still open to debate. In this study, six different computational (i.e. Natural Language Processing) and spatiotemporal analytical approaches were implemented to assess the relevance of risk information extracted from tweets obtained during the 2013 Colorado flood event. Primarily, tweets containing information about the flooding events and its impacts were analysed. Examination of the relationships between tweet volume and its content with precipitation amount, damage extent, and official reports revealed that relevant tweets provided information about the event and its impacts rather than any other risk information that public expects to receive via alert messages. However, only 14% of the geo-tagged tweets and only 0.06% of the total fire hose tweets were found to be relevant to the event. By providing insight into the quality of social media data and its usefulness to emergency management activities, this study contributes to the literature on quality of big data. Future research in this area would focus on assessing the reliability of relevant tweets for disaster related situational awareness.     

A novel approach to leveraging social media for rapid flood mapping: a case study of the 2015 South Carolina floods

Rapid flood mapping is critical for local authorities and emergency responders to identify areas in need of immediate attention. However, traditional data collection practices such as remote sensing and field surveying often fail to offer timely information during or right after a flooding event. Social media such as Twitter have emerged as a new data source for disaster management and flood mapping. Using the 2015 South Carolina floods as the study case, this paper introduces a novel approach to mapping the flood in near real time by leveraging Twitter data in geospatial processes. Specifically, in this study, we first analyzed the spatiotemporal patterns of flood-related tweets using quantitative methods to better understand how Twitter activity is related to flood phenomena. Then, a kernel-based flood mapping model was developed to map the flooding possibility for the study area based on the water height points derived from tweets and stream gauges. The identified patterns of Twitter activity were used to assign the weights of flood model parameters. The feasibility and accuracy of the model was evaluated by comparing the model output with official inundation maps. Results show that the proposed approach could provide a consistent and comparable estimation of the flood situation in near real time, which is essential for improving the situational awareness during a flooding event to support decision-making.     

Visualizing spatiotemporal trajectories of mobile social media users using space–time cube

The implementation of social network applications on mobile platforms has significantly elevated the activity of mobile social networking. Mobile social networking offers a channel for recording an individual’s spatiotemporal behaviors when location-detecting capabilities of devices are enabled. It also facilitates the study of time geography on an individual level, which has previously suffered from a scarcity of georeferenced movement data. In this paper, we report on the use of georeferenced tweets to display and analyze the spatiotemporal patterns of daily user trajectories. For georeferenced tweets having both location information in longitude and latitude values and recorded creation time, we apply a space–time cube approach for visualization. Compared to the traditional methodologies for time geography studies such as the travel diary-based approach, the analytics using social media data present challenges broadly associated with those of Big Data, including the characteristics of high velocity, large volume, and heterogeneity. For this study, a batch processing system has been developed for extracting spatiotemporal information from each tweet and then creating trajectories of each individual mobile Twitter user. Using social media data in time geographic research has the benefits of study area flexibility, continuous observation and non-involvement with contributors. For example, during every 30-minute cycle, we collected tweets created by about 50,000 Twitter users living in a geographic region covering New York City to Washington, DC. Each tweet can indicate the exact location of its creator when the tweet was posted. Thus, the linked tweets show a Twitter users’ movement trajectory in space and time. This study explores using data intensive computing for processing Twitter data to generate spatiotemporal information that can recreate the space–time trajectories of their creators.     

Modeling Use of Space from Social Media Data Using a Biased Random Walker

Individuals and other entities move through space as a function of local characteristics of place, their internal behavioral models, and the topological structure of the underlying space. When a collection of locations (i.e. geotagged photos or other geotagged social media information) from a large number of individuals is assembled, it becomes possible to understand the interrelationship between the individuals and the space they occupy. This research systematically considers this interrelationship through an examination of the effect of the intersection of behavioral and spatial characteristics on individuals moving on street networks. The research illustrates how social media data, in combination with a biased random walker, can be used to understand and model the interaction of spatial structure and social‐environmental factors on influencing individuals' use of their environment. The biased walker offers a flexible approach to incorporate consideration of both social‐environmental and structural factors into a model and we demonstrate this through a case study wherein we are able to use the random walker to model the characteristics of Flickr users in New York City.     

