A Comparative Evaluation of Speech-Text Effective Approaches for Arabic Sentiment Analysis
In recent years, the pervasive use of social media platforms has elevated the significance of users' emotions, ranging from satisfaction to anxiety, within digital discourse. This trend is particularly pronounced in the Arabic-speaking digital landscape, where the language holds a prominent pre...
محفوظ في:
| المؤلف الرئيسي: | |
|---|---|
| منشور في: |
2024
|
| الموضوعات: | |
| الوصول للمادة أونلاين: | https://bspace.buid.ac.ae/handle/1234/3254 |
| الوسوم: |
إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
|
| الملخص: | In recent years, the pervasive use of social media platforms has elevated the significance of users' emotions, ranging from satisfaction to anxiety, within digital discourse. This trend is particularly pronounced in the Arabic-speaking digital landscape, where the language holds a prominent presence across various communication platforms and social media networks. Effectively interpreting sentiment voiced in Arabic speech and script poses a significant difficulty owing to several elements. These consist of the ingrained sophistication of Arabic morphology, the scarcity of comprehensive Arabic corpora, and the divergence of Arabic accents and dialects. Conventional methods for sentiment analysis have struggled to provide accurate results in this context, prompting the exploration of deep learning techniques as a potential solution. This study tackles the challenge of Arabic sentiment analysis, encompassing modern standard Arabic as well as dialectal Arabic, by examining the effectiveness of assorted deep learning models in discerning sentiments expressed in both speech and its corresponding transcript data. Specifically, the study compares Arabic speech sentiment analysis with its transcript analysis, employing CNN, LSTM, BI-LSTM, and GRU models. A diverse dataset comprising Arabic speech and text samples, encompassing positive, negative, and neutral sentiments, was meticulously curated for this purpose. Each deep learning model was trained and evaluated separately on both speech and text data to assess its ability to discern sentiment nuances across these modalities. The results yielded insightful findings regarding the relative performance of CNN, LSTM, BI-LSTM, and GRU models across speech and text datasets. For Arabic speech sentiment analysis, BI-LSTM emerged as the most effective model, achieving an impressive accuracy rate of 89%, followed closely by CNN at 84%. In contrast, for Arabic text sentiment analysis, GRU and CNN techniques outperformed LSTM and BI-LSTM, achieving accuracy rates of 73% and 72%, respectively. Overall, this study contributes valuable insights into the domain of Arabic sentiment analysis, shedding light on the comparative effectiveness of CNN, LSTM, BI-LSTM, and GRU models in analysing sentiments expressed through both speech and text data. These findings hold significant implications for scholars and experts seeking to develop efficient emotion evaluation systems tailored to Arabic language contexts. |
|---|