Philipp Sorg

An experimental comparison of explicit semantic analysis implementations for cross-language retrieval

Explicit Semantic Analysis (ESA) has been recently proposed as an approach to computing semantic relatedness between words (and indirectly also between texts) and has thus a natural application in information retrieval, showing the potential to alleviate the vocabulary mismatch problem inherent in standard Bag-of-Word models. The ESA model has been also recently extended to cross-lingual retrieval settings, which can be considered as an extreme case of the vocabulary mismatch problem. The ESA approach actually represents a class of approaches and allows for various instantiations. As our first contribution, we generalize ESA in order to clearly show the degrees of freedom it provides. Second, we propose some variants of ESA along different dimensions, testing their impact on performance on a cross-lingual mate retrieval task on two datasets (JRC-ACQUIS and Multext). Our results are interesting as a systematic investigation has been missing so far and the variations between different basic design choices are significant. We also show that the settings adopted in the original ESA implementation are reasonably good, which to our knowledge has not been demonstrated so far, but can still be significantly improved by tuning the right parameters (yielding a relative improvement on a cross-lingual mate retrieval task of between 62% (Multext) and 237% (JRC-ACQUIS) with respect to the original ESA model).

Language resources extracted from Wikipedia

Wikipedia provides an interesting amount of text for more than hundred languages. This also includes languages where no reference corpora or other linguistic resources are easily available. We have extracted background language models built from the content of Wikipedia in various languages. The models generated from Simple and English Wikipedia are compared to language models derived from other established corpora. The differences between the models in regard to term coverage, term distribution and correlation are described and discussed. We provide access to the full dataset and create visualizations of the language models that can be used exploratory. The paper describes the newly released dataset for 33 languages, and the services that we provide on top of them.

Detect'11: international workshop on DETecting and Exploiting Cultural diversiTy on the social web

Sergej Sizov WeST – Institute for Web Science and Technologies University of Koblenz-Landau, 56070 Koblenz, Germany sizov@uni-koblenz.de Stefan Siersdorfer Intelligent Access to Information L3S Research Center, 30167 Hannover, Germany siersdorfer@l3s.de Thomas Gottron WeST – Institute for Web Science and Technologies University of Koblenz-Landau, 56070 Koblenz, Germany gottron@uni-koblenz.de Philipp Sorg Institut für Angewandte Informatik und Formale Beschreibungsverfahren Karlsruhe Institute of Technology, 76133 Karlsruhe, Germany philipp.sorg@kit.edu

Combining Data-Driven and Semantic Approaches for Text Mining

While the amount of structured data published on the Web keeps growing (fostered in particular by the Linked Open Data initiative), the Web still comprises of mainly unstructured—in particular textual—content and is therefore a Web for human consumption. Thus, an important question is which techniques are most suitable to enable people to effectively access the large body of unstructured information available on the Web, whether it is semantic or not. While the hope is that semantic technologies can be combined with standard Information Retrieval approaches to enable more accurate retrieval, some researchers have argued against this view. They claim that only data-driven or inductive approaches are applicable to tasks requiring the organization of unstructured (mainly textual) data for retrieval purposes. We argue that the dichotomy between data-driven/inductive and semantic approaches is indeed a false one. We further argue that bottom-up or inductive approaches can be successfully combined with top-down or semantic approaches and illustrate this for a number of tasks such as Ontology Learning, Information Retrieval, Information Extraction and Text Mining.

Learning Methods in Multi-grained Query Answering

This PhD proposal is about the development of new methods for information access. Two new approaches are proposed: Multi-Grained Query Answering that bridges the gap between Information Retrieval and Question Answering and Learning-Enhanced Query Answering that enables the improvement of retrieval performance based on the experience of previous queries and answers.