Capturing the "aboutness" of documents has been a key research focus throughout the history of automated textual information processing. In this work, we represent aboutness using words and phrases that best reflect the central topics of a document. We present a machine learning approach that learns to score and rank words and phrases in a document according to their relevance to the document. We use implicit user feedback available in search engine click logs to characterize the user-perceived notion of term relevance. Using a small set of manually generated training data, we show that the surrogate training data from click logs correlates well with this data, thus eliminating the need to create data for training manually which is both expensive and fundamentally difficult to obtain for such a task. Further, we use a diverse set of features in our learning model that capitalize heavily on the structural and visual properties of web documents. In our extensive experimentation, we pay particular attention to tail web pages and show that our approach trained on mainly head web pages generalizes and performs well on all kinds of documents. In several evaluation methods using manually generated summaries and term relevance judgments, our system shows 25% improvement over other aboutness solutions.

We address the problem of ranking question answerers according to their credibility, characterized here by the probability that a given question answerer (user) will be awarded a best answer on a question given the answerer's question-answering history. This probability (represented by θ) is considered to be a hidden variable that can only be estimated statistically from specific observations associated with the user, namely the number b of best answers awarded, associated with the number n of questions answered. The more specific problem addressed is the potentially high degree of uncertainty associated with such credibility estimates when they are based on small numbers of answers. We address this problem by a kind of Bayesian smoothing. The credibility estimate will consist of a mixture of the overall population statistics and those of the specific user. The greater the number of questions asked, the greater will be the contribution of the specific user statistics relative to those of the overall population. We use the Predictive Stochastic Complexity (PSC) as an accuracy measure to evaluate several methods that can be used for the estimation. We compare our technique (Bayesian Smoothing (BS)) with maximum a priori (MAP) estimation, maximum likelihood (ML) estimation and Laplace smoothing.

This paper presents a framework called Structure Sensitive CATegorization(SSCAT), that exploits document structure for improved categorization. There are two parts to this framework, viz. (1) Documents often have layout structure, such that logically coherent text is grouped together into fields using some mark-up language. We use a log-linear model, which associates one or more features with each field. Weights associated with the field features are learnt from training data and these weights quantify the per-class importance of the field features in determining the category for the document. (2) We employ a technique that exploits the parse tree of fields that are phrasal constructs, such as title and associates weights with words in these constructs while boosting weights of important words called focus words. These weights are learnt from example instances of phrasal constructs, marked with the corresponding focus words. The learning is accomplished by training a classifier that uses linguistic features obtained from the text's parse structure. The weighted words, in fields with phrasal constructs, are used in obtaining features for the corresponding fields in the overall framework. SSCAT was tested on the supervised categorization task of over one million products from Yahoo!'s on-line shopping data. With an accuracy of over 90%, our classifier outperforms Naive Bayes and Support Vector Machines. This not only shows the effectiveness of SSCAT but also strengthens our belief that linguistic features based on natural language structure can improve tasks such as text categorization.

Semi-supervised clustering algorithms partition a given data set using limited supervision from the user. In this paper, we propose a clustering algorithm that uses supervision in terms of relative comparisons, viz., x is closer to y than to z. The success of a clustering algorithm also depends on the kind of dissimilarity measure. The proposed clustering algorithm learns the underlying dissimilarity measure while finding compact clusters in the given data set. Through our experimental studies on high-dimensional textual data sets, we demonstrate that the proposed algorithm achieves higher accuracy than the algorithms using pair-wise constraints for supervision.

We present a question answering (QA) system which learns how to detect and rank answer passages by analyzing questions and their answers (QA pairs) provided as training data. We built our system in only a few person-months using o#- the-shelf components: a part-of-speech tagger, a shallow parser, a lexical network, and a few well-known supervised learning algorithms. In contrast, many of the top TREC QA systems are large group e#orts, using customized ontologies, question classifiers, and highly tuned ranking functions. Our ease of deployment arises from using generic, trainable algorithms that exploit simple feature extractors on QA pairs. With TREC QA data, our system achieves mean reciprocal rank (MRR) that compares favorably with the best scores in recent years, and generalizes from one corpus to another. Our key technique is to recover, from the question, fragments of what might have been posed as a structured query, had a suitable schema been available. One fragment comprises selectors: tokens that are likely to appear (almost) unchanged in an answer passage. The other fragment contains question tokens which give clues about the answer type, and are expected to be replaced in the answer passage by tokens which specialize or instantiate the desired answer type. Selectors are like constants in where-clauses in relational queries, and answer types are like column names. We present new algorithms for locating selectors and answer type clues and using them in scoring passages with respect to a question.