Institutional Repository of Vilnius University
Pokalbių roboto naudotojo ketinimų atpažinimas naudojant natūralios kalbos apdorojimą ir dirbtinius neuroninius tinklus /
Abstract
dc:descriptionThis paper addresses the task of recognizing user intents in a conversation chatbot using the Lithuanian language. It aims for the best results in order to try to develop natural language understanding areas for the Lithuanian language. To solve the problem of intention recognition, 3 models were created and compared, differing from each other in the vectorization methods used. To create the models, a dataset of intentions, consisting of everyday Lithuanian phrases, was first created. Then, preprocessing of the text was performed using natural language processing methods. Three different vectorization methods were used and compared: Bag of Words, TF-IDF, and Word2Vec. Training matrices were created for each model, and neural network architectures were developed and optimized using a hyperparameter optimization algorithm. The evaluation of the models' quality showed that the model based on the Word2Vec vectorization method provides the best results - when tested with a dedicated test dataset, it achieved an accuracy of 78.57%.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zaranka, Paulius,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192843298&prefLang=en_US