Below is the list of papers that, following the reviewers' recommendations, have been accepted for presentation at the 10th International Workshop on Semantic and Social Media Adaptation and Personalization. Congratulations to all authors and sincere thanks to the program committee.
Martin Lopez-Nores, Yolanda Blanco-Fernandez, Jose J. Pazos-Arias, Manuel Ramos-Cabrer and Alberto Gil-Solla. Augmented Reality, Smart Codes and Cloud Computing for Personalized Interactive Advertising on Billboards
Abstract: We present a system that enables m-commerce interactions with billboards through augmented reality and 2D smart codes. Specifically, extra information about the advertisements are embedded within the codes that are placed on the billboards, so that a mobile application can scan and decode them in order to support connectionless location-aware advertising experiences. In case of connectivity, the system relies on a server that offers cloud-based personalization functionalities in order to provide the users with tailor-made interactive spots that are automatically composed, customized according to their devices and locations, and personalized as per their particular preferences and needs.
Philip Healy, Graham Hunt, Steven Kilroy, Theo Lynn, John Morrison and Shankar Venkatagiri. Evaluation of Peak Detection Algorithms for Social Media Event Detection
Abstract: We evaluate the effectiveness of three peak detection algorithms when applied to collection of social media datasets. Each dataset is composed of a year's worth of tweets relating to a topic. The datasets were converted to time series composed of hourly tweet volumes. The objective of the analysis was to identify abnormal surges of communication, which are taken to be representative of the occurrence of events relevant to the topic under consideration. The ground truth was established by manually tagging the time series in order to identify peaks apparent to a human operator. Candidate algorithms were then evaluated in terms of the precision, recall, and F1 scores obtained when their output was compared to the manually identified peaks. A general-purpose algorithm is found to perform reasonably well, but seasonality in social media data limits the effectiveness of applying simple algorithms without filtering.
Jakub Simko and Maria Bielikova. Gaze-tracked Crowdsourcing
Abstract: When creating intelligent systems, we often need proper knowledge bases and resources annotated with metadata. Sometimes, we have no other option, than to utilize crowdsourcing, to acquire the data in necessary quantity. Crowdsourcing is a costly endeavor, always pressing for improvements in task solving quantity and quality. Studies show that consideration of implicit feedback (behavior of workers during task solving) helps to improve the overall crowd output. Gaze-tracking is a powerful source of implicit feedback as it records user's activity outside typical feedback channels (e.g. clicking, scrolling, typing) and reveals a great deal of cognitive processes. The paper argues that gaze-tracking will present a potent feedback source even for crowdsourcing, as it's costs are steadily dropping and the technology becomes available for wide worker pools. An example case study demonstrates the use of the gaze-tracking during a typical crowdsourcing task -- acquisition of training dataset for automated word sense disambiguation. Normally in such task, the worker explicitly selects a corresponding sense for a given word located in a text snippet and thus contributes to the dataset. With gaze-tracking involved, she also shares us other information useful for dataset enrichment: worker's reading pattern (which may indicate his confidence) and important sense-distinguishing words (e.g. words from snippet which triggered worker's decision).
Dong Zhou, Séamus Lawless, Jianxun Liu, Sanrong Zhang and Yu Xu. Query Expansion for Personalized Cross-Language Information Retrieval
Abstract: Cross-language information retrieval research has favored system-centered approaches in the past. The user is not an integral part of the translation and retrieval processes. In this paper, we investigate the problem of personalized cross-language information retrieval by exploiting query expansion techniques. The original query is augmented with terms mined from the user’s historical usage information in one language, with the aim of retrieving more relevant results in another language. Experiments semi-automatically constructed by using bilingual Wikipedia documents showed that in general personalized approaches work better than non-personalized approaches. We also found that an individual user model generated from one language can be used to enhance the personalized cross-language information retrieval.
Dimitris Kollias, George Marandianos, Amaryllis Raouzaiou and Andreas Stafylopatis. Interweaving Deep Learning and Semantic Techniques for Emotion Analysis in Human-Machine Interaction
Abstract: This paper presents a new data classification approach which is based on the one hand on deep learning neural networks for effectively extracting well defined categorical information from data and on the other hand on an adaptable support vector machine, which appropriately represents existing related knowledge about user and context specific data. The proposed approach is implemented and successfully tested experimentally for emotion analysis in human machine interaction.
Georgios Stratogiannis, Aggeliki Vlachostergiou, Georgios Siolas, George Caridakis, Phivos Mylonas, Andreas Stafylopatis and Stefanos Kollias. User and home appliances pervasive interaction in a sensor driven Smart Home environment: the SandS approach
Abstract: EU FIRE research project “Social and Smart” aims to formalize and build a complete ecosystem of users, context sensors and smart home appliances that interact following the ubiquitous computing paradigm in order to adapt and enhance the everyday user-appliance interaction. In this framework a user is modeled through the use of Personas stereotypes that are related through fuzzy relationships. Contextual information is collected via wireless ambient sensors, such as temperature and humidity ones, but can also include Smart City sensors and services. This contextual information is further related to each user’s model through the enforcement of home rules, expressed in a high level language. Knowledge representation is supported through Semantic Web technologies that also ensure the interoperability between all the actors of the ecosystem. Preliminary experimental results have been carried in a small scale Smart Home setting, but also in a larger scale using the FIWARE1 framework provided by the SmartSandander testbed.
