Viktor Slavkovikj

Using topic models for Twitter hashtag recommendation

Since the introduction of microblogging services, there has been a continuous growth of short-text social networking on the Internet. With the generation of large amounts of microposts, there is a need for effective categorization and search of the data. Twitter, one of the largest microblogging sites, allows users to make use of hashtags to categorize their posts. However, the majority of tweets do not contain tags, which hinders the quality of the search results. In this paper, we propose a novel method for unsupervised and content-based hashtag recommendation for tweets. Our approach relies on Latent Dirichlet Allocation (LDA) to model the underlying topic assignment of language classified tweets. The advantage of our approach is the use of a topic distribution to recommend general hashtags.

Hyperspectral Image Classification with Convolutional Neural Networks

Hyperspectral image (HSI) classification is one of the most widely used methods for scene analysis from hyperspectral imagery. In the past, many different engineered features have been proposed for the HSI classification problem. In this paper, however, we propose a feature learning approach for hyperspectral image classification based on convolutional neural networks (CNNs). The proposed CNN model is able to learn structured features, roughly resembling different spectral band-pass filters, directly from the hyperspectral input data. Our experimental results, conducted on a commonly-used remote sensing hyperspectral dataset, show that the proposed method provides classification results that are among the state-of-the-art, without using any prior knowledge or engineered features.

Unsupervised spectral sub-feature learning for hyperspectral image classification

ABSTRACT Spectral pixel classification is one of the principal techniques used in hyperspectral image (HSI) analysis. In this article, we propose an unsupervised feature learning method for classification of hyperspectral images. The proposed method learns a dictionary of sub-feature basis representations from the spectral domain, which allows effective use of the correlated spectral data. The learned dictionary is then used in encoding convolutional samples from the hyperspectral input pixels to an expanded but sparse feature space. Expanded hyperspectral feature representations enable linear separation between object classes present in an image. To evaluate the proposed method, we performed experiments on several commonly used HSI data sets acquired at different locations and by different sensors. Our experimental results show that the proposed method outperforms other pixel-wise classification methods that make use of unsupervised feature extraction approaches. Additionally, even though our approach does not use any prior knowledge, or labelled training data to learn features, it yields either advantageous, or comparable, results in terms of classification accuracy with respect to recent semi-supervised methods.

Automatic GEO-MASHUP generation of outdoor activities

In this paper, we describe a novel approach for the automatic generation of a GEO-MASHUP related to a user his outdoor activities. Each mashup consists of online geotagged media resources related to the geographic keypoints where the outdoor activity was performed. In order to detect candidate keypoints, we search low-activity locations based on the travelling distance over time. Subsequently, we filter out route-specific keypoints (such as traffic lights) using online trajectory information. Finally, the remaining keypoints are fed to a set of social media web services to retrieve the geotagged media which summarizes the users activity. The GEO-MASHUP demonstrator, which is evaluated in real-world conditions, shows the feasibility of our novel approach.

Image-Based Road Type Classification

The ability to automatically determine the road type from sensor data is of great significance for automatic annotation of routes and autonomous navigation of robots and vehicles. In this paper, we present a novel algorithm for content-based road type classification from images. The proposed method learns discriminative features from training data in an unsupervised manner, thus not requiring domain-specific feature engineering. This is an advantage over related road surface classification algorithms which are only able to make a distinction between pre-specified uniform terrains. In order to evaluate the proposed approach, we have constructed a challenging road image dataset of 20,000 samples from real-world road images in the paved and unpaved road classes. Experimental results on this dataset show that the proposed algorithm can achieve state-of-the-art performance in road type classification.

Map-Based Linking of Geographic User and Content Profiles for Hyperlocal Content Recommendation

In this paper we describe a novel approach for map-based linking of users with content (and vice versa) based on their geographic profiles. The proposed technique facilitates hyperlocal content recommendation targeted to the user’s geographic footprint. The generation of the geographic user profiles (GUP) is based on user-logged activity analysis. The result of this analysis is a heat map of the geographic keypoints where the outdoor activities were performed. For the geographic content profiling (GCP), we use the available geotags and perform address geocoding and geographic named entity recognition to extract additional locations from the media objects. In order to link the GCP to the GUP, and to be able to recommend the hyperlocal content that fits the user’s current profile, heat map analysis is performed using geographic analyzing tools. The GEOprofiling demonstrator, which is evaluated on real activity profiles and different media types, shows the feasibility of the proposed approach.

Automatic geographic enrichment by multi-modal bike sensing

This paper focuses on the automatic geo-annotation of road/terrain types by collaborative bike sensing. The proposed terrain classification system is mainly based on the analysis of volunteered geographic information gathered by cyclists. By using participatory accelerometer and GPS sensor data collected from the cyclists’ smartphones, which is enriched with image data from geographic web services or the smartphone camera, the proposed system is able to distinguish between 6 different terrain types. For the classification of the multi-modal bike data, the system employs a random decision forest (RDF), which compared favorably for the geo-annotation task against different classification algorithms. The system classifies the features of every instance of road (over a 5 seconds interval) and maps the results onto the corresponding GPS coordinates. Finally, based on all the collected instances, we can annotate geographic maps with the terrain types, create more advanced route statistics and facilitate geo-based recommender systems. The accuracy of the bike sensing system is 92 % for 6-class terrain classification. For the 2-class on-road/off-road classification an accuracy of 97 % is achieved, almost six percent above the state-of-the-art in this domain.

Automatic Geographic Enrichment by Multi-modal Bike Sensing

This paper focuses on the automatic geo-annotation of road/terrain types by collaborative bike sensing. The proposed terrain classification system is mainly based on the analysis of volunteered geographic information gathered by cyclists. By using participatory accelerometer and GPS sensor data collected from the cyclists’ smartphones, which is enriched with image data from geographic web services or the smartphone camera, the proposed system is able to distinguish between 6 different terrain types. For the classification of the multi-modal bike data, the system employs a random decision forest (RDF), which compared favorably for the geo-annotation task against different classification algorithms. The system classifies the features of every instance of road (over a 5 seconds interval) and maps the results onto the corresponding GPS coordinates. Finally, based on all the collected instances, we can annotate geographic maps with the terrain types, create more advanced route statistics and facilitate geo-based recommender systems. The accuracy of the bike sensing system is 92 % for 6-class terrain classification. For the 2-class on-road/off-road classification an accuracy of 97 % is achieved, almost six percent above the state-of-the-art in this domain.

Visitor-art interaction by motion path detection

This paper describes a method for video-based motion path detection which is applied in the creation of an interactive artwork. The proposed algorithm, based on the Hough transform, detects parametric motion trajectories in real-time (10 fps). In order to detect peoples motion under nonstatic background object occlusion we have also developed a video segmentation technique. The proposed interaction system adopts top-down camera view to extract spatiotemporal motion trajectories and discern predefined patterns of movement thus enabling the creation of new artistic choreographies. We present test results that illustrate the effectiveness of our method and discuss the practical applicability of our approach in other domains.