20.500.12556/DKUM-84914 Interpretacija odločitev nevronske mreže z metodo lime diplomsko delo Interpretation of neural network decisions using the lime method V diplomskem delu interpretiramo odločitve globokih nevronskih mrež za analizo sentimenta slovenskih in angleških besedil z metodo LIME. Nevronske mreže veljajo za črne škatle, saj je težko pojasniti njihove odločitve s samimi utežmi, kar predstavlja težave pri posvojitvi takšnih modelov v situacijah, kjer je pomembna transparentnost in zanesljivost odločitev. Problem črnih škatel lahko rešimo z metodami interpretiranja, kot je LIME, ki pojasni odločitev črne škatle z lokalnim nadomestkom. V tem delu interpretiramo napovedi treh nevronskih mrež z metodo LIME in poskušamo na podlagi interpretacij razložiti delovanje nevronskih mrež. In this thesis, we interpret the decisions of deep neural networks for sentiment analysis of Slovenian and English texts using the LIME method. Neural networks are considered black boxes because it is difficult to explain their decisions based solely on the weights, which poses challenges for adopting such models in situations where transparency and reliability of decisions are important. The problem of black boxes can be addressed with interpretation methods such as LIME, which explains the decision of a black box with a local surrogate. In this work, we interpret the predictions of three neural networks using the LIME method and attempt to explain the functioning of neural networks based on these interpretations. LIME razložljiva umetna inteligenca nevronska mreža analiza sentimenta LIME interpretable artificial intelligence neural network sentiment analysis true true false [K. Dajčman] Slovenski jezik Angleški jezik Diplomsko delo/naloga 2023-08-08 14:47:42 2023-09-21 09:38:54 2023-10-21 03:05:42 0000-00-00 00:00:00 2023 0 Maribor 0 Maribor 1 spletni vir (1 datoteka PDF (XIII, 46 f.)) 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.8(043.2) 169212931 UN_Dajcman_Kristijan_2023.pdf UN_Dajcman_Kristijan_2023.pdf 1 3AF0C83C4A174D077D76D81B1599EEFE 9e320ec198493636cdf9aff2cdfcc102ba37c66db92dbd2eb8355e489dc222d7 c337d41a-432d-11ee-b781-00155d000105 https://dk.um.si/Dokument.php?lang=slv&id=170373 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0