The KNOWITALL system aims to automate the tedious process of extracting large collections of facts (e.g., names of scientists or politicians) from the Web in an unsupervised, domain-independent, and scalable manner. The paper presents an overview of KNOW-ITALL’s novel architecture and design principles, emphasizing its distinctive ability to extract information without any hand-labeled training examples. In its first major run, KNOW-ITALL extracted over 50,000 facts, but suggested a challenge: How can we improve KNOW-ITALL’s recall and extraction rate without sacrificing precision? This paper presents three distinct ways to address this challenge and evaluates their performance. Pattern Learning learns domain-specific extraction rules, which enable additional extractions. Subclass Extraction automatically identifies sub-classes in order to boost recall. List Extraction locates lists of class instances, learns a “wrapper ” for each list, and extracts elements of each list. Since each method bootstraps from KNOWITALL’s domainindependent methods, the methods also obviate hand-labeled training examples. The paper reports on experiments, focused on named-entity extraction, that measure the relative efficacy of each method and demonstrate their synergy. In concert, our methods gave KNOW-ITALL a 4-fold to 8-fold increase in recall, while maintaining high precision, and discovered over 10,000 cities missing from the Tipster Gazetteer.

When documents are organized in a large number of topic categories, the categories are often arranged in a hierarchy. The U.S. patent database and Yahoo are two examples.

This paper shows how a text classifier's need for labeled training data can be reduced by a combination of active learning and Expectation Maximization (EM) on a pool of unlabeled data. Query-by-Committee is used to actively select documents for labeling, then EM with a naive Bayes model further improves classification accuracy by concurrently estimating probabilistic labels for the remaining unlabeled documents and using them to improve the model. We also present a metric for better measuring disagreement among committee members; it accounts for the strength of their disagreement and for the distribution of the documents. Experimental results show that our method of combining EM and active learning requires only half as many labeled training examples to achieve the same accuracy as either EM or active learning alone. Keywords: text classification active learning unsupervised learning information retrieval 1 Introduction In many settings for learning text classifiers, obtaining lab...

The World Wide Web is a vast source of information accessible to computers, but understandable only to humans. The goal of the research described here is to automatically create a computer understandable knowledge base whose content mirrors that of the World Wide Web. Such a knowledge base would enable much more effective retrieval of Web information, and promote new uses of the Web to support knowledge-based inference and problem solving. Our approach is to develop a trainable information extraction system that takes two inputs. The first is an ontology that defines the classes (e.g., company, person, employee, product) and relations (e.g., employed_by, produced_by) of interest when creating the knowledge base. The second is a set of training data consisting of labeled regions of hypertext that represent instances of these classes and relations. Given these inputs, the system learns to extract information from other pages and hyperlinks on the Web. This article describes our general a...

Recently there has been significant interest in supervised learning algorithms that combine labeled and unlabeled data for text learning tasks. The co-training setting [1] applies to datasets that have a natural separation of their features into two disjoint sets. We demonstrate that when learning from labeled and unlabeled data, algorithms explicitly leveraging a natural independent split of the features outperform algorithms that do not. When a natural split does not exist, co-training algorithms that manufacture a feature split may out-perform algorithms not using a split. These results help explain why co-training algorithms are both discriminative in nature and robust to the assumptions of their embedded classifiers. Categories and Subject Descriptors I.2.6 [Artificial Intelligence]: Learning; H.3.3 [Information Storage and Retrieval]: Information Search and Retrieval--- Information Filtering Keywords co-training, expectation-maximization, learning with labeled and unlabeled...

Recently, there has been much effort in making databases for molecular biology more accessible and interoperable. However, information in text form, such as MEDLINE records, remains a greatly underutilized source of biological information. We have begun a research effort aimed at automatically mapping information from text sources into structured representations, such as knowledge bases. Our approach to this task is to use machine-learning methods to induce routines for extracting facts from text. We describe two learning methods that we have applied to this task --- a statistical text classification method, and a relational learning method --- and our initial experiments in learning such information-extraction routines. We also present an approach to decreasing the cost of learning information-extraction routines by learning from "weakly" labeled training data.

Viral marketing takes advantage of networks of influence among customers to inexpensively achieve large changes in behavior. Our research seeks to put it on a firmer footing by mining these networks from data, building probabilistic models of them, and using these models to choose the best viral marketing plan. Knowledge-sharing sites, where customers review products and advise each other, are a fertile source for this type of data mining. In this paper we extend our previous techniques, achieving a large reduction in computational cost, and apply them to data from a knowledge-sharing site. We optimize the amount of marketing funds spent on each customer, rather than just making a binary decision on whether to market to him. We take into account the fact that knowledge of the network is partial, and that gathering that knowledge can itself have a cost. Our results show the robustness and utility of our approach.

Keyphrases are an important means of document summarization, clustering, and topic search. Only a small minority of documents have author-assigned keyphrases, and manually assigning keyphrases to existing documents is very laborious. Therefore it is highly desirable to automate the keyphrase extraction process. This paper shows that a simple procedure for keyphrase extraction based on the naive Bayes learning scheme performs comparably to the state of the art. It goes on to explain how this procedure’s performance can be boosted by automatically tailoring the extraction process to the particular document collection at hand. Results on a large collection of technical reports in computer science show that the quality of the extracted keyphrases improves significantly when domain-specific information is exploited.

In this paper we describe a trainable object detector and its instantiations for detecting faces and cars at any size, location, and pose. To cope with variation in object orientation, the detector uses multiple classifiers, each spanning a different range of orientation. Each of these classifiers determines whether the object is present at a specified size within a fixed-size image window. To find the object at any location and size, these classifiers scan the image exhaustively. Each classifier is based on the statistics of localized parts. Each part is a transform from a subset of wavelet coefficients to a discrete set of values. Such parts are designed to capture various combinations of locality in space, frequency, and orientation. In building each classifier, we gathered the class-conditional statistics of these part values from representative samples of object and non-object images. We trained each classifier to minimize classification error on the training set by using Adaboost with Confidence-Weighted Predictions (Shapire and Singer, 1999). In detection, each classifier computes the part values within the image window and looks up their associated classconditional probabilities. The classifier then makes a decision by applying a likelihood ratio test. For efficiency, the classifier evaluates this likelihood ratio in stages. At each stage, the classifier compares the partial likelihood ratio to a threshold and makes a decision about whether to cease evaluation—labeling the input as non-object—or to continue further evaluation. The detector orders these stages of evaluation from a low-resolution to a high-resolution search of the image. Our trainable object detector achieves reliable and efficient detection of human faces and passenger cars with out-of-plane rotation.

A central problem in machine learning is identifying a representative set of features from which to construct a classification model for a particular task. This thesis addresses the problem of feature selection for machine learning through a correlation based approach. The central hypothesis is that good feature sets contain features that are highly correlated with the class, yet uncorrelated with each other. A feature evaluation formula, based on ideas from test theory, provides an operational definition of this hypothesis. CFS (Correlation based Feature Selection) is an algorithm that couples this evaluation formula with an appropriate correlation measure and a heuristic search strategy. CFS was evaluated by experiments on artificial and natural datasets. Three machine learning algorithms were used: C4.5 (a decision tree learner), IB1 (an instance based learner), and naive Bayes. Experiments on artificial datasets showed that CFS quickly identifies and screens irrelevant, redundant, and noisy features, and identifies relevant features as long as their relevance does not strongly depend on other features. On natural domains, CFS typically eliminated well over half the features. In most cases, classification accuracy using the reduced feature set equaled or bettered accuracy using the complete feature set.