Uni-Logo
Sektionen
Sie sind hier: Startseite Projekte

Projekte

Aktuelle Forschungsprojekte der Professur

Semantic Search / Natural Language Processing

We research new methods for semantic search, using both shallow and deep natural language processing. Here are demos of some of the systems we have developed over the last years:

Contextual Sentence Decomposition (CSD and CSD-IE)

Link to demo and publications

CSD takes a given sentence and decomposes it into it's semantic unit. This is a routine of fundamental importance in search (whenever semantic proximity of words is of interest). A variant of CSD, called CSD-IE, can be used for Open Information Extraction (OIE): given a sentence, extract all subject-predidcate-object triples expressed in the sentence.

SPARQL+Text Search

Broccoli: Link to demo and publications
QLever: Link to demo and publications

Broccoli and QLever both perform combined search on a text and a knowledge, where the entities of the knowledge base are linked in the text. This provides powerful semantic search capabilities. Our first system, Broccoli, comes with a convenient user interface and search-as-you-type suggestions, but only works for tree-based queries with a one-dimensional result (list of entities). QLever is a full-featuered SPARQL Engine with the described combined text search capability. The QLever demo runs on a knowledge base with 1.9 billion triples and a text with 32.3 billion word occurrences and is still lightning fast.

Question Answering

Aqqu: Link to demo and publications
QA Completion: Link to demo

Aqqu answers arbitrary natural language questions on a knowledge base by automatically translating them to an appropriate SPARQL query. The automatic translation is learned from a training set of question-answer pairs. Our best version of Aqqu uses a deep neural network.

 

 

 

The rest of the page is out of date, sorry for that

Route Planning

The goal of this project is to enable efficient computation of journeys in public transportation networks, considering different modi of transportation (bus, train, subway, taxi, etc.), complex cost functions (including travel time, number of transfers, taxi costs) and realistic features like traffic and delay handling. A prototype for multi-modal and multi-criteria route planning for the Freiburg network can be found here.

Papers investigating certain problem aspects, like computing diverse sets of journeys or evaluating the delay robustness of route planning techniques can be found  here and here.

This project is supported by a Google Focused Research Award.

 

Research Paper Bibliographies and Management

We are developing Icecite, a research paper management system. In particular, Icecite performs automatic extraction of meta-data (title, authors, venue) and references with very high accuracy. It also features one-click download of papers in the bibliography. A prototype is available here.

We are also working on making bibliographical data more easily accessible. We maintain an instance of CompleteSearch, our powerful technology for fast search-as-you-type search, for the DBLP bibliography of computer science articles. You can find an instance here. Hannah Bast is currently speaker of the DBLP advisory board.

A paper describing the system can be found here.

 

Motion Capture (German Excellence Initiative)

This is a project with the BrainLinks BrainTools cluster of excellence. Using body sensors, we capture motion data from Parkinson patients as well as from healthy subjects. Our goal is the automatic extraction of features valuable for clinical use, in particular for diagnosis and for the calibration of devices for deep brain stimulation.

 

 

Benutzerspezifische Werkzeuge