Search

Given a large unlabeled document collection, the aim of this paper is to develop an accurate and efficient algorithm for solving the clustering problem over this collection. Document collections typically contain tens or hundreds of thousands of documents, with thousands or tens of thousands of features (i.e., distinct words). Most existing clustering algorithms struggle to find accurate solutions on such large data sets. The proposed algorithm overcomes this difficulty by an incremental approach, incrementing the number of clusters progressively from an initial value of one to a set value. At each iteration, the new candidate cluster is initialized using a partitioning approach which is guaranteed to minimize the objective function. Experiments have been carried out over six, diverse datasets and with different evaluation criteria, showing that the proposed algorithm has outperformed comparable state-of-the-art clustering algorithms in all cases.

Extracting meaningful features from documents can prove critical for a variety of tasks such as classification, clustering and semantic analysis. However, traditional approaches to document feature extraction mainly rely on first-order word statistics that are very high dimensional and do not capture well the semantic of the documents. For this reason, in this paper we present a novel approach that extracts document features based on a combination of a constructed word taxonomy and a word embedding in vector space. The feature extraction consists of three main steps: first, a word embedding technique is used to map all the words in the vocabulary onto a vector space. Second, the words in the vocabulary are organised into a hierarchy of clusters (word clusters) by using k-means hierarchically. Lastly, the individual documents are projected onto the word clusters based on a predefined set of keywords, leading to a compact representation as a mixture of keywords. The extracted features can be used for a number of tasks including document classification and clustering as well as semantic analysis of the documents generated by specific individuals over time. For the experiments, we have employed a dataset of transcripts of phone calls between claim managers and clients collected by the Transport Accident Commission of the Victorian Government. The experimental results show that the proposed approach has been capable of achieving comparable or higher accuracy than conventional feature extraction approaches and with a much more compact representation.

© 2018 Wiley Periodicals, Inc. Maximum-margin clustering is an extension of the support vector machine (SVM) to clustering. It partitions a set of unlabeled data into multiple groups by finding hyperplanes with the largest margins. Although existing algorithms have shown promising results, there is no guarantee of convergence of these algorithms to global solutions due to the nonconvexity of the optimization problem. In this paper, we propose a simulated annealing-based algorithm that is able to mitigate the issue of local minima in the maximum-margin clustering problem. The novelty of our algorithm is twofold, ie, (i) it comprises a comprehensive cluster modification scheme based on simulated annealing, and (ii) it introduces a new approach based on the combination of k-means++ and SVM at each step of the annealing process. More precisely, k-means++ is initially applied to extract subsets of the data points. Then, an unsupervised SVM is applied to improve the clustering results. Experimental results on various benchmark data sets (of up to over a million points) give evidence that the proposed algorithm is more effective at solving the clustering problem than a number of popular clustering algorithms.