Renato Domínguez García

Extended explicit semantic analysis for calculating semantic relatedness of web resources

Finding semantically similar documents is a common task in Recommender Systems. Explicit Semantic Analysis (ESA) is an approach to calculate semantic relatedness between terms or documents based on similarities to documents of a reference corpus. Here, usually Wikipedia is applied as reference corpus. We propose enhancements to ESA (called Extended Explicit Semantic Analysis) that make use of further semantic properties of Wikipedia like article link structure and categorization, thus utilizing the additional semantic information that is included in Wikipedia. We show how we apply this approach to recommendation of web resource fragments in a resource-based learning scenario for self-directed, on-task learning with web resources.

CROKODIL: a platform for collaborative resource-based learning

On-the-job learning is primarily a personal knowledge acquisition process accomplished increasingly based on resources found on the Web. These days, collaboratively learning from and with others on the Web is taking on a very prominent position in this learning process. CROKODIL aims to provide support for collaborative learning based on web resources. In this paper, we introduce our learning scenario and an evaluation of our target group. We describe our pedagogical concepts, and present the results of an evaluation of these concepts. CROKODIL supports the semantic tagging of resources as well as the collaborative use of these resources and their information. Social networking functionalities are integrated in the platform to encourage and support collaborative learning. We also present some extensions to the base functionality of the platform, such as resource recommendations and interfaces for the integration in existing learning management systems.

Exploiting semantic information for graph-based recommendations of learning resources

Recommender systems in e-learning have different goals as compared to those in other domains. This brings about new requirements such as the need for techniques that recommend learning resources beyond their similarity. It is therefore an ongoing challenge to develop recommender systems considering the particularities of e-learning scenarios like CROKODIL. CROKODIL is a platform supporting the collaborative acquisition and management of learning resources. It supports collaborative semantic tagging thereby forming a folksonomy. Research shows that additional semantic information in extended folksonomies can be used to enhance graph-based recommendations. In this paper, CROKODILs folksonomy is analysed, focusing on its hierarchical activity structure. Activities help learners structure their tasks and learning goals. AScore and AInheritScore are proposed approaches for recommending learning resources by exploiting the additional semantic information gained from activity structures. Results show that this additional semantic information is beneficial for recommending learning resources in an application scenario like CROKODIL.

Automatic taxonomy extraction in different languages using wikipedia and minimal language-specific information

Knowledge bases extracted from Wikipedia are particularly useful for various NLP and Semantic Web applications due to their co- verage, actuality and multilingualism. This has led to many approaches for automatic knowledge base extraction from Wikipedia. Most of these approaches rely on the English Wikipedia as it is the largest Wikipedia version. However, each Wikipedia version contains socio-cultural knowledge, i.e. knowledge with relevance for a specific culture or language. In this work, we describe a method for extracting a large set of hyponymy relations from the Wikipedia category system that can be used to acquire taxonomies in multiple languages. More specifically, we describe a set of 20 features that can be used for for Hyponymy Detection without using additional language-specific corpora. Finally, we evaluate our approach on Wikipedia in five different languages and compare the results with the WordNet taxonomy and a multilingual approach based on interwiki links of the Wikipedia.

Towards language-independent web genre detection

The term web genre denotes the type of a given web resource, in contrast to the topic of its content. In this research, we focus on recognizing the web genres blog, wiki and forum. We present a set of features that exploit the hierarchical structure of the web pages HTML mark-up and thus, in contrast to related approaches, do not depend on a linguistic analysis of the pages content. Our results show that it is possible to achieve a very good accuracy for a fully language independent detection of structured web genres.

FReSET: an evaluation framework for folksonomy-based recommender systems

FReSET is a new recommender systems evaluation framework aiming to support research on folksonomy-based recommender systems. It provides interfaces for the implementation of folksonomy-based recommender systems and supports the consistent and reproducible offline evaluations on historical data. Unlike other recommender systems framework projects, the emphasis here is on providing a flexible framework allowing users to implement their own folksonomy-based recommender algorithms and pre-processing filtering methods rather than just providing a collection of collaborative filtering implementations. FReSET includes a graphical interface for result visualization and different cross-validation implementations to complement the basic functionality.

Context determines content: an approach to resource recommendation in folksonomies

By means of tagging in social bookmarking applications, so called folksonomies emerge collaboratively. Folksonomies have shown to contain information that is beneficial for resource recommendation. However, as folksonomies are not designed to support recommendation tasks, there are drawbacks of the various recommendation techniques. Graph-based recommendation in folksonomies for example suffers from the problem of concept drift. Vector space based recommendation approaches in folksonomies suffer from sparseness of available data. In this paper, we propose the flexible framework VSScore which incorporates context-specific information into the recommendation process to tackle these issues. Additionally, as an alternative to the evaluation methodology LeavePostOut we propose an adaptation LeaveRTOut for resource recommendation in folksonomies. In a subset of resource recommendation tasks evaluated, the proposed recommendation framework VSScore performs significantly more effective than the baseline algorithm FolkRank.

Supporting resource-based learning on the web using automatically extracted large-scale taxonomies from multiple wikipedia versions

CROKODIL is a platform for the support of collaborative resource-based learning with Web resources. It enables the building of learning communities in which learners annotate their relevant resources using tags. In this paper, we propose the use of automatically generated large-scale taxonomies in different languages to cope with two challenges in CROKODIL: The multilingualism of the resources, i.e. web resources are in different languages and the connectivity of the semantic network, i.e. learners do not tag resources on the same topic with identical tags. More specifically, we describe a set of features that can be used for detecting hyponymy relations from the category system of Wikipedia.

Automatic acquisition of taxonomies in different languages from multiple Wikipedia versions

In the last years, the vision of the Semantic Web has led to many approaches that aim to automatically derive knowledge bases from Wikipedia. These approaches rely mostly on the English Wikipedia as it is the largest Wikipedia version and have lead to valuable knowledge bases. However, each Wikipedia version contains socio-cultural knowledge, i.e. knowledge with specific relevance for a culture or language. One difficulty of the application of existing approaches to multiple Wikipedia versions is the use of additional corpora. In this paper, we describe the adaptation of existing heuristics that make the extraction of large sets of hyponymy relations from multiple Wikipedia versions with little information about each language possible. Further, we evaluate our approach with Wikipedia versions in four different languages and compare results with GermaNet for German and WordNet for English.

CROKODIL: a platform supporting the collaborative management of web resources for learning purposes

CROKODIL is an ongoing project at the Darmstadt University of Technology. The aim of the project is to implement a platform for collaborative knowledge acquisition based on web resources. In this paper, we analyze according to a social search model, how CROKODIL provides support for all stages of the search process which is an important and integrated part in todays learning process.