ntvНаучно-технический вестник информационных технологий, механики и оптикиScientific and Technical Journal of Information Technologies, Mechanics and Optics2226-14942500-0373Университет ИТМО10.17586/2226-1494-2023-23-4-743-749ntv-204Research ArticleКОМПЬЮТЕРНЫЕ СИСТЕМЫ И ИНФОРМАЦИОННЫЕ ТЕХНОЛОГИИCOMPUTER SCIENCEМетод аугментации текстовых данных с сохранением стиля речи и лексики персоныText augmentation preserving persona speech style and vocabularyhttps://orcid.org/0000-0002-2935-991XМатвееваА. А.MatveevaA. A.

Матвеева Анастасия Андреевна — инженер

sc 57204215042

Санкт-Петербург, 197101

Anastasia A. Matveeva — Engineer

sc 57204215042

Saint Petersburg, 197101

anastasiamatveevaitmo@gmail.comhttps://orcid.org/0000-0002-8992-9654МахныткинаО. В.MakhnytkinaO. V.

Махныткина Олеся Владимировна — кандидат технических наук, доцент

sc 57208002090

Санкт-Петербург, 197101

Olesia V. Makhnytkina — PhD, Associate Professor

sc 57208002090

Saint Petersburg, 197101

makhnytkina@itmo.ruУниверситет ИТМОРоссияITMO UniversityRussian Federation202314122024234743749Copyright © Матвеева А.А., Махныткина О.В., 20242024Матвеева А.А., Махныткина О.В.Matveeva A.A., Makhnytkina O.V.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://ntv.elpub.ru/jour/article/view/204

Введение. В настоящее время часто для различных задач обработки естественного языка требуются большие наборы данных. Однако для многих задач сбор большого набора данных является трудоемким, дорогостоящим процессом и требует привлечения экспертов. Увеличение объема информации возможно достичь с использованием методов аугментации данных. Вместе с тем использование классических подходов может привести к включению в корпус данных фраз, которые отличаются по стилю речи и лексикону целевой персоны. Это сопровождается изменением целевого класса и появлением реплик с неестественным использованием лексики и отсутствием смысла. Предложен новый метод аугментации тестовых данных, учитывающий стиль и лексикон персоны. Метод. В работе разработан новый метод аугментации текстовых данных, сохраняющий индивидуальные речевые характеристики и словарный запас. Основная идея метода заключается в формировании индивидуальных шаблонов для каждого человека на основе анализа синтаксических деревьев высказываний и последующего создания новых реплик по сформированным шаблонам. Основные результаты. Метод апробирован на задаче оценки эмоционального состояния пользователя в диалоге. Исследования проведены для датасетов на английском и русском языках. Предложенный метод позволил повысить качество решения этих задач как для английского, так и для русского языков. Получено увеличений значений метрик accuracy и weighted F1 для разных моделей. Обсуждение. Результаты работы могут быть применены для повышения метрик accuracy и weighted F1 моделей, предназначенных для решения различных задач для английского и русского языков.

Currently, various natural language processing tasks often require large data sets. However, for many tasks, collecting large datasets is quite tedious and expensive, and requires the involvement of experts. An increase in the amount of data can be achieved using methods of data augmentation, however, the use of classical approaches can lead to the inclusion of phrases in the data corpus that differ in the speech style and vocabulary of the target person, which can lead to both a change in the target class as well as the appearance of replicas with unnatural vocabulary use and lack of meaning. In this context, a new method for test data enrichment is proposed that takes into account the person’s style and vocabulary. In this article, a new method for expanding text data that preserves individual language features and vocabulary is proposed. The core of the method is to create individual templates for each person based on the analysis of syntactic trees of propositions and then to create new replicas according to the generated templates. The method was tested on the task of assessing the user’s emotional state in a dialogue. The search was carried out for data sets in English and Russian. The proposed method made it possible to improve the quality of solving these problems for both the English and Russian languages. Up to a 2 % increase in accuracy and weighted F1 metrics has been noted for various models. The results of the work can be applied to improve the accuracy and weighted F1 metrics of models designed to solve various problems for the English and Russian languages.

