Data Mining
A. Beiranvand; M. Sarhadi; J. Salimi Sartakhti
Abstract
Background and Objectives: Large Language Models have demonstrated exceptional performance across various NLP tasks, especially when fine-tuned for specific applications. Full fine-tuning of large language models requires extensive computational resources, which are often ...
Read More
Background and Objectives: Large Language Models have demonstrated exceptional performance across various NLP tasks, especially when fine-tuned for specific applications. Full fine-tuning of large language models requires extensive computational resources, which are often unavailable in real-world settings. While Low-Rank Adaptation (LoRA) has emerged as a promising solution to mitigate these challenges, its potential remains largely untapped in multi-task scenarios. This study addresses this gap by introducing a novel hybrid approach that combines LoRA with an attention-based mechanism, enabling fine-tuning across tasks while facilitating knowledge sharing to improve generalization and efficiency. This study aims to address this gap by introducing a novel hybrid fine-tuning approach using LoRA for multi-task text classification, with a focus on inter-task knowledge sharing to enhance overall model performance.Methods: We proposed a hybrid fine-tuning method that utilizes LoRA to fine-tune LLMs across multiple tasks simultaneously. By employing an attention mechanism, this approach integrates outputs from various task-specific models, facilitating cross-task knowledge sharing. The attention layer dynamically prioritizes relevant information from different tasks, enabling the model to benefit from complementary insights. Results: The hybrid fine-tuning approach demonstrated significant improvements in accuracy across multiple text classification tasks. On different NLP tasks, the model showed superior generalization and precision compared to conventional single-task LoRA fine-tuning. Additionally, the model exhibited better scalability and computational efficiency, as it required fewer resources to achieve comparable or better performance. Cross-task knowledge sharing through the attention mechanism was found to be a critical factor in achieving these performance gains.Conclusion: The proposed hybrid fine-tuning method enhances the accuracy and efficiency of LLMs in multi-task settings by enabling effective knowledge sharing between tasks. This approach offers a scalable and resource-efficient solution for real-world applications requiring multi-task learning, paving the way for more robust and generalized NLP models.
Data Mining
R. Asgarnezhad; A. Monadjemi; M. SoltanAghaei
Abstract
Background and Objectives: With the extensive web applications, review sentiment classification has attracted increasing interest among text mining works. Traditional approaches did not indicate multiple relationships connecting words while emphasizing the preprocessing phase and data reduction techniques, ...
Read More
Background and Objectives: With the extensive web applications, review sentiment classification has attracted increasing interest among text mining works. Traditional approaches did not indicate multiple relationships connecting words while emphasizing the preprocessing phase and data reduction techniques, making a huge performance difference in classification. Methods: This study suggests a model as an efficient model for sentiment classification combining preprocessing techniques, sampling methods, feature selection methods, and ensemble supervised classification to increase the classification performance. In the feature selection phase of the proposed model, we applied n-grams, which is a computational method, to optimize the feature selection procedure by extracting features based on the relationships of the words. Then, the best-selected feature through the particle swarm optimization algorithm to optimize the feature selection procedure by iteratively trying to improve feature selection. Results: In the experimental study, a comprehensive range of comparative experiments conducted to assess the effectiveness of the proposed model using the best in the literature on Twitter datasets. The highest performance of the proposed model obtains 97.33, 92.61, 97.16, and 96.23% in terms of precision, accuracy, recall, and f-measure, respectively.Conclusion: The proposed model classifies the sentiment of tweets and online reviews through ensemble methods. Besides, two sampling techniques had applied in the preprocessing phase. The results confirmed the superiority of the proposed model over state-of-the-art systems.
Data Mining
R. Asgarnezhad; A. Monadjemi; M. SoltanAghaei
Abstract
Background and Objectives: Twitter Sentiment Classification is one of the most popular fields in information retrieval and text mining. Millions of people of the world intensity use social networks like Twitter. It supports users to publish tweets to tell what they are thinking about topics. There are ...
Read More
Background and Objectives: Twitter Sentiment Classification is one of the most popular fields in information retrieval and text mining. Millions of people of the world intensity use social networks like Twitter. It supports users to publish tweets to tell what they are thinking about topics. There are numerous web sites built on the Internet presenting Twitter. The user can enter a sentiment target and seek for tweets containing positive, negative, or neutral opinions. This is remarkable for consumers to investigate the products before purchase automatically.Methods: This paper suggests a model for sentiment classification. The goal of this model is to investigate what is the role of n-grams and sampling techniques in Sentiment Classification application using an ensemble method on Twitter datasets. Also, it examines both binary and multiple classifications, which are classified datasets into positive, negative, or neutral classes.Results: Twitter Classification is an outstanding problem, which has very few free resources and not available due to modified authorization status. However, all Twitter datasets are not labeled and free, except for our applied dataset. We reveal that the combination of ensemble methods, sampling techniques, and n-grams can improve the accuracy of Twitter Sentiment Classification.Conclusion: The results confirmed the superiority of the proposed model over state-of-the-art systems. The highest results obtained in terms of accuracy, precision, recall, and f-measure.