Please use this identifier to cite or link to this item: http://dspace.dtu.ac.in:8080/jspui/handle/repository/20808
Full metadata record
DC FieldValueLanguage
dc.contributor.authorBANSAL, AKASH-
dc.date.accessioned2024-08-05T08:57:22Z-
dc.date.available2024-08-05T08:57:22Z-
dc.date.issued2024-05-
dc.identifier.urihttp://dspace.dtu.ac.in:8080/jspui/handle/repository/20808-
dc.description.abstractIn this study, we introduce a approach to emotional classification of suicide note sentences. In our model, features are extract by CNN layers, sequential dependencies are captured using BiLSTM layers, and tokens are embedded using BERT. By playing to the strengths of each element, this strategy maximizes classification accuracy while conveying the text's subtle emotional connotations. The mode used is BERT which has rich, sensitive embeddings by providing tokenization to implement dense vector representations of tokens that take into account both the left and right word directions. Following the processing of these embeddings, a BiLSTM layer scans the sequences both like in backward direction as well as forward in order to identify long-term dependencies within the text. Through the bidirectional process, contextual information from past and future tokens is added to and optimized for the current token. From the concatenated states it obtained from the Bi-LSTM layer, the CNN layer extracts local patterns and hierarchical features using convolutional filers and other dense layers. The CNN removes important part which is more about textual, improving the model's recognition of intricate emotional cues.Long text sequences are eliminated by the CNN, which enhances the model's ability to recognize complex emotional cues. A neural network layer that creates a distribution over the six emotional classes that are proud, happy, sad, neutral,love, and hate and classify them . To not overfitting, the model is trained using an optimization method in conjunction with regularization, backpropagation, and categorical cross-entropy loss. Test set assessment is a amazing technique for increase the model's efficacy, and calculations like we have precision , recall and F1 score offer significant metrics into the model's working. We find that our hybrid model performs remarkably well at reliably classifying emotions in delicate textual input.en_US
dc.language.isoenen_US
dc.relation.ispartofseriesTD-7331;-
dc.subjectSENTIMENT CLASSIFICATIONen_US
dc.subjectSUICIDE NOTESen_US
dc.subjectDEEP LEARNINGen_US
dc.subjectBiLSTMen_US
dc.subjectBERTen_US
dc.subjectCNNen_US
dc.titleSENTIMENT CLASSIFICATION ON SUICIDE NOTES USING DEEP LEARNINGen_US
dc.typeThesisen_US
Appears in Collections:M.E./M.Tech. Computer Engineering

Files in This Item:
File Description SizeFormat 
AKASH BANSAL M.Tech.pdf1.54 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.