Our work brings together research expertise in information retrieval (IR), natural language processing and data science. We focus on core IR topics such as collaborative search, conversational search and data-hungry ranking models. We also focus on the application of IR to various domains, most prominently the educational domain and finance technology.
Web search engines are today considered to be the primary tool to assist and empower learners in finding information relevant to their learning goals - be it learning something new, improving their existing skills, or just fulfilling a curiosity. Search engines though are not optimized for human learning; instead they are optimized for the retrieval of information. What users do with the information or how they use it during their learning is not considered. In this research line we focus on exactly that issue: how to design search engine components (on the interface level or the ranking level) that are optimized for human learning. One concrete example of this research line is our work on instructional scaffolding: a traditional learning support strategy that has been studied in the traditional educational context but not incorporated as an element in search systems.
Today's web search engines are designed for single-user search. Over the years though, research efforts have shown that complex information needs (that are explorative, open-ended and multi-faceted) can be answered more efficiently and effectively when searching in collaboration. Collaborative search (and sensemaking) research is thus concerned with techniques, algorithms and interface affordances to gain insights and improve the collaborative search process. As standard web search engines (open-sour or commercial) are only designed for single-user search, we are not only developing algorithms for collaborative search but also invest in the implementation of an open-source collaborative search system.
Conversational search is concerned with creating agents that fulfill an information need by means of a mixed-initiative conversation through natural language interaction, rather than the turn-taking models exhibited in a standard search engine. It is an active area of research due to the widespread deployment of voice-based agents, such as Google Assistant and Microsoft Cortana. Voice-based agents are currently mostly used for simple closed domain tasks such as fact checking, initiating calls and checking the weather. They are not yet effective for conducting open domain complex and exploratory information seeking conversations. In this research line we investigate UI affordances (think about voice-only search for instance) and algorithmic approaches to improve conversational search such that complex information needs can be resolved via a conversation between a user and an agent.
Neural approach to information retrieval (BERT and Co) have become the new standard approach to retrieval algorithms. In this research line we investigate some of the issues that neural approaches have in terms of their runtime efficiency (huge models typically also mean slow inference), their training efficiency as well as their explainability. As one example, we have recently worked on the analysis of neural models for ranking based on an axiomatic approach. The axiomatic approach to has shown that the effectiveness of a retrieval method is connected to its fulfillment of axioms. This approach enabled researchers to identify shortcomings in existing approaches and ''fix'' them. With the new wave of neural net based approaches to IR, a theoretical analysis of those retrieval models is no longer feasible, as they potentially contain millions of parameters. We have proposed a pipeline to create diagnostic datasets for IR, each engineered to fulfill one axiom. This allows us to empirically determine to what extent neural models for IR are able to fulfill those common sense axioms (and this knowledge in turn helps us to identify shortcomings).
FinTech is short for finance technology. This application domain is concerned with the use of technology to improve financial services and processes. In our research we consider the use of NLP for the fully automatic extraction of information from different types of complex documents such as regulatory documents. We also investigate how to improve the workflow for employees of such businesses by improving their ability to query complex databases through a natural language interface (instead of a formal query language).