TY - GEN
T1 - Generating Student Progress Reports based on Keywords
AU - Kobashi, Shumpei
AU - Mine, Tsunenori
N1 - Funding Information:
This work was supported in part by e-sia Corporation and by Grant-in-Aid for Scientific Research proposal numbers (JP21H00907, JP20H01728, JP20H04300, JP19KK0257, JP18K18656). We would like to express our deepest gratitude to them.
Publisher Copyright:
© 2021 29th International Conference on Computers in Education Conference, ICCE 2021 - Proceedings. All rights reserved
PY - 2021/11/22
Y1 - 2021/11/22
N2 - In this paper, we propose a method that automatically generates a student learning status report based on keywords given by instructors at cram schools to reduce their burden on writing the report. For selecting sentences to generate the report, we propose two methods: Seq2Seq-based and Information Retrieval (IR)-based methods. The Seq2Seq-based method uses a Seq2Seq model to generate sentences using keywords given by the instructors. The IR-based method uses OkapiBM25 to select sentences from those written by the instructors based on the keywords. We conducted extensive experiments to evaluate the two methods on a test set of 197,493 sentences. The experimental results show that the Seq2Seq method generates more suitable sentences as the report than the IR-based method. Adding the attention mechanism to the Seq2Seq method further improved the performance of the Seq2Seq method. Considering the above experimental results, we discussed the generation of the lecturer report by keywords.
AB - In this paper, we propose a method that automatically generates a student learning status report based on keywords given by instructors at cram schools to reduce their burden on writing the report. For selecting sentences to generate the report, we propose two methods: Seq2Seq-based and Information Retrieval (IR)-based methods. The Seq2Seq-based method uses a Seq2Seq model to generate sentences using keywords given by the instructors. The IR-based method uses OkapiBM25 to select sentences from those written by the instructors based on the keywords. We conducted extensive experiments to evaluate the two methods on a test set of 197,493 sentences. The experimental results show that the Seq2Seq method generates more suitable sentences as the report than the IR-based method. Adding the attention mechanism to the Seq2Seq method further improved the performance of the Seq2Seq method. Considering the above experimental results, we discussed the generation of the lecturer report by keywords.
UR - http://www.scopus.com/inward/record.url?scp=85126598943&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85126598943&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85126598943
T3 - 29th International Conference on Computers in Education Conference, ICCE 2021 - Proceedings
SP - 75
EP - 80
BT - 29th International Conference on Computers in Education Conference, ICCE 2021 - Proceedings
A2 - Rodrigo, Maria Mercedes T.
A2 - Iyer, Sridhar
A2 - Mitrovic, Antonija
A2 - Cheng, Hercy N. H.
A2 - Kohen-Vacs, Dan
A2 - Matuk, Camillia
A2 - Palalas, Agnieszka
A2 - Rajenran, Ramkumar
A2 - Seta, Kazuhisa
A2 - Wang, Jingyun
PB - Asia-Pacific Society for Computers in Education
T2 - 29th International Conference on Computers in Education Conference, ICCE 2021
Y2 - 22 November 2021 through 26 November 2021
ER -