Aks-Nazar: Introducing a Persian-English Dataset for Multimodal Sentiment Analysis

In recent years, sentiment analysis research has primarily focused on textual data. With the increasing prevalence of multimedia content on social media platforms, multimodal sentiment analysis has gained greater significance. However, this field presents challenges for low-resource languages like Persian due to the lack of available resources. In this study, a dataset named Aks-Nazar is introduced, comprising both textual and visual samples categorized into two sentiment classes: positive and negative. Aks-Nazar, the largest multimodal sentiment analysis dataset for the Persian language, consists of ۱۰,۰۰۰ samples, each accompanied by precise English translations. The data were collected from Telegram and Twitter social media platforms and carefully annotated to ensure high-quality labeling. For data analysis, ResNet-۵۰ was used to extract visual features, while BiGRU, CNN, and BiLSTM models were employed for Persian text analysis. After feature extraction, a mid-level fusion approach was applied to combine textual and visual features, enhancing model performance. Evaluation results indicate that the multimodal model achieved a sentiment prediction accuracy of ۹۳.۳۹%. The Aks-Nazar dataset not only facilitates more precise sentiment analysis for Persian but also offers the potential for application in multilingual research, paving the way for more advanced studies in this field.

کلیدواژه ها:

Sentiment Analysis ، Data Mining ، Aks-Nazar Dataset ، Text Classification ، Image Classification