1st International KEYSTONE Conference

IKC 2015

Coimbra Portugal, 8-9 September 2015

• Navid Rekabsaz, Ralf Bierig, Mihai Lupu and Allan Hanbury. Toward Optimized Multimodal Concept Indexing.
  Abstract: Information retrieval on the (social) web moves from a pure term-frequency-based approach to an enhanced method that includes conceptual multimodal features on a semantic level. In this paper, we present an approach for semantic-based keyword search and focus especially on its optimization to scale it to real-world sized collections in the social media domain. Furthermore, we present a faceted indexing frame- work and architecture that relates content to semantic concepts to be indexed and searched semantically. We study the use of textual concepts in a social media domain and observe a significant improvement from using a concept-based solution for keyword searching. We address the problem of time-complexity that is critical issue for concept-based methods by focusing on optimization to enable larger and more real-world style applications.
  
  
  Piotr Kozikowski, Ekaterini Ioannou, Yannis Velegrakis and Francesco Guerra. Support of Part-whole Relations in Query Answering.
  Abstract:Part-whole relations are ubiquitous in our world, yet they do not get "first-class" treatment in the data managements systems most commonly used today. One aspect of part-whole relations that is particularly important is that of attribute transitivity. Some attributes of a whole are also attributes of its parts, and vice versa. We propose an extension to a generic entity-centric data model to support part-whole relations and attribute transitivity and provide more meaningful results to certain types of queries as a result. We describe how this model can be implemented using an RDF repository and three approaches to infer the implicit information necessary for query answering that adheres to the semantics of the model. The first approach is a naive implementation and the other two use indexing to improve performance. We evaluate several aspects of our implementations in a series of experimental results that show that the two approaches that use indexing are far superior to the naive approach and exhibit some advantages and disadvantages when compared to each other.
  
  
  Yiwei Zhou, Elena Demidova and Alexandra Cristea. Analysing Entity Context in Multilingual Wikipedia to Support Entity-Centric Retrieval Applications.
  Abstract:Representation of influential entities, such as famous people and multinational corporations, on the Web can vary across languages, reflecting language-specific entity aspects as well as divergent views on these entities in different communities. A systematic analysis of language- specific entity contexts can provide a better overview of the existing aspects and support entity-centric retrieval applications over multilingual Web data. An important source of cross-lingual information about influential entities is Wikipedia – an online community-created encyclopaedia – containing more than 280 language editions. In this paper we focus on the extraction and analysis of the language-specific entity contexts from different Wikipedia language editions over multilingual data. We discuss alternative ways such contexts can be built, including graph-based and article-based contexts. Furthermore, we analyse the similarities and the differences in these contexts in a case study including 80 entities and five Wikipedia language editions.
  
  
  Sara Cadegnani, Francesco Guerra, Sergio Ilarri, Maria Del Carmen Rodriguez-Hernandez, Raquel Trillo Lado and Yannis Velegrakis. Recommending Web Pages using Item-based Collaborative Filtering Approaches
  Abstract:Predicting the next page a user wants to see in a large website has gained importance along the last decade due to the fact that the Web has become the main communication media between a wide set of entities and users. This is true in particular for institutional government and public organization websites, where for transparency reasons a lot of information has to be provided. The "long tail" phenomenon affects also this kind of websites and users need support for improving the effectiveness of their navigation. For this reason, complex models and approaches for recommending web pages that usually require to process personal user preferences have been proposed. In this paper, we propose three different approaches to leverage information embedded in the structure of web sites and their logs to improve the effectiveness of web page recommendation by considering the context of the users, i.e., their current sessions when surfing a specific web site. This proposal does not require either information about the personal preferences of the users to be stored and processed or complex structures to be created and maintained. So, it can be easily incorporated to current large websites to facilitate the users’ navigation experience. Experiments using a real-world website are described and analyzed to show the performance of the three approaches.
  