#Earthquake: Twitter as a Distributed Sensor System

Social media feeds are rapidly emerging as a novel avenue for the contribution and dissemination of information that is often geographic. Their content often includes references to events occurring at, or affecting specific locations. Within this article we analyze the spatial and temporal characteristics of the twitter feed activity responding to a 5.8 magnitude earthquake which occurred on the East Coast of the United States (US) on August 23, 2011. We argue that these feeds represent a hybrid form of a sensor system that allows for the identification and localization of the impact area of the event. By contrasting this with comparable content collected through the dedicated crowdsourcing ‘Did You Feel It?’ (DYFI) website of the U.S. Geological Survey we assess the potential of the use of harvested social media content for event monitoring. The experiments support the notion that people act as sensors to give us comparable results in a timely manner, and can complement other sources of data to enhance our situational awareness and improve our understanding and response to such events.     

Detecting events from the social media through exemplar-enhanced supervised learning

ABSTRACT

Understanding and detecting the intended meaning in social media is challenging because social media messages contain varieties of noise and chaos that are irrelevant to the themes of interests. For example, conventional supervised classification approaches would produce inconsistent solutions to detecting and clarifying whether any given Twitter message is really about a wildfire event. Consequently, a renovated workflow was designed and implemented. The workflow consists of four sequential procedures: (1) Apply the latent semantic analysis and cosine similarity calculation to examine the similarity between Twitter messages; (2) Apply Affinity Propagation to identify exemplars of Twitter messages; (3) Apply the cosine similarity calculation again to automatically match the exemplars to known training results, and (4) Apply accumulative exemplars to classify Twitter messages using a support vector machine approach. The overall correction ratio was over 90% when a series of ongoing and historical wildfire events were examined.     

Chinese toponym recognition with variant neural structures from social media messages based on BERT methods

Journal of Geographical Systems - Many natural language tasks related to geographic information retrieval (GIR) require toponym recognition, and identifying Chinese toponyms from social media...     

Big location‐based social media messages from China's Sina Weibo network: Collection,storage, visualization,and potential ways of analysis

China's social media platform, Sina Weibo, like Twitter, hosts a considerable amount of big data: messages, comments, pictures. Collecting and analyzing information from this treasury of human behavior data is a challenge, although the message exchange on the network is readable by everyone through the web or app interface. The official Application Programming Interface (API) is the gateway to access and download public content from Sina Weibo and is used to collect messages for all mainland China. The nearby_timeline() request is used to harvest only messages with associated location information. This technical note serves as a reference for researchers who do not speak Mandarin but want to collect data from this rich source of information. Ways of data visualization are presented as a point cloud, density per areal unit, or clustered using Density‐Based Spatial Clustering of Applications with Noise (DBSCAN). The relation of messages to census information is also given.     

A spatial analysis of non‐English Twitter activity in Houston,TX

The use of social media data in geographic studies has become common, yet the question of social media's validity in such contexts is often overlooked. Social media data suffers from a variety of biases and limitations; nevertheless, with a proper understanding of the drawbacks, these data can be powerful. As cities seek to become “smarter,” they can potentially use social media data to creatively address the needs of their most vulnerable groups, such as ethnic minorities. However, questions remain unanswered regarding who uses these social networking platforms, how people use these platforms, and how representative social media data is of users' everyday lives. Using several forms of regression, I explore the relationships between a conventional data source (the U.S. Census) and a subset of Twitter data potentially representative of minority groups: tweets created by users with an account language other than English. A considerable amount of non‐stationarity is uncovered, which should serve as a warning against sweeping statements regarding the demographics of users and where people prefer to post. Further, I find that precisely located Twitter data informs us more about the digital status of places and less about users' day‐to‐day travel patterns.     

Geo‐parsing Messages from Microtext

Widespread use of social media during crises has become commonplace, as shown by the volume of messages during the Haiti earthquake of 2010 and Japan tsunami of 2011. Location mentions are particularly important in disaster messages as they can show emergency responders where problems have occurred. This article explores the sorts of locations that occur in disaster‐related social messages, how well off‐the‐shelf software identifies those locations, and what is needed to improve automated location identification, called geo‐parsing. To do this, we have sampled Twitter messages from the February 2011 earthquake in Christchurch, Canterbury, New Zealand. We annotated locations in messages manually to make a gold standard by which to measure locations identified by a Named Entity Recognition software. The Stanford NER software found some locations that were proper nouns, but did not identify locations that were not capitalized, local streets and buildings, or non‐standard place abbreviations and mis‐spellings that are plentiful in microtext. We review how these problems might be solved in software research, and model a readable crisis map that shows crisis location clusters via enlarged place labels.     

