TY  - JOUR
AU  - Kumar, Buddha Hari 
AU  - Perumal, Chitra 
AU  - Raja, Inakoti Ramesh 
AU  - Babu, Chukka Ramesh 
AU  - Gorre, Srinivas Rao 
AU  - Tripurana, Santosh 
PY  - 2025
TI  - A Deep Learning Approach for Telugu Domain Identification with Multichannel LSTM-CNN
JF  - Journal of Computer Science
VL  - 21
IS  - 9
DO  - 10.3844/jcssp.2025.2181.2190
UR  - https://thescipub.com/abstract/jcssp.2025.2181.2190
AB  - The vast growth of textual data has ushered into the limelight, a plethora of applications in information retrieval and natural language processing (NLP). Proper extraction of information from text is heavily dependent on recognizing the thematic content, which becomes crucial in the tasks of document summarization, information extraction, question answering, machine translation, and sentiment analysis. The great complexity of this challenge arises for regional languages such as Telugu, where unique linguistic features demand specialized approaches. In this work, we propose a Telugu Technical Domain Identification model based on a Multichannel Long Short-Term Memory Convolutional Neural Network (LSTM-CNN) architecture. This methodology benefits from the sequential data treatment capabilities of LSTM combined with the local feature extractive powers of CNN, which enable effective domain identification in Telugu texts. The model was assessed at the ICON Shared Challenge "TechDOfication 2020," scoring an F1 score of 90.01% on the validation set and 69.90% on the test set. The results indicate a great improvement over conventional models and show the tremendous efficacy of multichannel deep learning techniques for domain identification in Telugu. The proposed model will serve as a milestone toward enhancing NLP applications for regional languages while providing a scalable solution to the heightened demands for accurate thematic classification of techno-domain risks.