  
  Andrey Tagarev, Laura Tolosi and Vladimir Alexiev. Domain-specific modeling: Towards a Food and Drink Gazetteer.
  Abstract:Our goal is to build a Food and Drink (FD) gazetteer that can serve for classification of general, FD-related concepts, efficient faceted search or automated semantic enrichment. Fully supervised design of a domain-specific models ex novo is not scalable. Integration of several ready knowledge bases is tedious and does not ensure coverage. Completely data-driven approaches require a large amount of training data, which is not always available. In cases when the domain is not very specific (as the FD domain), re-using encyclopedic knowledge bases like Wikipedia may be a good idea. We propose here a semi-supervised approach, that uses a restricted Wikipedia as a base for the modeling, achieved by selecting a domain-relevant Wikipedia category as root for the model and all its subcategories, combined with expert and data-driven pruning of irrelevant categories.
  
  
  Michael Cochez, Vagan Terziyan and Vadim Ermolayev. Balanced Large Scale Knowledge Matching Using LSH Forest.
  Abstract:Evolving Knowledge Ecosystems were proposed recently to approach the Big Data challenge, following the hypothesis that knowledge evolves in a way similar to biological systems. Therefore, the inner working of the knowledge ecosystem can be spotted from natural evolution. An evolving knowledge ecosystem consists of Knowledge Organisms, which form a representation of the knowledge, and the environment in which they reside. The environment consists of contexts, which are composed of so-called knowledge tokens. These tokens are ontological fragments extracted from information tokens, in turn, which originate from the streams of information owing into the ecosystem. In this article we investigate the use of LSH Forest (a self-tuning indexing schema based on locality-sensitive hashing) for solving the problem of placing new knowledge tokens in the right contexts of the environment. We rst show that LSH Forest possesses required properties and could be used for large distributed set-ups. We also show the results of initial experiments with large ontologies.
  
  
  Andrea Calì, Thomas Lynch, Davide Martinenghi and Riccardo Torlone. Processing Keyword Queries under Access Limitations
  Abstract:The Deep Web is constituted by data accessible through Web pages, but not readily indexable by search engines, as they are returned in dynamic pages. In this paper we propose a framework for accessing Deep Web sources, represented as relational tables with so-called access limitations, with keyword-based queries. We formalize the notion of optimal answer and propose methods for query processing. To the best of our knowledge, ours is the first systematic approach to keyword search in such context.
  
  
  Mariem Harmassi, Daniela Grigori and Khalid Belhajjame. Mining Workflow Repositories for Improving Fragments Reuse
  Abstract:Public repositories of scientific and business workflows are gaining growing attention as a means to enable understanding, reuse and ultimately the reproducibility of the processes such workflows incarnate. However, as the number of workflows hosted by such repositories grows, their users face difficulties when it come to exploring and querying workflows. In this paper, we explore a functionality that can help repository administrators to index their workflows, and users to identify the workflows that are of interest to them. In particular, we investigate the problem of finding frequent and similar fragments in workflows using graph mining techniques. Our objective is not to come up with yet another graph mining or similarity technique. Instead, we explore different representations that can be used for encoding workflows before assessing their similarity taking into consideration the effectiveness and efficiency of the mining algorithm. Furthermore, we investigate how similarity detection can be improved by augmenting workflow specifications with knowledge extracted from external sources.
  
  
  Karol Draszawka, Julian Szymanski and Francesco Guerra. Improving css-KNN Classification Performance by Shifts in Training Data.
  Abstract:This paper presents a new approach to improve the performance of a css-k-NN classifier for categorization of text documents. The css-k-NN classifier (i.e., a threshold-based variation of a standard k-NN classifier we proposed in [1]) is a lazy-learning instance-based classifier. It does not have parameters associated with features and/or classes of objects, that would be optimized during off-line learning. In this paper we propose to introduce a training phase in the classifier. The idea is to compute training data modifications, such that class representative instances are optimized before the actual k-NN algorithm is employed. The empirical text classification experiments using mid-size Wikipedia data sets show that carefully cross-validated settings of such preprocessing yields significant improvements in k-NN performance compared to classification without this step. The proposed approach can be useful for improving the effectivenes of other classifiers as well as it can find applications in domain of recommendation systems and keyword-based search.
  