аугментация текстовых данныхраспознавание эмоцийоценка валентности высказыванийtext data augmentationemotion recognitionstatement valence evaluationИсследование выполнено за счет гранта Российского научного фонда (№ 22-11-00128, https://www.rscf.ru/ project/22-11-00128/).This research was supported by a grant from the Russian Science Foundation (22-11-00128 https://www.rscf.ru/project/22-11-00128/).ReferencesGiridhara P.K., Mishra C., Venkataramana R.K., Bukhari S.S., Dengel A.R. A study of various text augmentation techniques for relation classifcation in free text // Proc. of the 8th International Conference on Pattern Recognition Applications and Methods. 2019. P. 360–367 https://doi.org/10.5220/0007311003600367Giridhara P.K., Mishra C., Venkataramana R.K., Bukhari S.S., Dengel A.R. A study of various text augmentation techniques for relation classifcation in free text // Proc. of the 8th International Conference on Pattern Recognition Applications and Methods. 2019. P. 360–367 https://doi.org/10.5220/0007311003600367Papadaki M. Data Augmentation Techniques for Legal Text Analytics: A thesis submitted to Athens University of Economics and Business in fulfllment of the requirements for the degree of Master in Data Science. 2017. 33 p.Papadaki M. Data Augmentation Techniques for Legal Text Analytics: A thesis submitted to Athens University of Economics and Business in fulfllment of the requirements for the degree of Master in Data Science. 2017. 33 p.Zhang Z., Zweigenbaum P. GNEG: Graph-based negative sampling for word2vec // Proc. of the 56th Annual Meeting of the Association for Computational Linguistics. V. 2. 2018. P. 566–571. https://doi.org/10.18653/v1/P18-2090Zhang Z., Zweigenbaum P. GNEG: Graph-based negative sampling for word2vec // Proc. of the 56th Annual Meeting of the Association for Computational Linguistics. V. 2. 2018. P. 566–571. https://doi.org/10.18653/v1/P18-2090Wei J., Zou K. EDA: Easy data augmentation techniques for boosting performance on text classifcation tasks // Proc. of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2018. P. 6382–6388. https://doi.org/10.18653/v1/D19-1670Wei J., Zou K. EDA: Easy data augmentation techniques for boosting performance on text classifcation tasks // Proc. of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2018. P. 6382–6388. https://doi.org/10.18653/v1/D19-1670Wu X., Xia Y., Zhu J., Wu L., Xie S., Fan Y., Qin T. mixSeq: A simple data augmentation method for neural machine translation // Proc. of the 18th International Conference on Spoken Language Translation (IWSLT 2021). 2021. P. 192–197. https://doi.org/10.18653/v1/2021.iwslt-1.23Wu X., Xia Y., Zhu J., Wu L., Xie S., Fan Y., Qin T. mixSeq: A simple data augmentation method for neural machine translation // Proc. of the 18th International Conference on Spoken Language Translation (IWSLT 2021). 2021. P. 192–197. https://doi.org/10.18653/v1/2021.iwslt-1.23Kumar V., Choudhary A., Cho E. Data augmentation using pre-trained transformer models // Proc. of the 2nd Workshop on Life-long Learning for Spoken Language Systems. 2020. P. 18–26.Kumar V., Choudhary A., Cho E. Data augmentation using pre-trained transformer models // Proc. of the 2nd Workshop on Life-long Learning for Spoken Language Systems. 2020. P. 18–26.Kobayashi S. Contextual augmentation: Data augmentation by words with paradigmatic relations // Proc. of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers). 2018. P. 452–457. https://doi.org/10.18653/v1/N18-2072Kobayashi S. Contextual augmentation: Data augmentation by words with paradigmatic relations // Proc. of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers). 2018. P. 452–457. https://doi.org/10.18653/v1/N18-2072Yu A., Dohan D., Luong M., Zhao R., Chen K., Norouzi M., Le Q. QANet: Combining local convolution with global self-attention for reading comprehension // Proc. of the ICLR Conference. 2018.Yu A., Dohan D., Luong M., Zhao R., Chen K., Norouzi M., Le Q. QANet: Combining local convolution with global self-attention for reading comprehension // Proc. of the ICLR Conference. 2018.Mehdi R., Meyer M., Goutal S. Text Data Augmentation: Towards better detection of spear-phishing emails // arXiv. 2020. arXiv:2007.02033. https://doi.org/10.48550/arXiv.2007.02033Mehdi R., Meyer M., Goutal S. Text Data Augmentation: Towards better detection of spear-phishing emails // arXiv. 2020. arXiv:2007.02033. https://doi.org/10.48550/arXiv.2007.02033Edunov S., Ott M., Auli M., Grangier D. Understanding backtranslation at scale // Proc. of the 2018 Conference on Empirical Methods in Natural Language Processing. 2018. P. 489–500. https://doi.org/10.18653/v1/D18-1045Edunov S., Ott M., Auli M., Grangier D. Understanding backtranslation at scale // Proc. of the 2018 Conference on Empirical Methods in Natural Language Processing. 