Identifying disaster related social media for rapid response: a visual-textual fused CNN architecture

ABSTRACT

In recent years, social media platforms have played a critical role in mitigation for a wide range of disasters. The highly up-to-date social responses and vast spatial coverage from millions of citizen sensors enable a timely and comprehensive disaster investigation. However, automatic retrieval of on-topic social media posts, especially considering both of their visual and textual information, remains a challenge. This paper presents an automatic approach to labeling on-topic social media posts using visual-textual fused features. Two convolutional neural networks (CNNs), Inception-V3 CNN and word embedded CNN, are applied to extract visual and textual features respectively from social media posts. Well-trained on our training sets, the extracted visual and textual features are further concatenated to form a fused feature to feed the final classification process. The results suggest that both CNNs perform remarkably well in learning visual and textual features. The fused feature proves that additional visual feature leads to more robustness compared with the situation where only textual feature is used. The on-topic posts, classified by their texts and pictures automatically, represent timely disaster documentation during an event. Coupling with rich spatial contexts when geotagged, social media could greatly aid in a variety of disaster mitigation approaches.     

Mining Trajectory Data and Geotagged Data in Social Media for Road Map Inference

As mapping is costly and labor‐intensive work, government mapping agencies are less and less willing to absorb these costs. In order to reduce the updating cycle and cost, researchers have started to use user generated content (UGC) for updating road maps; however, the existing methods either rely heavily on manual labor or cannot extract enough information for road maps. In view of the above problems, this article proposes a UGC‐based automatic road map inference method. In this method, data mining techniques and natural language processing tools are applied to trajectory data and geotagged data in social media to extract not only spatial information – the location of the road network – but also attribute information – road class and road name – in an effort to create a complete road map. A case study using floating car data, collected by the National Commercial Vehicle Monitoring Platform of China, and geotagged text data from Flickr and Google Maps/Earth, validates the effectiveness of this method in inferring road maps.     

Space,time, and situational awareness in natural hazards: a case study of Hurricane Sandy with social media data

ABSTRACT

Various methods have been developed to investigate the geospatial information, temporal component, and message content in disaster-related social media data to enrich human-centric information for situational awareness. However, few studies have simultaneously analyzed these three dimensions (i.e. space, time, and content). With an attempt to bring a space–time perspective into situational awareness, this study develops a novel approach to integrate space, time, and content dimensions in social media data and enable a space–time analysis of detailed social responses to a natural disaster. Using Markov transition probability matrix and location quotient, we analyzed the Hurricane Sandy tweets in New York City and explored how people’s conversational topics changed across space and over time. Our approach offers potential to facilitate efficient policy/decision-making and rapid response in mitigations of damages caused by natural disasters.     

Spatiotemporal analysis of photo contribution patterns to Panoramio and Flickr

ABSTRACT

Photo-sharing services provide a rich resource of crowdsourced spatial data consisting of georeferenced imagery and metadata. Shared photos can provide valuable information for a variety of applications and geospatial analysis tasks, such as identifying tourist hot spots or traveled routes. Understanding the spatiotemporal patterns of photo contributions will allow analysts to assess the suitability of these data for related analysis tasks. Using California as a study area, this paper analyzes various aspects of photo contribution patterns of Panoramio and Flickr. It identifies areas where annual photo contributions are still growing and areas that undergo a decline in annual contributions. Multiple regression is used to identify which environmental correlates are associated with an increase in photo-sharing activities. Furthermore, panel data of annual contributions between 2006 and 2013 for California subcounties will be used in a regression model to demonstrate that there is a positive feedback effect between Panoramio and Flickr photo contributions, but no neighborhood effect. The results of this paper provide insight into the data quality of crowdsourced image collections. These collections are commonly used for geospatial applications, including tourist information services and the computation of scenic routes.     