Chiara Bernabei, Francesco Guerra and Raquel Trillo Lado. Keyword Search in structured data and Network Analysis: a preliminary experiment over DBLP
Abstract: Identifying similar items to the ones provided as input to a search system, is a challenging task. The main issues concern not only the management of large collections of data, but also the profiling of the users, who usually have different opinions, tastes and expertise. In this paper we make a preliminary investigation about the improvements in the accuracy of a search system provided by network analysis techniques supporting the discovery of relations among the items stored in the repository. For this reason, we have developed the SEEN prototype, a keyword search tool exploiting network analysis. SEEN has been evaluated against a relational version of the DBLP repository. The results of the preliminary experiments show that the the information provided by networks can improve the effectiveness of the results.
Evaggelos Spyrou, Ioannis Sofianos and Phivos Mylonas. Mining Tourist Routes from Flickr Photos
Abstract: Popular social networking sites like Flickr are nowadays overwhelmed by geo-tagged photos. Semi-automatic discovery of touristic routes and landmarks from this pool of photos form a challenging task. In this paper we attempt to semantically analyze user-generated routes within downtown city areas defined around a pre-selected geographical area and derived from a large geo-tagged Flickr dataset, by utilizing a novel two-level clustering scheme. Our goal is to select the semantically interesting routes for a given area of interest. Without loss of generality the latter is considered to be a predefined "window" around a city's most famous landmarks and touristic attractions. The herein proposed framework has been applied to a 100K geo-tagged Flickr photos dataset deriving from a major European metropolis, Athens, Greece and quantitatively evaluated against several state-of-the-art works.
Panos Alexopoulos and Manolis Wallace. Creating Domain-Specific Semantic Lexicons for Aspect-Based Sentiment Analysis
Abstract: Aspect based sentiment analysis (ABSA) is an opinion mining process where texts are analyzed to extract the sentiments that their authors express towards certain features and characteristics of particular entities, such as products or persons. Key role in the effectiveness of this process plays the accurate and complete identification of the entities’ discussed aspects within the text, as well as of the evaluation expressions that accompany these aspects. Nevertheless, what entities may be considered as aspects and what evaluation expressions may characterize them, depends largely on the domain at hand. With that in mind, in this paper we propose an approach for representing and populating semantic lexicons that contain domain-specific aspect-evaluation-polarity relations and, as such, can be (re-)used towards more effective ABSA in concrete domains and scenarios.
Gerasimos Razis, Ioannis Anagnostopoulos and Michalis Vafopoulos. Semantic Social Analytics and Linked Open Data Cloud
Abstract: In this paper, we propose an ontology schema towards linking semantified Twitter social analytics with the Linked Open Data cloud. The ontology is deployed over a publicly available service that measures how influential a Twitter account is by combining its social activity in Twitter. According to our knowledge this is the first work that combines social analytics with the Linked Open Data (LOD).
Ales Masiar and Jakub Simko. Short Video Metadata Acquisition Game
Abstract: Gathering proper descriptive metadata for multimedia resources is nowadays essential for effective information processing, recommendation and personalization. And, we still need to employ human workforce in crowdsourcing scenarios for solving particular metadata acquisition tasks. In this paper we present a human computation game, which acquires metadata for short videos found in the Vine social media service. Player’s task is to watch a given short video (without any further description) and formulate a search query, which will make that video appear in top results of the search. The player queries are then processed for keywords characterizing the videos. Our experiments show, that these keywords are both correct and from the great part, represent new information atop existing video descriptions.
Manolis Wallace and Nikos Platis. The uncertain tag cloud
Abstract: Tag clouds provide an excellent means of visualization of weighted semantic information. When, on the other hand, this information is not definitive but is rather accompanied by a measurable degree of uncertainty, conventional tag clouds are no longer suitable visualization tools. In this paper we extend the conventional approach to tag cloud generation and propose the utilization of the degree of opaqueness as a means to visualize the degree of certainty. In order to experimentally assess the efficacy of the proposed approach we have developed the corresponding software tools and have applied the conventional and proposed approached to tag visualization in a real life scenario of probabilistic data.
Dimitris Sgouropoulos, Evaggelos Spyrou, Giorgos Siantikos and Theodoros Giannakopoulos. Counting and Tracking People in a Smart Room: an IoT Approach
Abstract: In this paper we present an approach for counting and tracking people within a meeting room scenario. The sensing and processing modules are incorporated within the context of an IoT framework, that follows a message oriented architecture. The proposed algorithm consists of a motion detection module, a background subtraction module, a people counting module and a tracking module, while its output is primarily used as input to a decision making module that controls the room's environment. We show that we may achieve satisfactory results, using simple low resolution cameras. We evaluate our method using a publicly available, real world data set. Experimental results indicate the effectiveness of our approach.