Search

Due to the enormous amount of user-generated content being generated on the web, labeling such data is a time-consuming and expensive endeavor. As a result, we have limited annotated data and the vast majority of data are unlabeled. Analysis reveals that extracting (external) knowledge from unlabeled data and integrating it with knowledge extracted from labeled data is a beneficial task for text information processing, in particular text classification. In this paper, we present a hybrid approach for classifying sentiments that employs external knowledge, which is categorized as either general-purpose sentiment knowledge or domain-related knowledge. General-purpose sentiment knowledge is extracted from sentiment lexicons, whereas domain-related knowledge is extracted from unlabeled data from the same or related domains. Similar domains for a given domain are identified based on their similarity score in terms of overlapping features. The proposed approach utilizes both forms of external knowledge and combines them with logistic regression to train an improved classification model. The classification model uses the conventional gradient descent algorithm for optimization, and its convergence analysis indicates that it is convex and converges to the global optimum. The proposed approach is empirically evaluated and compared to three baselines and one state-of-the-art method using standard performance evaluation metrics on a multi-domain sentiment dataset. The experiment results are encouraging, demonstrating that the proposed approach considerably outperforms the baseline approaches and outperforms the state-of-the-art approach by up to 2% in terms of both f-score and accuracy. • A hybrid approach for document-level sentiment classification that makes use of external knowledge. • A domain-specific knowledge extraction approach from unlabeled text documents. • A detailed convergence analysis of the proposed optimization model used in the proposed approach. [ABSTRACT FROM AUTHOR]

• Two novel term weighting schemes are proposed for text classification. • The proposed schemes are named as SQRT_TF-IGM imp and TF-IGM imp. • These schemes are derived from standard inverse gravity moment formula. • SQRT_TF-IGM imp method generally outperformed all other schemes. • The performance gains of the SQRT_TF-IGM imp scheme are statistically significant. Text classification is one of the popular high dimensional classification problems where providing better feature vector representations explicitly improve classification performances. Thus, assigning appropriate weights to features or terms are crucial for obtaining effective feature vector representations. The methods used for weighting terms in text classification are called term weighting schemes. Although there exist some term weighting schemes for text classification, they are not fully effective and researchers still focus on proposing new term weighting schemes. In this study, two novel term weighting schemes namely SQRT_TF-IGM imp and TF-IGM imp derived from standard inverse gravity moment formula are proposed to improve weighting behaviors of existing TF-IGM scheme especially for some extreme cases. The performances of proposed schemes are compared with two standard IGM based schemes and five other state-of-the-art term weighting methods on both unbalanced (Reuters-21578) and balanced (20 Mini Newsgroups and 20 Newsgroups) datasets with KNN, SVM, and NN classifiers. Micro-F1 and macro-F1 are used as success measures. The experiments are conducted with various different feature sizes to examine the effects of the feature size on the success of weighting. The experimental results showed that the proposed SQRT_TF-IGM imp method generally outperformed all schemes including both standard TF-IGM and SQRT_TF-IGM schemes. However, the proposed TF-IGM imp scheme also showed mostly better performance than standard TF-IGM. To demonstrate validity of the proposed weighting scheme having maximum performance, t -test is also used and it can be stated that the performance gains obtained by the proposed SQRT_TF-IGM imp weighting scheme compared to standard SQRT_TF-IGM are statistically significant. [ABSTRACT FROM AUTHOR]