Modeling Remote Sensing Satellite Collection Opportunity Likelihood for Hurricane Disaster Response

State and local agencies involved in emergency response to natural disasters such as hurricanes have explicitly indicated they need imagery covering the disaster area within three days of the event; and more desirably within 24 hours of the event. Airborne image collections have often been used but suffer from several problems, most noticeably the collection time (days or week) required for larger areas. The use of remote sensing satellites carrying high spatial resolution sensors has often been touted as the logical response for rapidly collecting post-disaster event imagery for emergency response. Unfortunately, satellites are maintained on fixed orbits. The repeat interval for remote sensing satellites carrying high spatial resolution sensors, even with pointable sensors, is on the order of several days, depending on the latitude for the disaster event. Fortunately, more than one satellite carries high spatial resolution imagery. This combination of requirements and restrictions may result in either a relatively high (or low) likelihood of collecting imagery within the three-day window of opportunity. This research investigated the likelihood of collecting imagery over a hurricane disaster area based on the orbital cycles of three high spatial resolution imaging satellites. Using the spatial-temporal distribution of historic hurricane landfall locations as a proxy for the probability distribution of future hurricanes by latitude, the "visibility" of each landfall location to future satellite imaging opportunities was determined. The results indicate that the likelihood of collecting imagery within one day of the event varied between 17 and 39 percent by relying on one satellite image provider. However, if either of three satellite imagery sources (i.e., Ikonos-2, Quickbird-2, and Orbview-3) could be used, then the likelihood increased to 61 percent. By relying on three satellite imagery providers there is a likelihood of between 94 and 100 percent of collecting imagery within two or three days, respectively, after the event.     

Inferring the Location of Twitter Messages Based on User Relationships

User interaction in social networks, such as Twitter and Facebook, is increasingly becoming a source of useful information on daily events. The online monitoring of short messages posted in such networks often provides insight on the repercussions of events of several different natures, such as (in the recent past) the earthquake and tsunami in Japan, the royal wedding in Britain and the death of Osama bin Laden. Studying the origins and the propagation of messages regarding such topics helps social scientists in their quest for improving the current understanding of human relationships and interactions. However, the actual location associated to a tweet or to a Facebook message can be rather uncertain. Some tweets are posted with an automatically determined location (from an IP address), or with a user‐informed location, both in text form, usually the name of a city. We observe that most Twitter users opt not to publish their location, and many do so in a cryptic way, mentioning non‐existing places or providing less specific place names (such as “Brazil”). In this article, we focus on the problem of enriching the location of tweets using alternative data, particularly the social relationships between Twitter users. Our strategy involves recursively expanding the network of locatable users using following‐follower relationships. Verification is achieved using cross‐validation techniques, in which the location of a fraction of the users with known locations is used to determine the location of the others, thus allowing us to compare the actual location to the inferred one and verify the quality of the estimation. With an estimate of the precision of the method, it can then be applied to locationless tweets. Our intention is to infer the location of as many users as possible, in order to increase the number of tweets that can be used in spatial analyses of social phenomena. The article demonstrates the feasibility of our approach using a dataset comprising tweets that mention keywords related to dengue fever, increasing by 45% the number of locatable tweets.     

Exploring Geomorphometry through User Generated Content: Comparing an Unsupervised Geomorphometric Classification with Terms Attached to Georeferenced Images in Great Britain

User generated content such as the georeferenced images and their associated tags found in Flickr provides us with opportunities to explore how the world is described in the non‐scientific, everyday language used by contributors. Geomorphometry, the quantitative study of landforms, provides methods to classify Digital Elevation Models (DEMs) according to attributes such as slope and convexity. In this article we compare the terms used in Flickr and Geograph in Great Britian to describe georeferenced images to a quantitative, unsupervised classification of a DEM, using a well established method, and explore the variation of terms across geomorphometric classes and space. Anthropogenic terms are primarily associated with more gentle slopes, while terms which refer to objects such as mountains and waterfalls are typical of steeper slopes. Terms vary both across and within classes, and the source of the user generated content has an influence on the type of term used with Geograph, a collection which aims to document the geography of Great Britain, dominated by features which might be observed on a map.