ellibsЭлектронные библиотекиRussian Digital Libraries Journal1562-5419Казанский (Приволжский) федеральный университет10.26907/1562-5419-2021-24-2-271-292ellibs-273Research ArticleСтатьиПрименение машинного обучения к задаче генерации поисковых запросовApplying Machine Learning to the Task of Generating Search QueriesГусенковА. М.GusenkovA. M.gusenkov.a.m@gmail.comСиттиковаА. Р.SittikovaA. R.sitti.alina@mail.ru202128042021242272293Copyright © Гусенков А.М., Ситтикова А.Р., 20212021Гусенков А.М., Ситтикова А.Р.Gusenkov A.M., Sittikova A.R.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://ellibs.elpub.ru/jour/article/view/273

Исследованы две модификации рекуррентных нейронных сетей: сети с долгой краткосрочной памятью и сети с управляемым рекуррентным блоком с добавлением механизма внимания к обеим сетям, а также модель Transformer в задаче генерации запросов к поисковым системам. В качестве модели Transformer использована модель GPT-2 от OpenAI, которая обучалась на запросах пользователей. Проведен латентно-семантический анализ для определения семантических сходств между корпусом пользовательских запросов и запросов, генерируемых нейронными сетями. Для проведения анализа корпус был переведен в формат bag of words, к нему применена модель TFIDF, проведено сингулярное разложение. Семантическое сходство вычислялось на основе косинусной меры. Также для более полной оценки применимости моделей к задаче был проведен экспертный анализ для оценки связности слов в искусственно созданных запросах.

In this paper we research two modifications of recurrent neural networks – Long Short-Term Memory networks and networks with Gated Recurrent Unit with the addition of an attention mechanism to both networks, as well as the Transformer model in the task of generating queries to search engines. GPT-2 by OpenAI was used as the Transformer, which was trained on user queries. Latent-semantic analysis was carried out to identify semantic similarities between the corpus of user queries and queries generated by neural networks. The corpus was convert-ed into a bag of words format, the TFIDF model was applied to it, and a singular value decomposition was performed. Semantic similarity was calculated based on the cosine measure. Also, for a more complete evaluation of the applicability of the models to the task, an expert analysis was carried out to assess the coherence of words in artificially created queries.

