A Novel Hybrid Text Summarization System for Punjabi Text

Results of proposed system are compared with different baseline systems, and it is found that F score, Precision, Recall and ROUGE-2 score of the system are reasonably well as compared to other baseline systems.

Abstract

Text summarization is the task of shortening text documents but retaining their overall meaning and information. A good summary should highlight the main concepts of any text document. Many statistical-based, location-based and linguistic-based techniques are available for text summarization. This paper has described a novel hybrid technique for automatic summarization of Punjabi text. Punjabi is an official language of Punjab State in India. There are very few linguistic resources available for Punjabi. The proposed summarization system is hybrid of conceptual-, statistical-, location- and linguistic-based features for Punjabi text. In this system, four new location-based features and two new statistical features (entropy measure and Z score) are used and results are very much encouraging. Support vector machine-based classifier is also used to classify Punjabi sentences into summary and non-summary sentences and to handle imbalanced data. Synthetic minority over-sampling technique is applied for over-sampling minority class data. Results of proposed system are compared with different baseline systems, and it is found that F score, Precision, Recall and ROUGE-2 score of our system are reasonably well as compared to other baseline systems. Moreover, summary quality of proposed system is comparable to the gold summary.