Alexandru Todor

ChemCloud: Chemical e-Science Information Cloud

Our Chemical e-Science Information Cloud (ChemCloud) - a Semantic Web based eScience infrastructure - integrates and automates a multitude of databases, tools and services in the domain of chemistry, pharmacy and bio-chemistry available at the Fachinformationszentrum Chemie (FIZ Chemie), at the Freie Universitaet Berlin (FUB), and on the public Web. Based on the approach of the W3C Linked Open Data initiative and the W3C Semantic Web technologies for ontologies and rules it semantically links and integrates knowledge from our W3C HCLS knowledge base hosted at the FUB, our multi-domain knowledge base DBpedia (Deutschland) implemented at FUB, which is extracted from Wikipedia (De) providing a public semantic resource for chemistry, and our well-established databases at FIZ Chemie such as ChemInform for organic reaction data, InfoTherm the leading source for thermophysical data, Chemisches Zentralblatt, the complete chemistry knowledge from 1830 to 1969, and ChemgaPedia the largest and most frequented e-Learning platform for Chemistry and related sciences in German language.

Enriching Topic Models with DBpedia

Traditional Topic Modeling approaches only consider the words in the document. By using an entity-topic modeling approach and including background knowledge about the entities such as the occupation of persons, the location of organizations, the band of a musician etc., we can better cluster related documents together, and produce semantic topic models that can be represented in a knowledge base. In our approach we first reduce the text documents to a set of entities and then enrich this set with background knowledge from DBpedia. Topic modeling is performed on the enriched set of entities and various feature combinations are evaluated in order to determine the combination that achieves the best classification precision or perplexity compared to using word-based topic models alone.

Optimized Processing of Subscriptions to DBpedia Live

DBpedia Live enables access to structured data extracted from Wikipedia in real-time. A data stream that is generated from Wikipedia changes is instantly loaded in the DBpedia RDF store. Applications can benefit by subscribing to the RDF update stream and receive continuous results from DBpedia. Providing a continuous update stream of changes to subscribed DBpedia queries is a challenging task due to the load it places on the RDF store.

Human Perception of Enriched Topic Models.

Topic modeling algorithms, such as LDA, find topics, hidden structures, in document corpora in an unsupervised manner. Traditionally, applications of topic modeling over textual data use the bag-of-words model, i.e. only consider words in the documents. In our previous work we developed a framework for mining enriched topic models. We proposed a bag-of-features approach, where a document consists not only of words but also of linked named entities and their related information, such as types or categories.

Improving Semantic Search through Entity-Based Document Ranking

Traditional keyword-based IR approaches take into account the document context only in a limited manner. In our paper we present a novel document ranking approach based on the semantic relationships between named entities. In the first step we annotate all documents with named entities from a knowledge base (for example people, places and organisations). In the next step these annotations in combination with the relationships from the knowledge base are used to rank documents in order to perform a semantic search. Documents that contain the specific named entity that was searched for as well as other strongly related entities, receive a higher ranking. The inclusion of the document context in the ranking approach achieves a higher precision in the Top-K results.