Afaan Oromo News Text Summarization Using Sentence Scoring Method

Abstract

Nowadays information is available in both electronic (soft copy) and hard copy format. Due to presence of huge amount of electronic format information it needs lot of time and money to access information. So, to get information in short period of time with minimum amount of money it needs a system which summarize and present it for readers. Therefore, this research attempt on the Afaan Oromo News Text Summarization Using Sentence Scoring Method. The researcher used features like thematic words, word frequency, title words, term weight, cue phrases, name of numbers and sentence position in this work to achieve the study of way of designing and developing single document summarizer for Afaan Oromo news text. So, using extractive method the researcher did experiments on ten selected topics out of 30 gathered topics. Manual summary is prepared by three Afaan Oromo speaker domain expert. The system is developed by NLTK using python programming language. The developed system calculates the score of the sentence by adding the score of each individual words and the score is computed for sentence. The system generates the summary by extracting n top scored sentences at three extraction rate i.e. at 20%, 30% and 40%. The system was evaluated based on the nine experimental situations both subjectively and objectively. Subjective evaluation focused on the structure of the summary referential clarity, to check as there is any redundancy or not, in-formativeness, grammatical correctness and coherence of the summary. So, at 20%, 30% and 40% extraction rate grammatical correctness is 90%, 90% and 92% respectively, concerning redundancy at 20%, 30% and 40% extraction rate performance of the summarizer system is 72%, 82% and 84% respectively. And at 20%, 30% and 40% extraction rate performs 66%, 74% and 86% in concerning referential clarity. Coherency of the summary evaluation performed at 20%, 30% and 40% extraction rate 62%, 66% and 72% respectively. And concerning informativeness at 20%, 30% and 40% extraction rate the performance of automatic summary was 74%, 78% and 86%. And with that of objective evaluation the three metrics recall, precision and F-score computed and 86.1% was performed by the system.