Hybrid Algorithm for Multilingual Summarization of Hindi and Punjabi Documents

摘要

This paper concentrates on hybrid algorithm for multilingual summarization of Hindi and Punjabi documents. It combines the features of Hindi summarizer as suggested by CDAC Noida and Punjabi summarizer as suggested by Gupta and Lehal in 2012. In addition to this, it also suggests some new features for summarizing Hindi and Punjabi multilingual text. It is first time that this multilingual text summarizer has been proposed which supports both Hindi and Punjabi text. Nine features used in this algorithm for summarizing multilingual Hindi and Punjabi text are: 1) Key phrase extraction 2) Font feature 3) Nouns and Verbs Extraction 4) Position feature 5) Cue-phrase feature 6) Negative keywords extraction 7) Named Entities extraction 8) Relative length feature 9) extraction of number data. For each sentence, scores of each feature is calculated and then machine learning based mathematical regression is applied for identifying weights of these nine features. Sentence final-scores are calculated from feature weight equations. Top scored sentences in proper order (in same order as in input) are selected for final summary. Default summary is made at 30% compression ratio. This algorithm performs well at 30% compression ratio for both intrinsic and extrinsic measures of summary evaluation. This algorithm has been thoroughly tested on 30 Hindi-Punjabi documents and reports F-Score equal to 92.56% which is reasonably good.