Jade Goldstein

The use of MMR, diversity-based reranking for reordering documents and producing summaries

This paper presents a method for combining query-relevance with information-novelty in the context of text retrieval and summarization. The Maximal Marginal Relevance (MMR) criterion strives to reduce redundancy while maintaining query relevance in re-ranking retrieved documents and in selecting apprw priate passages for text summarization. Preliminary results indicate some benefits for MMR diversity ranking in document retrieval and in single document summarization. The latter are borne out by the recent results of the SUMMAC conference in the evaluation of summarization systems. However, the clearest advantage is demonstrated in constructing non-redundant multi-document summaries, where MMR results are clearly superior to non-MMR passage selection.

The Use of MMR and Diversity-Based Reranking for Reodering Documents and Producing Summaries

This paper presents a method for combining query-relevance with information-novelty in the context of text retrieval and summarization. The Maximal Marginal Relevance (MMR) criterion strives to reduce redundancy while maintaining query relevance in re-ranking retrieved documents and in selecting appropriate passages for text summarization. Preliminary results indicate some benefits for MMR diversity ranking in document retrieval and in single document summarization. The latter are borne out by the recent results of the SUMMAC conference in the evaluation of summarization systems. However, the clearest advantage is demonstrated in constructing non-redundant multi-document summaries, where MMR results are clearly superior to non-MMR passage selection.

Summarizing text documents: sentence selection and evaluation metrics

Human-quality text summarization systems are di cult to design, and even more di cult to evaluate, in part because documents can di er along several dimensions, such as length, writing style and lexical usage. Nevertheless, certain cues can often help suggest the selection of sentences for inclusion in a summary. This paper presents our analysis of news-article summaries generated by sentence selection. Sentences are ranked for potential inclusion in the summary using a weighted combination of statistical and linguistic features. The statistical features were adapted from standard IR methods. The potential linguistic ones were derived from an analysis of news-wire summaries. To evaluate these features we use a normalized version of precision-recall curves, with a baseline of random sentence selection, as well as analyze the properties of such a baseline. We illustrate our discussions with empirical results showing the importance of corpus-dependent baseline summarization standards, compression ratios and carefully crafted long queries.

Multi-document summarization by sentence extraction

This paper discusses a text extraction approach to multi-document summarization that builds on single-document summarization methods by using additional, available information about the document set as a whole and the relationships between the documents. Multi-document summarization differs from single in that the issues of compression, speed, redundancy and passage selection are critical in the formation of useful summaries. Our approach addresses these issues by using domain-independent techniques based mainly on fast, statistical processing, a metric for reducing redundancy and maximizing diversity in the selected passages, and a modular framework to allow easy parameterization for different genres, corpora characteristics and user requirements.

Interactive graphic design using automatic presentation knowledge

We present three novel tools for creating data graphics: (1) SageBrush, for assembling graphics from primitive objects like bars, lines and axes, (2) SageBook, for browsing previously created graphics relevant to current needs, and (3) SAGE, a knowledge-based presentation system that automatically designs graphics and also interprets a users specifications conveyed with the other tools. The combination of these tools supports two complementary processes in a single environment: design as a constructive process of selecting and arranging graphical elements, and design as a process of browsing and customizing previous cases. SAGE enhances userdirected design by completing partial specifications, by retrieving previously created graphics based on their appearance and data content, by creating the novel displays that users specify, and by designing alternatives when users request them. Our approach was to propose interfaces employing styles of interaction that appear to support graphic design. Knowledge-based techniques were then applied to enable the interfaces and enhance their usability.

Using aggregation and dynamic queries for exploring large data sets

When working with large data sets, users perform three primary types of activities: data manipulation, data analysis, and data visualization. The data manipulation process involves the selection and transformation of data prior to viewing. This paper addresses user goals for this process and the interactive interface mechanisms that support them. We consider three classes of data manipulation goals: controlling the scope (selecting the desired portion of the data), selecting the focus of attention (concentrating on the attributes of data that are relevant to current analysis), and choosing the level of detail (creating and decomposing aggregates of data). We use this classification to evaluate the functionality of existing data exploration interface techniques. Based on these results, we have expanded an interface mechanism called the Aggregate Manipulator (AM) and combined it with Dynamic Query (DQ) to provide complete coverage of the data manipulation goals. We use real estate sales data to demonstrate how the AM and DQ synergistically function in our interface.

Creating and evaluating multi-document sentence extract summaries

This paper discusses passage extraction approaches to multidocument summarization that use available information about the document set as a whole and the relationships between the documents to build on single document summarization methodology. Multi-document summarization di ers from single in that the issues of compression, speed, redundancy and passage selection are critical in the formation of useful summaries, as well as the users goals in creating the summary. Our approach addresses these issues by using domain-independent techniques based mainly on fast, statistical processing, a metric for reducing redundancy and maximizing diversity in the selected passages, and a modular framework to allow easy parameterization for di erent genres, corpora characteristics and user requirements. We examined how humans create multi-document summaries as well as the characteristics of such summaries and use these summaries to evaluate the performance of various multidocument summarization algorithms.

A Framework for Knowledge-based, Interactive Data Exploration

Abstract In this paper we propose a framework that combines the functionality of data exploration and automatic presentation systems to create a knowledge-based interactive data exploration system. The purpose of a data exploration system is to enable users to uncover and extract relationships hidden in large data sets. The purpose of an automatic presentation system is to reduce the need for users to have graphic design expertise and for them to spend a lot of time interacting with graphics packages to view their data. Previous work on data exploration was limited to query mechanisms that were often complex to learn and difficult to use, data manipulation mechanisms that did not provide complete coverage of the operations needed by users (especially the ability to form ad hoc groupings of data), and graphics that were restricted to a small set of predefined visualizations. Automatic presentation research, although addressing these issues, has been limited to the display of small data sets. Furthermore, this research has not supported interactive, user-directed processes of design and data manipulation in automatic presentation systems. We propose a framework that overcomes these limitations of current data exploration systems and integrates new interactive capabilities with automatic presentation components. This approach to supporting data exploration integrates recent work on SageTools, an environment for interactive and automatic presentation design, with a prototypical interactive data manipulation system called IDES. In this paper we present our work on the IDES data manipulation capabilities and discuss requirements for coordinating them with automatic presentation of large data sets.

Creating charts by demonstration

“Gold” is a new interactive editor that allows a user to draw examples of the desired picture for business graphics and the system automatically produces a visualization. To specify a custom visualization in other systems, code must be written or a bewildering array of dialog boxes and commands must be used. In Gold, as the user is drawing an example of the desired visualization, knowledge of properties of the data and the typical graphics in business charts are used to generalize the example and create a picture for the actual data. The goal is to make designing a complex, composite chart almost as easy as sketching a picture on a napkin.

SUMMARIZATION: (1) USING MMR FOR DIVERSITY- BASED RERANKING AND (2) EVALUATING SUMMARIES

This paper develops a method for combining query-relevance with information-novelty in the context of text retrieval and summarization. The Maximal Marginal Relevance (MMR) criterion strives to reduce redundancy while maintaining query relevance in reranking retrieved documents and in selecting appropriate passages for text summarization. Preliminary results indicate some benefits for MMR diversity ranking in ad-hoc query and in single document summarization. The latter are borne out by the trial-run (unofficial) TREC-style evaluation of summarization systems. However, the clearest advantage is demonstrated in the automated construction of large document and non-redundant multi-document summaries, where MMR results are clearly superior to non-MMR passage selection. This paper also discusses our preliminary evaluation of summarization methods for single documents.