Ilija Subašić

Discovery of interactive graphs for understanding and searching time-indexed corpora

Rich information spaces (like the Web or scientific publications) are full of “stories”: sets of statements that evolve over time, manifested as, for example, collections of news articles reporting events that relate to an evolving crime investigation, sets of news articles and blog posts accompanying the development of a political election campaign, or sequences of scientific papers on a topic. In this paper, we formulate the problem of discovering such stories as Evolutionary Theme Pattern Discovery, Summary and Exploration (ETP3). We propose a method and a visualisation tool for solving ETP3 by understanding, searching and interacting with such stories and their underlying documents. In contrast to existing approaches, our method concentrates on relational information and on local patterns rather than on the occurrence of individual concepts and global models. In addition, it relies on interactive graphs rather than natural language as the abstracted story representations. Furthermore, we present an evaluation framework. Two real-life case studies are used to illustrate and evaluate the method and tool.

STORIES in Time: A Graph-Based Interface for News Tracking and Discovery

We present the STORIES methods and tool for (a) learning an abstracted story representation from a collection of time-indexed documents; (b) visualising it in a way that encourages users to interact and explore in order to discover temporal “story stages” depending on their interests; and (c) supporting the search for documents and facts that pertain to the user-constructed story stages. In addition, we give an overview of evaluation studies of the tool.

Story graphs: Tracking document set evolution using dynamic graphs

With the growing number of document sets accessible online, tracking their evolution over time story tracking became an increasingly interesting problem. In this paper we propose a story tracking method based on the dynamics of keyword-association graphs. We create a graph representation of the story evolution that we call story graphs, and investigate how graph structure can be used for detecting and discovering new developments in the story. First we investigate the possibly interesting graph properties for development detection. We continue by investigating how graph structure can be linked to the sentences representing developments. For this we create an evaluation framework which bridges the gap between temporal text mining patterns and sentences. We apply this framework to evaluate our method against other temporal text mining methods. Our experiments show that story graphs perform at similar levels overall, but provide distinctive advantages in some settings.

Investigating query bursts in a web search engine

The Internet has become for many the most important medium for staying informed about current news events. Some events cause heightened interest on a topic, which in turn yields a higher frequency of the search queries related to it. These queries are going through a “query burst”. In this paper we examine the behavior of search engine users during a query burst, compared to before and after the burst. We are interested in how this behavior changes, and how it affects other stake-holders in web search.We analyze one year of web-search and news-search logs, looking at query bursts from multiple perspectives. First, we adopt the perspective of search engine users, describing changes in their effort and interest while searching. Second, we adopt the perspective of news providers by comparing web search and news search query bursts. Third, we look at the burst from the perspective of content providers.We study the conditions under which content providers can “ride” a wave of increased interest on a topic, and obtain a share of the users increased attention. We do so by identifying the class of queries that can be considered as an opportunity for content providers that are “late-comers” for a query, in the sense of not being among the first to write about its topic. We also present a simple model for predicting the click share content providers could obtain if they decide to provide content about these queries.

Sailing the corpus sea: Visual exploration of news stories

Rich information spaces like blogs or news are full of “stories”: sets of statements that evolve over time, made in fast-growing streams of documents. Even if one reads a specific source every day and/or subscribes to a selection of feeds, one may easily lose track; in addition, it is difficult to reconstruct a story already in the past. In this paper, we present the STORIES methods and tool for (a) learning an abstracted story representation from a collection of time-indexed documents; (b) visualizing it in a way that encourages users to interact and explore in order to discover temporal “story stages” depending on their interests; (c) supporting the search for documents and facts that pertain to the user-constructed story stages; (d) discovering the most important facts in the corpora; and (e) navigating in document space along multiple meaningful dimensions of document similarity and relatedness. This combination provides users with more control, progressing from “surfing” the Web to “sailing” selected corpora of it, semantically in story space as well as between the underlying documents. An evaluation demonstrates that machine learning and interaction lead to representations that serve to retrieve coherent and relevant document subsets and that help users learn facts about the story.

How to Carve up the World: Learning and Collaboration for Structure Recommendation

Structuring is one of the fundamental activities needed to understand data. Human structuring activity lies behind many of the datasets found on the internet that contain grouped instances, such as file or email folders, tags and bookmarks, ontologies and linked data. Understanding the dynamics of large-scale structuring activities is a key prerequisite for theories of individual behaviour in collaborative settings as well as for applications such as recommender systems. One central question is to what extent the “structurer” – be it human or machine – is driven by his/its own prior structures, and to what extent by the structures created by others such as one’s communities.