Abstract: Sentiment analysis is an emerging research field that can be integrated with other domains, including data mining, natural language processing and machine learning. In political articles, it is difficult to understand and summarise the state or overall views due to the diversity and size of social media information. A number of studies were conducted in the area of sentiment analysis, especially using English texts, while Arabic language received less attention in the literature. In this study, we propose a detection model for political orientation articles in the Arabic language. We introduce the key assumptions of the model, present and discuss the obtained results, and highlight the issues that still need to be explored to further our understanding of subjective sentences. The main purpose of applying this new approach based on Rough Set (RS) theory is to increase the accuracy of the models in recognizing the orientation of the articles. We present extensive simulation results, which demonstrate the superiority of the proposed model over other algorithms. It is shown that the performance of the proposed approach significantly improves by adding discriminating features. To summarize, the proposed approach demonstrates an accuracy of 85.483%, when evaluating the orientation of political Arabic datasets, compared to 72.58% and 64.516% for the Support Vector Machines and Naïve Bayes methods, respectively.
The Challenge of Arabic Political Sentiment
While sentiment analysis—the process of determining whether a text is positive, negative, or neutral—is a mature field for the English language, it remains in its infancy for Arabic. Arabic presents unique linguistic challenges, including a complex morphology, a lack of capitalization, and a right-to-left script.
Beyond simple polarity, this research tackles the more nuanced problem of political orientation. In the volatile world of political discourse, simply knowing if an article is "positive" isn't enough; researchers and analysts need to know if the content aligns with Reformist, Conservative, or Revolutionary ideologies. Existing machine learning models often struggle with the inherent vagueness and subjectivity of political language, creating a need for a more robust mathematical approach to handle uncertainty.
The LRST Model: Bridging Lexicons and Rough Sets
To address these challenges, the researchers proposed a hybrid model called LRST (Lexicon and Rough Set Theory). The workflow consists of several key stages designed to handle the linguistic complexity of Arabic and the inherent vagueness of political discourse.
Methodology Overview
- Preprocessing: This is critical for Arabic. The team performed "tatwel" removal (removing elongated letters), normalization (standardizing variations of letters like Alif and Ya), and "light stemming" to strip words down to their essential forms without losing semantic meaning.
- N-gram Extraction: The text was broken down into N-grams (sequences of 1 to 5 words). This helps capture context that single words might miss.
- Rough Set Theory: This is the core innovation. Rough Set theory is a mathematical tool used to deal with "imperfect" or vague data. The authors specifically utilized the lower approximation. In simple terms, if you have a category (like "Reformist"), the lower approximation consists of all words that certainly belong to that category based on the training data.
- Classification: Instead of a standard "black box" classifier, the LRST model compares a test article against these "certain" sets in the lexicon. The article is assigned to the class where it has the highest number of matches within the Rough Set lower approximations.
Experimental Setup and Datasets
The study utilized two primary datasets to validate the model:
- Arabic Political Corpus: A manually annotated dataset created by the authors containing 206 articles categorized into Reformist, Conservative, and Revolutionary classes.
- BBC News Dataset: To test the generalization of their model, the authors applied it to a standard English dataset of 2,225 records across five categories (Business, Entertainment, Politics, Sport, and Tech).
Key Results and Insights
The LRST model showed clear superiority in the political Arabic context, achieving an accuracy of 85.48% (using 2-grams). This significantly outperformed traditional machine learning benchmarks:
- LRST Accuracy: 85.48%
- SVM Accuracy: 72.58%
- Naïve Bayes Accuracy: 64.51%
The "Bigram Sweet Spot"
The researchers found that 2-grams (bigrams) yielded the highest accuracy. While unigrams (single words) are common, they often lack political context. Conversely, 4-grams and 5-grams became too specific, leading to "data sparsity" where the model couldn't find enough matches to make an accurate call.
When tested on the English BBC dataset, the model achieved a staggering 96.7% accuracy across five different news categories. This suggests the LRST approach is highly robust for document-level classification across different languages, proving that a linguistically-informed hybrid model (Lexicon + Math) can be more effective than standard statistical models when dealing with specialized datasets.
Conclusion and Future Work
The study demonstrates that embracing the inherent vagueness of political language through Rough Set theory leads to more precise classification. While the results are strong, future work will focus on scaling the dataset size and exploring more granular political classifications. Additionally, the authors suggest that the next frontier for the LRST method is multimodal analysis, combining text analysis with video feedback to better understand public sentiment.
Summary updated