Academia.eduAcademia.edu

Figure 1 – uploaded by IRJET Journal

See full PDFdownloadDownload figure
3.1 Load raw text

Figure 1 3.1 Load raw text

Related Figures (5)

BERT (Bidirectional Encoder Representations from Transformers) is a neural network-based technique for natural language processing. It is a pre-trained open-sourced model from Google. It helps computers to understand the language a bit more as humans do. The input text is summarized using the BERTSUM model, which is fine-tuned BERT for extractive summarization. The architecture of BERTSUM is shown in Fig- 2. In the BERTSUM model, at the start of each sentence, a [CLS] token is added, and between every two sentences, a [SEP] token is added to separate the sentences. Here, a [CLS] token is added to collect the preceding sentence context. Now each word of the sentence is tokenized using token embeddings. There is also a difference in segment embeddings. In the BERTSUM model, each sentence is assigned an embedding of Ea or Eb depending on whether the sentence is even or odd. If the sequence is [s1, s2, s3] then the segment embeddings are [Ea, Eb, Ea]. This way, all sentences are embedded and sent into further layers. BERTSUM assigns scores to each sentence that represents how much value that sentence adds to the overall document. So, [s1, s2, s3] is assigned [score1, score2, score3]. The sentences with the highest scores are then collected and rearranged to summarise the input text. In the output, the precise and summarized text is generated.
BERT (Bidirectional Encoder Representations from Transformers) is a neural network-based technique for natural language processing. It is a pre-trained open-sourced model from Google. It helps computers to understand the language a bit more as humans do. The input text is summarized using the BERTSUM model, which is fine-tuned BERT for extractive summarization. The architecture of BERTSUM is shown in Fig- 2. In the BERTSUM model, at the start of each sentence, a [CLS] token is added, and between every two sentences, a [SEP] token is added to separate the sentences. Here, a [CLS] token is added to collect the preceding sentence context. Now each word of the sentence is tokenized using token embeddings. There is also a difference in segment embeddings. In the BERTSUM model, each sentence is assigned an embedding of Ea or Eb depending on whether the sentence is even or odd. If the sequence is [s1, s2, s3] then the segment embeddings are [Ea, Eb, Ea]. This way, all sentences are embedded and sent into further layers. BERTSUM assigns scores to each sentence that represents how much value that sentence adds to the overall document. So, [s1, s2, s3] is assigned [score1, score2, score3]. The sentences with the highest scores are then collected and rearranged to summarise the input text. In the output, the precise and summarized text is generated.
3.3.3 Scoring and selecting words , International Research Journal of Engineering and Technology (IRJET) _ e- T Volume: 08 Issue: 05 | May 2021 www.irjet.net p- After creating a word co-occurrence matrix, RAKE calculates the keyword score. This score of candidate word is computed by taking the degree of a word in the matrix (i.e. the sum of the number of co-occurrences the word has with any other content word in the text), as the word frequency (i.e. the number of times the word appears in the text) which is given by,
3.3.3 Scoring and selecting words , International Research Journal of Engineering and Technology (IRJET) _ e- T Volume: 08 Issue: 05 | May 2021 www.irjet.net p- After creating a word co-occurrence matrix, RAKE calculates the keyword score. This score of candidate word is computed by taking the degree of a word in the matrix (i.e. the sum of the number of co-occurrences the word has with any other content word in the text), as the word frequency (i.e. the number of times the word appears in the text) which is given by,
If we were to compute the keyword score for phrases we will get as following: If we were to compute the keyword score for each words in our example, we will get as following:
If we were to compute the keyword score for phrases we will get as following: If we were to compute the keyword score for each words in our example, we will get as following:
Table -1: Test set results on the CNN/DailyMail dataset using ROUGE F1 The results and accuracy of the proposed system are comparatively higher than other systems. For the summarizer, we have used the BERTSUM model, which outperforms the other models. As shown in Table- 1 [8], all BERT-based models outperformed previous state-of-the-art models by a large margin. BERTSUM with Transformer achieved the best performance on all three metrics. The BERTSUM with LSTM model does not have an obvious influence on the summarization performance than the Classifier model. So among all the models, we have chosen BERTSUM + Transformer model, for the text summarization which has the best score compared to all the previously proposed systems. which has the best score compared to all the previously
Table -1: Test set results on the CNN/DailyMail dataset using ROUGE F1 The results and accuracy of the proposed system are comparatively higher than other systems. For the summarizer, we have used the BERTSUM model, which outperforms the other models. As shown in Table- 1 [8], all BERT-based models outperformed previous state-of-the-art models by a large margin. BERTSUM with Transformer achieved the best performance on all three metrics. The BERTSUM with LSTM model does not have an obvious influence on the summarization performance than the Classifier model. So among all the models, we have chosen BERTSUM + Transformer model, for the text summarization which has the best score compared to all the previously proposed systems. which has the best score compared to all the previously
Table -1: Results of Hypernym Validation 5. Conclusion & Future scope Multiple Choice Questions (MCQs) are generated successfully. Efficient questions are produced with good quality distractors. The problem of manually creating questions is solved with the proposed system. The proposed system creates automated questions with the help of NLP that reduces human intervention and it is a cost and time- effective system. And the accuracy of the distractor generated is reasonably high. This system not only helps teachers with E-assessments but also helps students who are preparing for competitive exams. Students can test their ability to solve the questions and can also check their understanding of the concepts. For generating distractors, the wordnet approach is used We evaluate the quality of wordnet by examining the validity of its constituent synonyms and its hypernym-hyponym pairs [9]. The hypernymy validation procedure was testec on the set of all direct hypernyms for noun synsets in the Princeton WordNet (v2.1). There are a total of 79297 hypernym-hyponym pairs constituting this set. The number of pairs validated by the algorithm at different steps of the approach is summarized in Table 2. In all, the algorithm was able to validate 56203 out of 79297 noun hypernym relation pairs in the Princeton Wordnet, giving a validatior percentage of 70.88%. Validation of the hypernym- hyponym relation for a pair of synsets is a strong indicator o: hypernym relation between them. However, the failure tc validate a synset pair is not a definitive indicator of erroneous construction and has to be treated as a flag for human inspection.
Table -1: Results of Hypernym Validation 5. Conclusion & Future scope Multiple Choice Questions (MCQs) are generated successfully. Efficient questions are produced with good quality distractors. The problem of manually creating questions is solved with the proposed system. The proposed system creates automated questions with the help of NLP that reduces human intervention and it is a cost and time- effective system. And the accuracy of the distractor generated is reasonably high. This system not only helps teachers with E-assessments but also helps students who are preparing for competitive exams. Students can test their ability to solve the questions and can also check their understanding of the concepts. For generating distractors, the wordnet approach is used We evaluate the quality of wordnet by examining the validity of its constituent synonyms and its hypernym-hyponym pairs [9]. The hypernymy validation procedure was testec on the set of all direct hypernyms for noun synsets in the Princeton WordNet (v2.1). There are a total of 79297 hypernym-hyponym pairs constituting this set. The number of pairs validated by the algorithm at different steps of the approach is summarized in Table 2. In all, the algorithm was able to validate 56203 out of 79297 noun hypernym relation pairs in the Princeton Wordnet, giving a validatior percentage of 70.88%. Validation of the hypernym- hyponym relation for a pair of synsets is a strong indicator o: hypernym relation between them. However, the failure tc validate a synset pair is not a definitive indicator of erroneous construction and has to be treated as a flag for human inspection.
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved