Catálogo de publicaciones - libros

Information is often represented in text form and classified into categories for efficient browsing, retrieval, and dissemination. Unfortunately, automatic classifiers may conduct many misclassifications. One of the reasons is that the documents for training the classifiers are mainly from the categories, leading the classifiers to derive category profiles for distinguishing each category from others, rather than measuring the extent to which a document’s content overlaps that of a category. To tackle the problem, we present a technique DP4FC to help various classifiers to improve the mining of category profiles. Upon receiving a document, DP4FC helps to create dynamic category profiles with respect to the document, and accordingly helps to make proper filtering and classification decisions. Theoretical analysis and empirical results show that DP4FC may make a classifier’s performance both better and more stable.

This paper presents a method to extract citation types from scientific articles, viewed as an intrinsic part of emerging trend detection (ETD) in scientific literature. There are two main contributions in this work: (1) Definition of six categories (types) of citations in the literature that are extractable, human-understandable, and appropriate for building the interest and utility functions in emerging trend detection models, and (2) A method to classify citation types using finite-state machines which does not require user-interactions or explicit knowledge. The experimental comparative evaluations show the high performance of the method and the proposed ETD model shows the crucial role of classified citation types in the detection of emerging trends in scientific literature.

Although much work has been done on duplicate document detection (DDD) and its applications, we observe the absence of a systematic study of the performance and scalability of large-scale DDD. It is still unclear how various parameters of DDD, such as similarity threshold, precision/recall requirement, sampling ratio, document size, correlate mutually. In this paper, correlations among several most important parameters of DDD are studied and the impact of sampling ratio is of most interest since it heavily affects the accuracy and scalability of DDD algorithms. An empirical analysis is conducted on a million documents from the TREC .GOV collection. Experimental results show that even using the same sampling ratio, the precision of DDD varies greatly on documents with different size. Based on this observation, an adaptive sampling strategy for DDD is proposed, which minimizes the sampling ratio within the constraint of a given precision threshold. We believe the insights from our analysis are helpful for guiding the future large scale DDD work.

This paper addresses a real world problem: the classification of text documents in the medical domain. There are a number of approaches to classifying text documents. Here, we use a approach and argue that it is effective and computationally efficient for real-world problems. The approach uses a two-step strategy to cut down on the effort required to label each document for classification. Only a small set of positive documents are labeled initially, with others being labeled automatically as a result of the first step. The second step builds the actual text classifier. There are a number of methods that have been proposed for each step. A comprehensive evaluation of various combinations of methods is conducted to compare their performances using real world medical documents. The results show that using EM based methods to build the classifier yields better results than SVM. We also experimentally show that careful selection of a subset of features to represent the documents can improve the performance of the classifiers.

We present a novel clustering algorithm to group the XML documents by similar structures. We introduce a format to represent the XML documents for efficient processing. We develop a criterion function that do not require the pair-wise similarity to be computed between two individual documents, rather measures the similarity at clustering level utilising structural information of the XML documents. The experimental analysis shows the method to be fast and accurate.

In this paper we introduce a novel document clustering approach that solves some major problems of traditional document clustering approaches. Instead of depending on traditional vector space model, this approach represents a set of documents as bipartite graphs using domain knowledge in ontology. In this representation, the concepts of the documents are classified according to their relationships with documents that are reflected on the bipartite graph. Using the concept groups, documents are clustered based on the concepts’ contribution to each document. Through the mutual-refinement relationship with concept groups and document groups, the two groups are recursively refined. Our experimental results on MEDLINE articles show that our approach outperforms two leading document clustering algorithms: BiSecting K-means and CLUTO. In addition to its decent performance, our approach provides a meaningful explanation for each document cluster by identifying its most contributing concepts, thus helps users to understand and interpret documents and clustering results.

In the literature of web search and mining, researchers used to consider the World Wide Web as a flat network, in which each page as well as each hyperlink is treated identically. However, it is the common knowledge that the Web is organized with a natural hierarchical structure according to the URLs of pages. Exploring the hierarchical structure, we found several level-biased characteristics of the Web. First, the distribution of pages over levels has a spindle shape. Second, the average indegree in each level decreases sharply when the level goes down. Third, although the indegree distributions in deeper levels obey the same power law with the global indegree distribution, the top levels show a quite different statistical characteristic. We believe that these new discoveries might be essential to the Web, and by taking use of them, the current web search and mining technologies could be improved and thus better services to the web users could be provided.

Existing web usage mining techniques focus only on discovering knowledge based on the statistical measures obtained from the characteristics of web usage data. They do not consider the dynamic nature of web usage data. In this paper, we present an algorithm called (ustering of vlutionary ten-based web ccess sequences) to cluster web access sequences based on their . In this approach, Web access sequences that have similar change patterns in their support counts in the history are grouped into the same cluster. The intuition is that often are event/task-driven. As a result, related to the same event/task are expected to be accessed in similar ways over time. Such clusters are useful for several applications such as intelligent web site maintenance and personalized web services.

Online auction Web sites are fast changing and highly dynamic. It is difficult to digest the poorly organized and vast amount of information contained in the auction sites. We develop a framework aiming at automatically extracting the product features and summarizing the hot item features across different auction Web sites. One challenge of this problem is to extract useful information from the product descriptions provided by the sellers, which vary largely in the layout format. We formulate the problem as a single graph labeling problem using conditional random fields which can model the relationship among the neighbouring tokens in a Web page, the tokens from different pages, as well as various information such as the hot item features across different auction sites. We have conducted extensive experiments from several real-world auction Web sites to demonstrate the effectiveness of our framework.

Session similarity is a key issue in web session clustering. Existing approaches vary on session representation and similarity computation. However, they do not consider the similarity between pages, which is crucial due to the semantic gap between URLs and corresponding application events. This paper presents a domain taxonomy-based clustering approach, which extends the WLCS technique by integrating page similarity to compute session similarity. The approach can be applied to both usage and navigation clustering purposes.