Semi-supervised deep learning for the classification of eldercare workers’ sentiments
Tekstin luokitteluun on olemassa laaja tutkimuksen kirjo, mutta vain osa siitä on puoliohjattujen syvien neuroverkkojen pohjalta tehtyä – etenkin, kun opetusaineisto on ollut englannin kielellä, tai muulla huomattavan paljon tutkitulla kielellä. Tässä pro gradussa käymme läpi puoliohjattujen syväoppimismenetelmien kirjallisuutta tekstin luokittelussa, ja luomme käytännön toteutuksen kolmelle puoliohjatulle tekstin luokittelumenetelmälle. Nämä menetelmät opetetaan ja testataan pienenpuoleisella, suomenkielisellä aineistolla. Tulosten perusteella voitaisiin sanoa, että puoliohjattujen menetelmien yhteydessä on kannattavaa käyttää regularisointimenetelmiä ylisovittumisen ehkäisemiseksi, varsinkin kun opetusaineisto on pieni. Jotta voitaisiin saada kokonaisvaltaisempi kuva eri puoliohjattujen menetelmien kannattavuudesta ja luotettavuudesta luonnollisen kielen luokittelutehtävässä, olisi suomenkielisistä syväoppimismalleista ja regularisoinnista hyvä tehdä lisää tutkimusta. There exists extensive research for text classification, but only a handful of it is put into practice by deep neural networks that use semi-supervised learning – especially when semi-supervised deep neural networks are not trained in English, or other majorly studied languages. In this thesis we go through previous literature regarding semi-supervised deep learning methods for text classification, and then build a hands-on solution for three semi-supervised text classification methods. These methods are trained and tested on a small dataset, that is in Finnish. The results suggest that regularization methods should be taken into consideration when using semi-supervised methods for training – particularly when using smaller datasets that easily leads to overfitting. More research on regularization and Finnish deep learning models should be conducted to have a more comprehensive view on the applicability and reliability of text classification in natural language processing.
Keywords
Metadata
Show full item recordCollections
- Pro gradu -tutkielmat [29758]
License
Related items
Showing items with similar title or keywords.
-
The Truth is Out There : Focusing on Smaller to Guess Bigger in Image Classification
Terziyan, Vagan; Kaikova, Olena; Malyk, Diana; Branytskyi, Vladyslav (Elsevier, 2023)In Artificial Intelligence (AI) in general and in Machine Learning (ML) in particular, which are important and integral components of modern Industry 4.0, we often deal with uncertainty, e.g., lack of complete information ... -
Problem Transformation Methods with Distance-Based Learning for Multi-Target Regression
Hämäläinen, Joonas; Kärkkäinen, Tommi (ESANN, 2020)Multi-target regression is a special subset of supervised machine learning problems. Problem transformation methods are used in the field to improve the performance of basic methods. The purpose of this article is to test ... -
Automatic identification of architecture and endianness using binary file contents
Kairajärvi, Sami (2019)This thesis explores how architecture and endianness of executable code can be identified using binary file contents, as falsely identifying the architecture caused about 10% of failures of firmware analysis in a recent ... -
Artificial intelligence in the cyber security environment
Vähäkainu, Petri; Lehto, Martti (Academic Conferences International, 2019)Artificial Intelligence (AI) is intelligence exhibited by machines. Any system that perceives its environment and takes actions that maximize its chance of success at some goal may be defined as AI. The family of AI ... -
Do Randomized Algorithms Improve the Efficiency of Minimal Learning Machine?
Linja, Joakim; Hämäläinen, Joonas; Nieminen, Paavo; Kärkkäinen, Tommi (MDPI AG, 2020)Minimal Learning Machine (MLM) is a recently popularized supervised learning method, which is composed of distance-regression and multilateration steps. The computational complexity of MLM is dominated by the solution of ...