Spatially Aware Term Selection for Geotagging

Laere, Olivier Van; Quinn, Jonathan; Schockaert, Steven; Dhoedt, Bart

doi:10.1109/tkde.2013.42

Cited by 35 publications

(40 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, most social images still lack location information, thus driving the recent interest in location estimation for social images as an active research topic [Zhang and Kosecka, 2006, Schindler et al, 2007, Hays and Efros, 2008,Serdyukov et al, 2009,Van Laere et al, 2011,Li et al, 2013b,Van Laere et al, 2014. With the rapid growth in social images, new challenges have been raised: 1) data quality of social images is an issue because they are often associated with noise and error-prone user-generated content, such as junk comments and misspelled words; and 2) data sparsity exists in social images despite the large volume of images, since they are mostly unevenly distributed around the world and their contextual information is often missing or incomplete.…”

Section: Location Estimationmentioning

confidence: 99%

“…A two-step approach was proposed by [Van Laere et al, 2011] which first determined the most likely area for the query image based on a language model and then performed a similarity search within this area to find the most textually similar candidate for a final location estimation. They further improved the classification accuracy by spatial-aware terms selection [Van Laere et al, 2014] which reduces the negative impact of noisy tags by detecting the most location-indicative tags from the whole corpus. Another approach to improve the language model is to combine it with a geographic topic model.…”

Section: Location Estimationmentioning

confidence: 99%

“…To increase the weightings of the tags that carry geographical cues, we apply the method proposed in [Van Laere et al, 2014] to calculate the spatial-aware weighting for each tag. Given a tag t and the image set L t consisting of the images with this tag, the weight of t, denoted as s(t), is computed as: 1) where N t = |L t | is the total number of occurrences of tag t, p and q are images in L t , dist(·) is the geographic distance (e.g., Haversine distance), λ is the distance threshold, and w is the hyper-parameter controlling the weight to favor tags with occurrences centered around only a few locations.…”

Section: Spatial-aware Tag Weightingmentioning

confidence: 99%

“…To avoid overfitting on the test dataset, for parameters tuning of the proposed spatial-aware language model (SLM) and topic model (SRLSI), we randomly sample 50,000 images from the training dataset as a validation dataset, and the rest are used for training. Note that when separating the validation dataset from the training dataset, we ensure all the images from the same user are either in the training dataset, or in the validation dataset, to avoid an unfair exploitation of user-specific tags [ Van Laere et al, 2014]. This validation dataset is also used for training the listwise L2R model.…”

Section: Benchmark and Data Preprocessmentioning

confidence: 99%

“…As a compensation, large collections of georeferenced text, such as the associated tags of geotagged Flickr images, can be used to identify correlations between occurrences of terms and particular places. A vital step for these approaches is to extract spatial-aware terms [Van Laere et al, 2014] that are most indicative of geographic locations from a given text collection. Language model smoothing is required due to data sparseness.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Large-scale content management and retrieval for social media

Cao¹

View full text Add to dashboard Cite

With online social network and mobile devices flourishing, social media is becoming ubiquitous in our daily lives. However, the poor data quality and the diversity of multimedia content are major obstacles to finding desired information for the users. Furthermore, efficient retrieval becomes a non-trivial task due to the massive, and growing amount of media data. Nowadays, social media systems require data cleaning and understanding techniques for intelligent management, as well as efficient and accurate retrieval frameworks to handle the large amount of data. In this thesis, we study different content management and retrieval methodologies to address the above problems.The first part of this thesis aims to improve the data quality of social media by automatically completing the missing information. Specifically, we present a spatial-aware multimodal location estimation framework to predict the unknown location for some media data. Our method consists of multiple models which utilize various information sources, including textual and visual content of the social media.Understanding the visual content plays a crucial role in multimedia retrieval. With the help of deep learning techniques, machines are now able to recognize certain objects from images and videos. In the second part of this thesis, we design a new neural network architecture to automatically learn the discriminative visual features from the visual content of a large-scale annotated dataset. The network is able to capture the fine-grained differences among images.To tackle the large data volume and high feature dimensionality, in the third part of this thesis, we propose an efficient retrieval framework which combines the bag-of-visual-words model and deep learning features. The burstiness issue and the quantization error are revisited and addressed in our scenario. Efficient storage and fast retrieval is achieved by incorporating the inverted index into our framework. Sometimes, the initial retrieval results might not be satisfactory because of the noisy and irrelevant data. Therefore, in the last part of this thesis, we introduce a query adaptive reranking method to further refine the search result. The key idea is to suppress the negative effects of background noise when calculating the importance of result candidates. We formulate the procedure as a quadratic programming problem and solve it with an efficient optimization algorithm.Integrating the all the components in to one system, we are able to accurately predict the location of social media and as a result, we improve the data quality for better content management. The whole system is also capable to provide fast and accurate image retrieval, which is crucial for large-scale applications.iii

show abstract

Section: Location Estimationmentioning

confidence: 99%

Section: Location Estimationmentioning

confidence: 99%

Section: Spatial-aware Tag Weightingmentioning

confidence: 99%

Section: Benchmark and Data Preprocessmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Large-scale content management and retrieval for social media

Cao¹

View full text Add to dashboard Cite

show abstract

High‐level event identification in social media

Dashdorj

Altangerel

2018

Concurrency and Computation

View full text Add to dashboard Cite

Summary The increasing numbers of large data sets generated by information technologies provide a great opportunity to better understand emerging topics in human society. Retrieving real‐world events from such data, particularly free‐text data, is a complicated task in Natural Language Processing and Location‐based Social Networks. In this work, we propose a new approach, which recognizes geo‐referenced high‐level events/activities mentioned in web sources adopting open gazetteers: OpenStreetMap and Google Maps. Our approach demonstrated on sampled news articles identifies events associated with the relevant topics using a latent Dirichlet allocation. This research is an essential step towards recommendation systems, urban planning, and monitoring.

show abstract

Embedding Geographic Locations for Modelling the Natural Environment Using Flickr Tags and Structured Data

Jeawak

Jones

Schockaert

2019

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Meta-data from photo-sharing websites such as Flickr can be used to obtain rich bag-of-words descriptions of geographic locations, which have proven valuable, among others, for modelling and predicting ecological features. One important insight from previous work is that the descriptions obtained from Flickr tend to be complementary to the structured information that is available from traditional scientific resources. To better integrate these two diverse sources of information, in this paper we consider a method for learning vector space embeddings of geographic locations. We show experimentally that this method improves on existing approaches, especially in cases where structured information is available.

show abstract

Spatially Aware Term Selection for Geotagging

Cited by 35 publications

References 20 publications

Large-scale content management and retrieval for social media

Large-scale content management and retrieval for social media

High‐level event identification in social media

Embedding Geographic Locations for Modelling the Natural Environment Using Flickr Tags and Structured Data

Contact Info

Product

Resources

About