обработка естественного языкагенерация естественного языкамашинное обучениенейронные сетиnatural language processingnatural language generationmachine learningneural networksReferencesVan Deemter K., Krahmer E., Theune M. Real vs. template-based natural language generation: a false opposition? URL: https://wwwhome.ewi.utwente.nl/~theune/PUBS/templates-squib.pdfVan Deemter K., Krahmer E., Theune M. Real vs. template-based natural language generation: a false opposition? URL: https://wwwhome.ewi.utwente.nl/~theune/PUBS/templates-squib.pdfXie Z. Neural Text Generation: A Practical Guide.Xie Z. Neural Text Generation: A Practical Guide.URL: https://arxiv.org/pdf/1711.09534.pdfURL: https://arxiv.org/pdf/1711.09534.pdfA Comprehensive Guide to Natural Language Generation, 2019. URL: https://medium.com/sciforce/a-comprehensive-guide-to-natural-language-generation-dd63a4b6e548A Comprehensive Guide to Natural Language Generation, 2019. URL: https://medium.com/sciforce/a-comprehensive-guide-to-natural-language-generation-dd63a4b6e548Arrington M. AOL proudly releases massive amounts of user search data, 2006. URL: https://techcrunch.com/2006/08/06/aol-proudly-releases-massive-amounts-of-user-search-data/Arrington M. AOL proudly releases massive amounts of user search data, 2006. URL: https://techcrunch.com/2006/08/06/aol-proudly-releases-massive-amounts-of-user-search-data/Reiter E. NLG vs Templates: Levels of Sophistication in Generating Text, 2016. URL: https://ehudreiter.com/2016/12/18/nlg-vs-templatesReiter E. NLG vs Templates: Levels of Sophistication in Generating Text, 2016. URL: https://ehudreiter.com/2016/12/18/nlg-vs-templatesGagniuc P. Markov Chains: From Theory to Implementation and Experimentation, 2017. USA, NJ: John Wiley & Sons.Gagniuc P. Markov Chains: From Theory to Implementation and Experimentation, 2017. USA, NJ: John Wiley & Sons.Press O., Bar A., Bogin B., Berant J., Wolf L. Language Generation with Recurrent Generative Adversarial Networks without Pre-training. URL: https://arxiv.org/pdf/1706.01399.pdfPress O., Bar A., Bogin B., Berant J., Wolf L. Language Generation with Recurrent Generative Adversarial Networks without Pre-training. URL: https://arxiv.org/pdf/1706.01399.pdfWilliams R.J., Hinton G.E., Rumelhart D.E. Learning representations by back-propagating errors. URL: http://www.cs.utoronto.ca/~hinton/absps/naturebp.pdfWilliams R.J., Hinton G.E., Rumelhart D.E. Learning representations by back-propagating errors. URL: http://www.cs.utoronto.ca/~hinton/absps/naturebp.pdfHochreiter S., Bengio Y., Frasconi P., Schmidhuber J. Gradient Flow in Recurrent Nets: the Difficulty of Learning Long-Term Dependencies.Hochreiter S., Bengio Y., Frasconi P., Schmidhuber J. Gradient Flow in Recurrent Nets: the Difficulty of Learning Long-Term Dependencies.URL: https://www.bioinf.jku.at/publications/older/ch7.pdfURL: https://www.bioinf.jku.at/publications/older/ch7.pdfHochreiter S., Schmidhuber J. Long-Short Term Memory. URL: http://web.archive.org/web/20150526132154/http:// deeplearning.cs.cmu.edu/pdfs/Hochreiter97_lstm.pdfHochreiter S., Schmidhuber J. Long-Short Term Memory. URL: http://web.archive.org/web/20150526132154/http:// deeplearning.cs.cmu.edu/pdfs/Hochreiter97_lstm.pdfHeck J., Salem F. Simplified Minimal Gated Unit Variations for Recurrent Neural Networks. URL: https://arxiv.org/abs /1701.03452Heck J., Salem F. Simplified Minimal Gated Unit Variations for Recurrent Neural Networks. URL: https://arxiv.org/abs /1701.03452Bahdanau D., Cho K.m Bengio Y. Neural Machine Translation by Jointly Learning to Align and Translate. URL: https://arxiv.org/pdf/1409.0473.pdfBahdanau D., Cho K.m Bengio Y. Neural Machine Translation by Jointly Learning to Align and Translate. URL: https://arxiv.org/pdf/1409.0473.pdfFelbo B., Mislove A., Søgaard A., Rahwan I., Lehmann S. Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasm. URL: https://arxiv.org/pdf/1708.00524.pdfFelbo B., Mislove A., Søgaard A., Rahwan I., Lehmann S. Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasm. URL: https://arxiv.org/pdf/1708.00524.pdfBisong E. Google Colaboratory. In: Building Machine Learning and Deep Learning Models on Google Cloud Platform, 2019. Apress, Berkeley, CA.Bisong E. Google Colaboratory. In: Building Machine Learning and Deep Learning Models on Google Cloud Platform, 2019. Apress, Berkeley, CA.Chollet F. Keras, 2015. URL: https://keras.ioChollet F. Keras, 2015. URL: https://keras.ioKingma D., Ba J. Adam: A Method for Stochastic Optimization. URL: https://arxiv.org/abs/1412.6980Kingma D., Ba J. Adam: A Method for Stochastic Optimization. URL: https://arxiv.org/abs/1412.6980Learning Rate Scheduler. URL: https://keras.io/api/callbacks/learning_rate_ scheduler/Learning Rate Scheduler. URL: https://keras.io/api/callbacks/learning_rate_ scheduler/Schuster M., Paliwal K. Bidirectional recurrent neural networks. URL: https://www.researchgate.net/publication/ 3316656_Bidirectional_recurrent_neural_networksSchuster M., Paliwal K. Bidirectional recurrent neural networks. URL: https://www.researchgate.net/publication/ 3316656_Bidirectional_recurrent_neural_networksVaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A., Kaizer L., Polosukhin I. Attention Is All You Need. URL: https://arxiv.org/pdf/1706.03762.pdfVaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A., Kaizer L., Polosukhin I. Attention Is All You Need. URL: https://arxiv.org/pdf/1706.03762.pdfRadford A., Wu J., Child R., Luan D., Amodei D., Sutskever I. Language Models Are Unsupervised Multitask Learners. URL: https://d4mucfpksywv.cloudfront.net/better-language-models/language-models.pdfRadford A., Wu J., Child R., Luan D., Amodei D., Sutskever I. Language Models Are Unsupervised Multitask Learners. URL: https://d4mucfpksywv.cloudfront.net/better-language-models/language-models.pdfDevlin J., Chang M., Lee K., Toutanova K. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. URL: https://arxiv.org/pdf/1810.04805.pdfDevlin J., Chang M., Lee K., Toutanova K. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. URL: https://arxiv.org/pdf/1810.04805.pdfBrown T., Mann B., Ryder N., Subbiah M., Kaplan J. Language Models Are Few-Shot Learners. URL: https://arxiv.org/abs/2005.14165Brown T., Mann B., Ryder N., Subbiah M., Kaplan J. Language Models Are Few-Shot Learners. URL: https://arxiv.org/abs/2005.14165Gage P. A New Algorithm for Data Compression. URL: https://www.derczynski.com/papers/archive/BPE_Gage.pdfGage P. A New Algorithm for Data Compression. URL: https://www.derczynski.com/papers/archive/BPE_Gage.pdfDeerwester S., Harshman R. Indexing by Latent Semantic Analysis. URL: https://www.cs.bham.ac.uk/ ~pxt/IDA/lsa_ind.pdfDeerwester S., Harshman R. Indexing by Latent Semantic Analysis. URL: https://www.cs.bham.ac.uk/ ~pxt/IDA/lsa_ind.pdfNakov P. Getting Better Results with Latent Semantic Indexing. URL: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.59.6406&rep=rep1&type=pdfNakov P. Getting Better Results with Latent Semantic Indexing. URL: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.59.6406&rep=rep1&type=pdfRehurek R., Sojka P. Software Framework for Topic Modelling with Large Corpora // Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks. University of Malta. 2010.Rehurek R., Sojka P. Software Framework for Topic Modelling with Large Corpora // Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frameworks. University of Malta. 2010.

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