Search

In non-negative matrix factorization, it is difficult to find the optimal non-negative factor matrix in each iterative update. However, with the help of transformation matrix, it is able to derive the optimal non-negative factor matrix for the transformed cost function. Transformation matrix based nonnegative matrix factorization method is proposed and analyzed. It shows that this new method, with comparable complexity as the priori schemes, is efficient in enhancing nonnegative matrix factorization and achieves better performance in NMF based text clustering. [ABSTRACT FROM AUTHOR]

Web content mining is an important way of Internet information collection and analysis, but most of web pages are non-XML documents, how to extract useful information efficiently from massive web pages is a interesting research topic. On the basis of analyzing the features of web content mining, a XML-based web content mining method is proposed. Firstly, it defines the authority web page using the HITS algorithms, then transforms the non-XML documents into structured XML documents after the data cleaning and extracting by HTML Tidy, finally does data mining on the XML document using text clustering techniques. A science paper web site is chosen as a case study for Web content extracting. Experimental results show that the proposed method works well, it can extract web content efficiently and automatically. [ABSTRACT FROM AUTHOR]

Clustering web snippet results returned from search engine helps facilitate browsing and navigating for users. Due to the extremely short length of web snippets, many traditional clustering techniques which adopt the bag of words model often yields unsatisfactory clustering results. In this paper, we propose a method of text enrichment for improving performance of web snippet clustering. The main idea is to expand the original snippets with some related conceptual terms. We apply the Open Directory Project (ODP), a web taxonomy organized by humans, to provide the concept hierarchy of the web contents. Using a test data set of 240 queries, we performed the experiments by using two clustering techniques: K-means clustering as the non-overlapping approach and the Suffix Tree Clustering (STC) as the overlapping approach. Using the proposed text enrichment method, the K-means clustering yielded the overall performance improvement up to 15.51% based on the F1 measure. On the other hand, the Suffix Tree Clustering with text enrichment helped improve the performance up to 53.71%. [ABSTRACT FROM AUTHOR]

In this research, the NTSO (Neural Text Self Organizer) is proposed as the approach to text clustering. It is required to encode documents into numerical vectors for using a traditional approach to text clustering. The two main problems, huge dimensionality and sparse distribution are caused by encoding so. The idea of this research is to encode documents into string vectors and use the NTSO as the approach to text clustering. As the empirical validation, we will compare the NTSO with other text clustering approaches with respect to the speed and the performance. [ABSTRACT FROM AUTHOR]

Most text clustering techniques are based on words and/or phrases weights in the text. Such representation is often unsatisfactory because it ignores the relationships between terms, and considers them as independent features. In this paper, a new semantic similarity based model (SSBM) is proposed. The semantic similarity based model computes semantic similarities by utilizing WordNet as an ontology. The proposed model captures the semantic similarities between documents that contain semantically similar terms but unnecessarily syntactically identical. The semantic similarity based model assigns a new weight to document terms reflecting the semantic relationships between terms that co-occur literally in the document. Our model in conjunction with the extended gloss overlaps measure and the adapted Lesk algorithm solves ambiguity, synonymy problems that are not detected using traditional term frequency based text mining techniques. The proposed model is evaluated on the Reuters-21578 and the 20-Newsgroups text collections datasets. The performance is assessed in terms of the Fmeasure, Purity and Entropy quality measures. The obtained results show promising performance improvements compared to the traditional term based vector space model (VSM) as well as other existing methods that include semantic similarity measures in text clustering. [ABSTRACT FROM AUTHOR]

In this paper, we proposed a bayesian mixture model, in which introduce a context variable, which has Dirichlet prior, in a bayesian framework to model text multiple topics and then clustering. It is a novel unsupervised text learning algorithm to cluster large-scale web data. In addition, parameters estimation we adopt Maximum Likelihood (ML) and EM algorithm to estimate the model parameters, and employed BIC principle to determine the number of clusters. Experimental results show that method we proposed distinctly outperformed baseline algorithms. [ABSTRACT FROM AUTHOR]

Fuzzy Kernel C-Means (FKCM) algorithm can improve accuracy significantly compared with classical Fuzzy C-Means algorithms for nonlinear separability, high dimension and clusters with overlaps in input space. Despite of these advantages, several features are subjected to the applications in real world such as local optimal, outliers, the c parameter must be assigned in advance and slow convergence speed. To overcome these disadvantages, Semi-Supervised learning and validity index are employed. Semi-Supervised learning uses limited labeled data to assistant a bulk of unlabeled data. It makes the FKCM avoid drawbacks proposed. The number of cluster will great affect clustering performance. It isn΄t possible to assume the optimal number of clusters especially to large text corps. Validity function makes it possible to determine the suitable number of cluster in clustering process. Sparse format, Cscatter and gathering strategy save considerable store space and computation time. Experimental results on the Reuters-21578 benchmark dataset demonstrate that the algorithm proposed is more flexibility and accuracy than the state-of-art FKCM. [ABSTRACT FROM AUTHOR]

The paper presents an autonomous text classification module for a news web portal for the Romanian language. Statistical natural language processing techniques are combined in order to achieve a completely autonomous functionality of the portal. The news items are automatically collected from a large number of news sources using web syndication. Afterward, machine-learning techniques are used for achieving an automatic classification of the news stream. Firstly, the items are clustered using an agglomerative algorithm and the resulting groups correspond to the main news topics. Thus, more information about each of the main topics is acquired from various news sources. Secondly, text classification algorithms are applied to automatically label each cluster of news items in a predetermined number of classes. More than a thousand news items were employed for both the training and the evaluation of the classifiers. The paper presents a complete comparison of the results obtained for each method. [ABSTRACT FROM AUTHOR]