  
  Ranka Stanković, Cvetana Krstev, Ivan Obradović and Olivera Kitanović. Indexing of textual databases based on lexical resources: A case study for Serbian.
  Abstract:In this paper we describe an approach to improvement of information retrieval results for large textual databases by pre-indexing documents using bag-of-words and named entity recognition. The approach was applied on a database of geological projects financed by the Republic of Serbia in the last half century. Each document within this database is described by metadata, consisting of several fields such as title, domain, keywords, abstract, geographical location and the like. A bag of words was produced from these metadata using morphological dictionaries and transducers, and named entities within the metadata were recognized using a rule-based system. They were then used for indexing the document where ranking was based on tf-idf measure. Evaluation of ranked retrieval results based on data obtained by pre-indexing are compared to results obtained by informational retrieval without pre-indexing with Eleven Point Precision-Recall Curve, showing a significant improvement in terms of Mean Average Precision measure (MAP).
  
  
  Claudia Ifrim, Florin Pop, Mariana Mocanu and Valentin Cristea. Agile DBLP: A Search-based Mobile Application for Structured Digital Libraries
  Abstract:Internet access on mobile devices used to be thought of as something only for people on the move. Some mobile browser users are simply quickly looking for access to specific information. They may be checking their online grocery list, looking up for an address, or trying to find the best Italian restaurant within a five minute walk. The abilities of applications on native platforms stayed rather consistent for over 10 years, but in the past several years we have seen the web platform increase its ability to handle web applications, with almost the same fidelity as native applications. Expose APIs and web services are available for a wide range of data sets or services like weather, maps and locations. In this paper we propose an easy and fast alternative of mobile application development, using existing resources as web services or exposed APIs. We present the design of a search-based mobile application prototype (”Agile DBLP”) using the exposed data set of DBLP Computer Science Bibliography. We consider that this application will be useful for the scientists or for common people.
  
  
  Dror Mughaz, Yaakov Hacohen-Kerner and Dov Gabbay. Meaningful Key-Phrases as Means to Improve the Estimation of the Birth and Death Years of Jewish Authors. Abstract:In this study, we will try to find the time-frame in which the author of a given document lived. The discussed documents are rabbinic documents written in Hebrew, Aramaic and Yiddish languages. The documents are usually undated and do not contain any bibliographic section, which leaves us with an interesting challenge to find the desired time-frame .In order to do that, we define a set of keyphrases and formulate various kinds of rules: "Iron-clad", Heuristic and Greedy constraints defining the time-frame. These rules are based on Keyphrases and key-words in the documents of the authors. Identifying the time-frame of an author can help us obtain the generation in which specific documents were written, can help in the examination of forged documents, to conclude if the document was edited, and in some cases to identify an anonymous author. We tested these rules on two corpuses of documents, which authored by 12 and 24 rabbinic authors and the results are promising.
  
  
  Tarcísio Souza, Elena Demidova, Thomas Risse, Helge Holzmann, Gerhard Gossen and Julian Szymanski. Semantic URL Analytics to Support Efficient Annotation of Large Scale Web Archives
  Abstract:Long-term Web archives comprise Web documents gathered over longer time periods and can easily reach hundreds of terabytes in size. Semantic annotations such as named entities can facilitate intelligent access to the Web archive data. However, the annotation of the entire archive content on this scale is often infeasible. The most efficient way to access the documents within Web archives is provided through their URLs, which are typically stored in dedicated index files. The URLs of the archived Web documents can contain semantic information and can offer an efficient way to obtain initial semantic annotations for the archived documents. In this paper, we analyse the applicability of semantic analysis techniques such as named entity extraction to the URLs in a Web archive. We evaluate the precision of the named entity extraction from the URLs in the Popular German Web dataset and analyse the proportion of the archived URLs from 1444 popular domains in the time interval from 2000 to 2012 to which these techniques are applicable. Our results demonstrate that named entity recognition can be successfully applied to a large number of URLs in our Web archive and provide a good starting point to efficiently annotate large scale collections of Web documents.
  