2018. P. 489–500. https://doi.org/10.18653/v1/D18-1045Guo H., Mao Y., Zhang R. Augmenting data with mixup for sentence classifcation: An empirical study // arXiv. 2019. arXiv:1905.08941. https://doi.org/10.48550/arXiv.1905.08941Guo H., Mao Y., Zhang R. Augmenting data with mixup for sentence classifcation: An empirical study // arXiv. 2019. arXiv:1905.08941. https://doi.org/10.48550/arXiv.1905.08941Coulombe C. Text data augmentation made simple by leveraging NLP cloud APIs // arXiv. 2018. arXiv:1812.04718. https://doi.org/10.48550/arXiv.1812.04718Coulombe C. Text data augmentation made simple by leveraging NLP cloud APIs // arXiv. 2018. arXiv:1812.04718. https://doi.org/10.48550/arXiv.1812.04718Shen T., Lei T., Barzilay R., Jaakkola T. Style transfer from nonparallel text by cross-alignment // Advances in Neural Information Processing Systems. 2017. V. 30.Shen T., Lei T., Barzilay R., Jaakkola T. Style transfer from nonparallel text by cross-alignment // Advances in Neural Information Processing Systems. 2017. V. 30.Yang S., Huang X., Lau J.H., Erfani S. Robust task-oriented dialogue generation with contrastive pre-training and adversarial fltering // Findings of the Association for Computational Linguistics (EMNLP 2022). 2022. P. 1220–1234.Yang S., Huang X., Lau J.H., Erfani S. Robust task-oriented dialogue generation with contrastive pre-training and adversarial fltering // Findings of the Association for Computational Linguistics (EMNLP 2022). 2022. P. 1220–1234.Kovriguina L., Shilin I., Shipilo A., Putintseva A. Russian tagging and dependency parsing models for stanford CoreNLP natural language toolkit // Communications in Computer and Information Science. 2017. V. 786. P. 101–111. https://doi.org/10.1007/978-3-319-69548-8_8Kovriguina L., Shilin I., Shipilo A., Putintseva A. Russian tagging and dependency parsing models for stanford CoreNLP natural language toolkit // Communications in Computer and Information Science. 2017. V. 786. P. 101–111. https://doi.org/10.1007/978-3-319-69548-8_8Matveev Y., Matveev A., Frolova O., Lyakso E., Ruban N. Automatic speech emotion recognition of younger school age children // Mathematics. 2022. V. 10. N 14. P. 2373. https://doi.org/10.3390/math10142373Matveev Y., Matveev A., Frolova O., Lyakso E., Ruban N. Automatic speech emotion recognition of younger school age children // Mathematics. 2022. V. 10. N 14. P. 2373. https://doi.org/10.3390/math10142373Lyakso E., Frolova O., Matveev A., Matveev Y., Grigorev A., Makhnytkina O., Ruban N. Recognition of the emotional state of children with down syndrome by video, audio and text modalities: human and automatic // Lecture Notes in Computer Science. 2022. V. 13721. P. 438–450. https://doi.org/10.1007/978-3-031-20980-2_38Lyakso E., Frolova O., Matveev A., Matveev Y., Grigorev A., Makhnytkina O., Ruban N. Recognition of the emotional state of children with down syndrome by video, audio and text modalities: human and automatic // Lecture Notes in Computer Science. 2022. V. 13721. P. 438–450. https://doi.org/10.1007/978-3-031-20980-2_38Kim T., Vossen P. EmoBERTa: Speaker-Aware Emotion Recognition in Conversation with RoBERTa // arXiv. 2021. arXiv:2108.12009. https://doi.org/10.48550/arXiv.2108.12009Kim T., Vossen P. EmoBERTa: Speaker-Aware Emotion Recognition in Conversation with RoBERTa // arXiv. 2021. arXiv:2108.12009. https://doi.org/10.48550/arXiv.2108.12009Song X., Zang L., Zhang R., Hu S., Huang L. Emotionfow: Capture the dialogue level emotion transitions // Proc. of the ICASSP 2022– 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2022. P. 8542–8546. https://doi.org/10.1109/ICASSP43922.2022.9746464Song X., Zang L., Zhang R., Hu S., Huang L. Emotionfow: Capture the dialogue level emotion transitions // Proc. of the ICASSP 2022– 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). 2022. P. 8542–8546. https://doi.org/10.1109/ICASSP43922.2022.9746464Shen W., Chen J., Quan X., Xie Z. DialogXL: All-in-One XLNet for multi-party conversation emotion recognition // Proceedings of the AAAI Conference on Artifcial Intelligence. 2021. V. 35. N 15. P. 13789–13797 https://doi.org/10.1609/aaai.v35i15.17625Shen W., Chen J., Quan X., Xie Z. DialogXL: All-in-One XLNet for multi-party conversation emotion recognition // Proceedings of the AAAI Conference on Artifcial Intelligence. 2021. V. 35. N 15. P. 13789–13797 https://doi.org/10.1609/aaai.v35i15.17625Shen W., Wu S., Yang Y., Quan X. Directed acyclic graph network for conversational emotion recognition // Proc. of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021. P. 1551–1560.Shen W., Wu S., Yang Y., Quan X. Directed acyclic graph network for conversational emotion recognition // Proc. of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021. P. 1551–1560.

The authors declare that there are no conflicts of interest present.