We present a family of margin based online learning algorithms for various prediction tasks. In particular we derive and analyze algorithms for binary and multiclass categorization, regression, uniclass prediction and sequence prediction. The update steps of our different algorithms are all based on analytical solutions to simple constrained optimization problems. This unified view allows us to prove worst-case loss bounds for the different algorithms and for the various decision problems based on a single lemma. Our bounds on the cumulative loss of the algorithms are relative to the smallest loss that can be attained by any fixed hypothesis, and as such are applicable to both realizable and unrealizable settings. We demonstrate some of the merits of the proposed algorithms in a series of experiments with synthetic and real data sets.

Information filtering agents and collaborative filtering both attempt to alleviate information overload by identifying which items a user will find worthwhile. Information filtering (IF) focuses on the analysis of item content and the development of a personal user interest profile. Collaborative filtering (CF) focuses on identification of other users with similar tastes and the use of their opinions to recommend items. Each technique has advantages and limitations that suggest that the two could be beneficially combined. This paper shows that a CF framework can be used to combine personal IF agents and the opinions of a community of users to produce better recommendations than either agents or users can produce alone. It also shows that using CF to create a personal combination of a set of agents produces better results than either individual agents or other combination mechanisms. One key implication of these results is that users can avoid having to select among ag...

The optimal settings of retrieval parameters often depend on both the document collection and the query, and are usually found through empirical tuning. In this paper, we propose a family of two-stage language models for information retrieval that explicitly captures the different influences of the query and document collection on the optimal settings of retrieval parameters. As a special case, we present a two-stage smoothing method that allows us to estimate the smoothing parameters completely automatically. In the first stage, the document language model is smoothed using a Dirichlet prior with the collection language model as the reference model. In the second stage, the smoothed document language model is further interpolated with a query background language model. We propose a leave-one-out method for estimating the Dirichlet parameter of the first stage, and the use of document mixture models for estimating the interpolation parameter of the second stage. Evaluation on five different databases and four types of queries indicates that the twostage smoothing method with the proposed parameter estimation methods consistently gives retrieval performance that is close to— or better than—the best results achieved using a single smoothing method and exhaustive parameter search on the test data.

We present a probabilistic relational algebra (PRA) which is a generalization of standard relational algebra. Here tuples are assigned probabilistic weights giving the probability that a tuple belongs to a relation. Based on intensional semantics, the tuple weights of the result of a PRA expression always confirm to the underlying probabilistic model. We also show for which expressions extensional semantics yields the same results. Furthermore, we discuss complexity issues and indicate possibilities for optimization. With regard to databases, the approach allows for representing imprecise attribute values, whereas for information retrieval, probabilistic document indexing and probabilistic search term weighting can be modelled. As an important extension, we introduce the concept of vague predicates which yields a probabilistic weight instead of a Boolean value, thus allowing for queries with vague selection conditions. So PRA implements uncertainty and vagueness in combination with the...

The standard strategies for evaluation based on precision and recall are examined and their relative advantages and disadvantages are discussed. In particular, it is suggested that relevance feedback be evaluated from the perspective of the user. A number of different statistical tests are described for determining if differences in performance between retrieval methods are significant. These tests have often been ignored in the past because most are based on an assumption of normality which is not strictly valid for the standard performance measures. However, one can test this assumption using simple diagnostic plots, and if it is a poor approximation, there are a number of non-parametric alternatives.

Query expansion methods have been studied for a long time - with debatable success in many instances. In this paper we present a probabilistic query expansion model based on a similarity thesaurus which was constructed automatically. A similarity thesaurus reflects domain knowledge about the particular collection from which it is constructed. We address the two important issues with query expansion: the selection and the weighting of additional search terms. In contrast to earlier methods, our queries are expanded by adding those terms that are most similar to the concept of the query, rather than selecting terms that are similar to the query terms. Our experiments show that this kind of query expansion results in a notable improvement in the retrieval effectiveness when measured using both recall-precision and usefulness.

The Smart information retrieval project emphasizes completely automatic approaches to the understanding and retrieval of large quantities of text. We continue our work in TREC 3, performing runs in the routing, ad-hoc, and foreign language environments. Our major focus is massive query expansion: adding from 300 to 530 terms to each query. These terms come from known relevant documents in the case of routing, and from just the top retrieved documents in the case of ad-hoc and Spanish. This approach improves effectiveness from 7% to 25% in the various experiments. Other ad-hoc work extends our investigations into combining global similarities, giving an overall indication of how a document matches a query, with local similarities identifying a smaller part of the document which matches the query. Using an overlapping text window definition of "local", we achieve a 16% improvement. Introduction For over 30 years, the Smart project at Cornell University has been interested in the analy...

This paper presents an application of nonlinear neural networks to topic spotting. Neural networks allow us to model higherorder interaction between document terms and to simultaneously predict multiple topics using shared hidden features. In the context of this model, we compare two approaches to dimensionality reduction in representation: one based on term selection and another based on Latent Semantic Indexing (LSI). Two different methods are proposed for improving LSI representations for the topic spotting task. We find that term selection and our modified LSI representations lead to similar topic spotting performance, and that this performance is equal to or better than other published results on the same corpus. 1 Introduction Topic spotting is the problem of identifying which of a set of predefined topics are present in a natural language document. More formally, given a set of n topics and a document, the task is to output for each topic the probability that the topic is prese...

Though many systems exist for content-based retrieval of images, little work has been done on the audio portion of the multimedia stream. This paper presents a system to retrieve audio documents by acoustic similarity. The similarity measure is based on statistics derived from a supervised vector quantizer, rather than matching simple pitch or spectral characteristics. The system is thus able to learn distinguishing audio features while ignoring unimportant variation. Both theoretical and experimental results are presented, including quantitative measures of retrieval performance. Retrieval was tested on a corpus of simple sounds as well as a corpus of musical excerpts. The system is purely data-driven and does not depend on particular audio characteristics. Given a suitable parameterization, this method maythus be applicable to image retrieval as well.

The Smart information retrieval project emphasizes completely automatic approaches to the understanding and retrieval of large quantities of text. We continue our work in TREC 4, performing runs in the routing, ad-hoc, confused text, interactive, and foreign language environments. Introduction For over 30 years, the Smart project at Cornell University has been interested in the analysis, search, and retrieval of heterogeneous text databases, where the vocabulary is allowed to vary widely, and the subject matter is unrestricted. Such databases may include newspaper articles, newswire dispatches, textbooks, dictionaries, encyclopedias, manuals, magazine articles, and so on. The usual text analysis and text indexing approaches that are based on the use of thesauruses and other vocabulary control devices are difficult to apply in unrestricted text environments, because the word meanings are not stable in such circumstances and the interpretation varies depending on context. The applicabil...