Can we look at individual search sessions and identify users who are similar in terms of search goals and behaviors? One of the potential goals of this research would be to identify users who are similar and predict their search outcomes (success or failure). Another goal is to make suggestions on what information could be potentially useful for the user (information fostering).
Information search is no longer limited to typing queries in the search box. Query-less searches, common in voice-based assistants, make use of natural language conversations between the user and the system. With increased knowledge about the capabilities of the system and the requirements of the user, the system understands how to improve the search results. This is a result of increased personalization and better relevance assessments. This project will aim to improve the dialogue or conversations which occur between the user and the system during an information gathering session.
Automatic reformulation of questions in Community Question Answering (CQA) sites in an effort to improve the quality of questions. It is common for CQA sites to discard poor-quality questions. The goal of this research is to design an approach to automatically reformulate such questions, with or without any feedback from the users, to match the quality requirements of the site. The system is to be trained using questions which are similar and have been frequently upvoted and answered in the community.
Collecting and identifying micro-actions that comprise information seeking sessions, and then recombining them in order to create novel sessions and potentially improved outcomes. These actions comprise larger information behaviors such as browsing, but can be arranged into different patterns thus creating unique and previously unidentified behaviors. This project is based on the research into digital infinity and the possibility for a small set of ordered items to create near infinite possibilities.
Generating summaries in multiple formats including tables, narratives, abstracts, and other visualizations from a collection of documents, with a particular focus on government and financial documents. The goal is to take advantage of underlying textual structures, such as formatting and tagging, as well as implied structures, such as time series, hierarchical relationships and other dependencies.
Generating automated multimedia answers in the form of text, appropriate image, videos from the Community Question Answering (CQA) sites, with a focus on answer quality. The goal is to design an algorithm which will take account of the question type and its intended audience to generate automated answers. The source of the answers can be some combination of existing answers provided to previously asked similar questions, or some other resources depending on the context.
NeuIR (pronounced “New IR”) Group is focused on solving problems at the intersection of Deep Neural Network, Big Data, and Multimedia Information Retrieval. We understand the importance of different forms of data (text, speech, images, sensor data), and strive to achieve the following research goals:
(a) Exploring different deep learning techniques in an effort to advance artificial intelligence in information retrieval;
(b) Identifying challenges in handling big data and developing best practices to solve these problems;
(c) Developing newer retrieval models and evaluation metrics specific to conversational information retrieval.