[dianek | vijayd | fu] @ email.unc.edu In this paper we investigate the effectiveness of a documentindependent technique for eliciting feedback from users about their information problems. We propose that such a technique can be used to elicit terms from users for use in query expansion and as a follow-up when ambiguous queries are initially posed by users. We design a feedback form to obtain additional information from users, administer the form to users after initial querying, and create a series of experimental runs based on the information that we obtained from the form. Results demonstrate that the form was successful at eliciting more information from users and that this additional information significantly improved retrieval performance. Our results further demonstrate a strong relationship between query length and performance.

Although no unified definition of the concept of search strategy in Information Retrieval (IR) exists so far, its importance is manifest: nonexpert users, directly interacting with an IR system, apply alimited portfolio of simple actions; they do not know how to react in critical situations; and they often do not even realize that their difficulties are due to strategic problems. Auser interface to an IR system should therefore provide some strategic help, focusing user’s attention on strategic issues and providing tools to generate better strategies. Because neither the user nor the system can autonomously solve the information problem, but they complement each other, we propose acollaborative coaching approach,inwhichthetwopartnerscooperate:theuser retains the control of the session and the system provides suggestions. The effectiveness of the approach is demonstrated by a conceptual analysis, a prototype knowledge-based system named FIRE, and its evaluation through informal laboratory experiments.

We extend the traditional active learning framework to include feedback on features in addition to labeling instances, and we execute a careful study of the effects of feature selection and human feedback on features in the setting of text categorization. Our experiments on a variety of categorization tasks indicate that there is significant potential in improving classifier performance by feature re-weighting, beyond that achieved via membership queries alone (traditional active learning) if we have access to an oracle that can point to the important (most predictive) features. Our experiments on human subjects indicate that human feedback on feature relevance can identify a sufficient proportion of the most relevant features (over 50 % in our experiments). We find that on average, labeling a feature takes much less time than labeling a document. We devise an algorithm that interleaves labeling features and documents which significantly accelerates standard active learning in our simulation experiments. Feature feedback can complement traditional active learning in applications such as news filtering, e-mail classification, and personalization, where the human teacher can have significant knowledge on the relevance of features.

Abstract. It has long been a tradition of evaluating information retrieval systems with very simple user models and very simple tasks: the task is to retrieve relevant documents to a user need described by a query. TREC, the Text REtrieval Conference sponsored by NIST, raised the bar by providing large scale collections, well defined user needs, independently judged documents, and a specified form of success. Groups from around the world all tackled this same task that allowed wide analysis of just what factors influenced system performance. Yet there was concern, as system performance improvement did not always lead to human performance improvement, so a concerted effort to study how people interact with information retrieval systems was undertaken in the Interactive Track of TREC. This paper describes this track, some of the experiments that we have undertaken in this track, and highlights some of the real problems in such evaluation. There are two key issues that we have often observed in interactive information retrieval. The first issue is that human preference is often not correlated with

features and structure of Web-based online IR systems

We show how personalization techniques can be exploited to implement more adaptive and effective information access systems in electronic publishing. We distinguish persistent (or long term) and ephemeral (or short term) personalization, and we describe how both of them can be profitably applied in information filtering and retrieval systems used, via a specialized Web portal, by physicists in their daily job. By means of several experimental results, we demonstrate that persistent personalization is needed and useful for information filtering systems, and ephemeral personalization leads to more effective and usable information retrieval systems.

Abstract. Modern Information Retrieval Systems match the terms contained in a user’s query with available documents through the use of an index. In this work, we propose a method for expanding the query with its associated terms, in order to increase the system recall. The proposed method is based on a novel fuzzy clustering of the index terms, using their common occurrence in documents as clustering criterion. The clusters which are relevant to the terms of the query form the query context. The terms of the clusters that belong to the context are used to expand the query. Clusters participate in the expansion according to their degree of relevance to the query. Precision of the result is thus improved. This statistical approach for query expansion is useful when no a priori semantic knowledge is available. 1

Abstract not found

Abstract. We have developed an interactive query refinement tool that helps users search a knowledge base for solutions to problems with electronic equipment. The system is targeted towards non-technical users, who are often unable to formulate precise problem descriptions on their own. Two distinct but interrelated functionalities support the refinement of a vague, non-technical initial query into a more precise problem description: a synonymy mechanism that allows the system to match non-technical words in the query with corresponding technical terms in the knowledge base, and a novel refinement mechanism that helps the user build up successively longer and more precise problem descriptions starting from the seed of the initial query. A natural language parser is used both in the application of context-sensitive synonymy rules and the construction of the refinement tree. 1

Abstract. Persistent and ephemeral personalization techniques can be exploited to implement more adaptive and effective information access systems in electronic publishing. Within the TIPS project, we effectively applied both these techniques in information filtering and retrieval systems used, via the specialized Torii portal, by physicists in their daily job. 1.