H.3.8. Natural Language Processing
P. Kavehzadeh; M. M. Abdollah Pour; S. Momtazi
Abstract
Over the last few years, text chunking has taken a significant part in sequence labeling tasks. Although a large variety of methods have been proposed for shallow parsing in English, most proposed approaches for text chunking in Persian language are based on simple and traditional concepts. In this paper, ...
Read More
Over the last few years, text chunking has taken a significant part in sequence labeling tasks. Although a large variety of methods have been proposed for shallow parsing in English, most proposed approaches for text chunking in Persian language are based on simple and traditional concepts. In this paper, we propose using the state-of-the-art transformer-based contextualized models, namely BERT and XLM-RoBERTa, as the major structure of our models. Conditional Random Field (CRF), the combination of Bidirectional Long Short-Term Memory (BiLSTM) and CRF, and a simple dense layer are employed after the transformer-based models to enhance the model's performance in predicting chunk labels. Moreover, we provide a new dataset for noun phrase chunking in Persian which includes annotated data of Persian news text. Our experiments reveal that XLM-RoBERTa achieves the best performance between all the architectures tried on the proposed dataset. The results also show that using a single CRF layer would yield better results than a dense layer and even the combination of BiLSTM and CRF.
Document and Text Processing
S. Momtazi; A. Rahbar; D. Salami; I. Khanijazani
Abstract
Text clustering and classification are two main tasks of text mining. Feature selection plays the key role in the quality of the clustering and classification results. Although word-based features such as term frequency-inverse document frequency (TF-IDF) vectors have been widely used in different applications, ...
Read More
Text clustering and classification are two main tasks of text mining. Feature selection plays the key role in the quality of the clustering and classification results. Although word-based features such as term frequency-inverse document frequency (TF-IDF) vectors have been widely used in different applications, their shortcoming in capturing semantic concepts of text motivated researches to use semantic models for document vector representations. Latent Dirichlet allocation (LDA) topic modeling and doc2vec neural document embedding are two well-known techniques for this purpose. In this paper, we first study the conceptual difference between the two models and show that they have different behavior and capture semantic features of texts from different perspectives. We then proposed a hybrid approach for document vector representation to benefit from the advantages of both models. The experimental results on 20newsgroup show the superiority of the proposed model compared to each of the baselines on both text clustering and classification tasks. We achieved 2.6% improvement in F-measure for text clustering and 2.1% improvement in F-measure in text classification compared to the best baseline model.