meta data for this page

Recent Research Topics (Only selected ones)

Open Domain Question Answering

Automatic question answering has received increasing attention recently in NLP community. Open domain question answering assumes providing answers to natural language questions which are extracted from large text collections. We focus in particular on answering questions from news article collections that span many years or decades. This requires putting special attention on temporal aspects of text such as document timestamp and embedded temporal expressions in content of documents.

Automatic Timeline Generation

Timeline summarization generates timelines from news collections for arbitrary queries. For example, one can generate timeline of covid-19 related events in Austria. In this research we investigate generation of multiple timelines at the same time from underlying document collections rather than single timelines as it has been done so far. In addition, we have also conducted researches on contrastive timeline generation and financial document timeline generation.

Novel Access Methods to Large News Collections

Access to longitudinal news collections requires novel user-friendly solutions as average users tend to lack contextual information needed to understand the news stories. We propose ranking news articles by their contemporary relevance which is the degree to which the news relate to present, as well as we provide methods for the estimation of news interestingness. Furthermore, we propose various approaches for computing temporal analogy.

Text Readability Estimation

Many users have problems with understanding complex texts, especially, ones that require knowledge of specialized domain (e.g., law, medicine) or high cognitive processing skills. We propose approaches for automatically estimating text difficulty levels. An example can be seen in the recently released online service for estimating the readability of text in German: Ablesbarkeitsmesser. One of the other application examples for this research can be readability-aware search engine which outputs not only relevant documents but also ones that are easy-to-read and understandable for difficult queries or topics.

Temporal Text Analysis & Understanding

This research assumes generating diverse temporal estimations for textual content. The examples of such estimator computation are publication date prediction, text focus date detection, or analysis of semantic evolution of words in past texts. Another range of temporal estimators deal with deep understanding of events expressed in text, e.g., estimation of event dates, or prediction of action continuity and validity (temporal natural language inference).

Domain-specific Business Applications

This research focuses on adaptation and application of Natural Language Processing, Information Retrieval and Knowledge Extraction methods to diverse business areas including insurance, economics and law. Examples are financial news summarization and timeline generation, risk assessment for insurance, international relation prediction or judicial support for conducting trial debates by judge's question generation.