  
  Yaakov Hacohen-Kerner, Asaf Sabag, Dimitris Liparas, Anastasia Moumtzidou, Stefanos Vrochidis and Ioannis Kompatsiaris. Classification using various ML Methods and Combinations of Key-Phrases and Visual Features. Abstract:In this research, we present a comparative classification study of news documents using different ML methods and different combinations of keyphrases and visual features. The keyphrases are word n-grams (excluding stopwords) extracted from the news documents. The visual features are extracted from the biggest image of each document, as it is assumed to be the representative one. The application do-main is news documents written in English that belong to four domains: Health, Lifestyle-Leisure, Science-Technology, and Nature-Environment downloaded from large number of news web-sites. Many various classification experiments have been performed with three supervised machine learning methods. The n-gram textual feature set containing 624 features obtained an accuracy result of 81.0%, which is regarded as a baseline result. This result was much better than the accuracy result (58.4%) obtained by the visual feature set containing 4000 low-level features. A competition between three supervised machine learning methods, a feature filtering method, and parameter tuning led to an improved ac-curacy result of 86.3% achieved by the Random Forest method containing 800 trees, and a combination of 59 n-gram textual features and 31 visual features.
  
  
  Miran Pobar and Marina Ivasic-Kos. Multimodal Image Retrieval Based on Keywords and Low-level Image Features.
  Abstract:Image retrieval approaches dealing with the complex problem of image search and retrieval in very large image datasets proposed so far can be roughly divided into those that use text descriptions of images (text-based image retrieval) and those that compare visual image content (content-based image retrieval). Both approaches have their strengths and drawbacks especially in the case of searching for images in general unconstrained domain. To take advantage of both approaches, we propose a multimodal framework that uses both keywords and visual properties of images. Keywords are used to determine the semantics of the query while the example image presents the visual impression (perceptual and structural information) that retrieved images should suit. In the paper, the overview of the proposed multimodal image retrieval framework is presented. For computing the content-based similarity between images different feature sets and metrics were tested. The procedure is described with Corel and Flickr images from the domain of outdoor scenes.
  
  

• José R.R. Viqueira, Alberto Bugarín, Joaquín Trinanes and Jaime Martínez-Urtaza. Keyword-Based Search over Environmental Datasets
• Gjorgji Madjarov and Sanda Sanda Martincic-Ipsic. Extracting keywords from images: bag-of-visual- words enriched with graph techniques
• Dragan Ivanovic and Marjan Čeh. Automatic Georeferencing and Search on Structured Georeferenced Bibliographic Data
• Elena Demidova, Sergej Zerr, Julian Szymanski and Karol Draszawka. Efficient Entity Annotation for Lage Scale Web Archives
• Georgia Kapitsaki and John Breslin. Searching over mobile user activity data
• Dragan Ivanovic and Georgia Kapitsaki. Personalization of Keyword-based Search on Structured Data Sources
• Márius Šajgalík, Michal Barla, Maria Bielikova and Julian Szymański. Connecting Words and Linked Data Concepts by Latent Features

• Manolis Wallace and Thanos Triantos. Visualization of Uncertain Information in Tag Clouds
• Andrea Calì, Leonardo Coaccioli, Mirko Michele Dimartino, Riccardo Frosini and Federico Pastori. Semantic Keyword Search in Linked Data

• strong accept 3
• accept 2
• weak accept 1
• borderline 0
• weak reject -1
• reject -2
• strong reject -3

• All papers sent to the Challenge Track since the average score of all of them was greater than 0 (borderline)
• All papers sent to the In-use Track since the average score for all of them was greater than -0.33
• 13 papers with average score greater than 0 (borderline) as FULL papers in the Research Track (acceptance rate 62%)
• 2 papers with average score equal to 0 (borderline) as SHORT papers in the Research Track