Talking about Places:Considering Context for the Geolocation of Images Extracted from Tweets

Abstract

This paper investigates the extraction of geolocated images from social media. Pictures taken with a mobile device are typically georeferenced, but social media may or may not provide geo-coordinates, depending on their privacy policies. Our goal is to geolocate images extracted from Twitter to support emergency services in natural disasters. As the number of tweets with native georeferences is limited, we introduce algorithms that take advantage of various contextual clues included in social media posts to help increase the proportion of posts that can be geolocated. Using an explorative approach, we also investigate how to locate, in other social media, images that were originally embedded in tweets. The application of these context-based algorithms to a case study is discussed.

Social Media Data Processing and Analysis by Means of Machine Learning for Rapid Detection, Assessment and Mapping the Impact of Disasters

“…identifying topicality) within unstructured or semi-structured web-harvested data, unknown quality (there may be little or no relevant metadata), and difficulty integrating it with other sources (which may, in turn, have their own issues of quality and uncertainty). When using social media, biases may also be presented, for example, due to a lack of digital engagement within certain demographics of the populations of particular areas (Francalanci, C., et al, 2018, Houston, J.B., et al, 2015. So, to address this issue the following steps were taken: -algorithms for collecting and initial filtering of relevant text and photo data were developed; -an approach for extraction of features, important for damage assessment, from text information was developed; -approach for extracting useful features from user-shared photo images was designed.…”

Image-Based Social Sensing: Combining AI and the Crowd to Mine Policy-Adherence Indicators from Twitter

“…In this paper we adopt the CIME geolocation algorithm proposed in the E2mC project [26], [27], which for nongeolocated tweets extracts a possible location from the text and metadata of the post, using the Stanford Core Named Entity Extraction algorithm [28] and OpenStreetMap [29] with the Nominatim API 7 as a gazeteer and a context-based approach for disambiguation [27].…”

“…In order to evaluate the social impact of COVID-19 in different countries, it is necessary to associate a location to each post. The geolocation was performed using the CIME service [26], [27] described in Section II, applying it on the textual part of the tweet, combined with the textual user location, if present. The geolocation was not performed whenever already available from Twitter itself, in which case the original one is used.…”

Crowd4ems: A Crowdsourcing Platform for Gathering And Geolocating Social Media Content in Disaster Response

“…As a basis of this work we focus here on the description of the results of the geolocalization service called CIME (Contextbased Media Extractor) which has been developed for the project. The CIME algorithm is briefly illustrated in (Francalanci et al, 2018) and its presentation is not the focus of this work. Many services are also provided as a support to filtering relevant images, to eliminate duplicates and to identify images with a poor quality (e.g.…”

“…This paper proposes a hybrid approach, combining automatic extraction of relevant social media content and crowdsouring. Automatic extraction is based on keyword matching and allows the automatic extraction of locations with a NER (Named Entity Recognition) based and disambiguation approach called CIME (Context based IMage Recognition) (Francalanci et al, 2018). Crowdsourcing, uses Crowd4EMS 10 , a facilitating platform for coordinating volunteer contributions and creating reliable and actionable data for disaster response.…”