Generation of lay summaries for scientific articles based on neural networks
| Jazyk: | ruština |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
transformer аÑÑ Ð¸ÑекÑÑÑа
ÑиÑÑаÑоÑ-деÑиÑÑаÑÐ¾Ñ Ð¼Ð¾Ð´ÐµÐ»Ñ rouge авÑомаÑиÑеÑкое ÑеÑеÑиÑование пÑедваÑиÑелÑÐ½Ð°Ñ Ð¾Ð±ÑабоÑка даннÑÑ data pre-processing transformer architecture t5 automatic summarization encoder-decoder model |
| DOI: | 10.18720/spbpu/3/2021/vr/vr21-747 |
| Popis: | Рданной вÑпÑÑкной квалиÑикаÑионной ÑабоÑе опиÑан пÑоÑеÑÑ ÑазÑабоÑки ÑиÑÑÐµÐ¼Ñ Ð´Ð»Ñ Ð³ÐµÐ½ÐµÑаÑии неÑÐµÑ Ð½Ð¸ÑеÑкого ÑезÑме из наÑÑно -иÑÑледоваÑелÑÑÐºÐ¸Ñ ÑÑаÑей. РпеÑвом Ñазделе изложен Ð¾Ð±Ð·Ð¾Ñ Ð¿Ð¾Ð´Ñ Ð¾Ð´Ð¾Ð² в облаÑÑи обÑабоÑки еÑÑеÑÑвенного ÑзÑка. Ðа оÑнове вÑбÑанного Ð¿Ð¾Ð´Ñ Ð¾Ð´Ð° ÑаÑÑмаÑÑиваÑÑÑÑ Ð¿ÑеимÑÑеÑÑва ÑÑÑеÑÑвÑÑÑÐ¸Ñ Ð½ÐµÐ¹ÑоннÑÑ Ð¼Ð¾Ð´ÐµÐ»ÐµÐ¹ и вÑбиÑаеÑÑÑ Ð½Ð°Ð¸Ð±Ð¾Ð»ÐµÐµ Ð¿Ð¾Ð´Ñ Ð¾Ð´ÑÑÐ°Ñ Ð² ÑооÑвеÑÑÑвии Ñ Ð¿ÑиведеннÑми кÑиÑеÑиÑми. РпоÑледÑÑÑÐ¸Ñ ÑÐ°Ð·Ð´ÐµÐ»Ð°Ñ Ð¾Ð¿Ð¸ÑÑваеÑÑÑ Ð¿ÑоÑеÑÑ ÑазÑабоÑки ÑиÑÑÐµÐ¼Ñ Ð´Ð»Ñ ÑеÑÐµÐ½Ð¸Ñ Ð¿Ð¾ÑÑавленной задаÑи. ÐÑиводиÑÑÑ Ð°ÑÑ Ð¸ÑекÑÑÑа вÑбÑанной модели, иÑполÑзÑемÑе алгоÑиÑÐ¼Ñ Ð¾Ð±ÑÑÐµÐ½Ð¸Ñ Ð¸ ÑпоÑÐ¾Ð±Ñ Ð¿ÑедобÑабоÑки даннÑÑ . Ðалее обоÑновÑваеÑÑÑ Ð²ÑÐ±Ð¾Ñ Ð±Ð¸Ð±Ð»Ð¸Ð¾Ñек и опиÑÑваÑÑÑÑ ÑÑÐ°Ð¿Ñ ÑÑÑановки и наÑÑÑойки ÑÑедÑ. РпоÑледнем Ñазделе пÑиводиÑÑÑ Ð¾Ð¿Ð¸Ñание меÑодик оÑÐµÐ½Ð¸Ð²Ð°Ð½Ð¸Ñ Ð¸ пÑоÑеÑÑа Ð¸Ð·Ð¼ÐµÐ½ÐµÐ½Ð¸Ñ Ð³Ð¸Ð¿ÐµÑпаÑамеÑÑов в модели. РконÑе Ñаздела пÑÐ¸Ð²ÐµÐ´ÐµÐ½Ñ ÑгенеÑиÑованнÑе ÑезÑме и вÑÐ²Ð¾Ð´Ñ ÑделаннÑе, на оÑновании оÑенок, полÑÑеннÑÑ Ð¾Ñ ÑеÑензенÑов. This final qualification paper describes the process of developing a system for generating lay summaries from scientific articles. The first section provides an overview of approaches to natural language processing. Based on the chosen approach, the advantages of existing neural models are considered and the most suitable one is selected in accordance with the specified criteria. The following sections describe the process of developing a system to solve this problem. The architecture of the selected model, the training algorithms used, and the data preprocessing methods are described. At the same time, the selected libraries are explained and the steps for installing and configuring the environment are described. The last section describes the evaluation methods and the process of changing the hyperparameters of the model. At the end of the section, the generated lay summaries and conclusions based on the ratings received from the reviewers are presented. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|