To better understand the key terms associated with each sentiment, word clouds were generated for positive, neutral, and negative reviews. These visualizations highlight the most frequently mentioned words in each category, offering valuable insights into customer opinions about KFC restaurants
Figure 4.25: Word cloud of positive reviews
The word cloud for positive reviews shows words like "phục vụ" (service), "thân thiện" (friendly), "sạch sẽ" (clean), and "ngon" (delicious), indicating that customers often appreciate the service quality, cleanliness, and food taste. These terms reflect the aspects customers are most satisfied with.
Figure 4.26: Word cloud of neutral reviews
49
The neutral review word cloud is dominated by terms such as "không" (not), "được"
(received), and "bình thường" (normal). These words suggest that customers in this category had average or mixed experiences, neither particularly positive nor negative.
Figure 4.27: Word cloud of negative reviews
For negative reviews, prominent words include "nhân viên" (staff), "thái độ" (attitude),
"đồ ăn" (food), and "khách hàng" (customer). These terms point to dissatisfaction with staff behavior, food quality, or customer service. The frequent appearance of "thất vọng" (disappointed) underscores the negative sentiment.
By analyzing these word clouds, we can identify specific strengths and weaknesses in customer experiences. Positive feedback emphasizes strengths in food and service, while negative and neutral reviews highlight areas for improvement, such as staff attitudes and food quality. These insights can guide KFC in enhancing customer satisfaction and addressing common complaints.
50
CONCLUSION Summary of Findings
This research investigated the use of sentiment analysis on customer reviews of KFC restaurants in Hanoi, Vietnam, aiming to extract valuable insights for enhancing customer satisfaction and refining business strategies. Some models like Logistic Regression, Multinomial Nạve Bayes, Support Vector Classifier (SVC), and PhoBERT, were applied and assessed based on key performance. Among these models, PhoBERT achieved the highest performance, attaining an accuracy of 90.64%, highlighting its strong capability in handling Vietnamese text data.
The analysis of customer sentiment distribution revealed key patterns in positive, neutral, and negative reviews. Positive feedback often highlighted food quality, friendly staff, and cleanliness, while negative reviews frequently criticized staff attitudes, service quality, and specific aspects of the dining experience. Word clouds provided a visual representation of these sentiment trends, offering a clearer understanding of the aspects that influence customer opinions.
To address challenges such as imbalanced data, techniques like SMOTE (Synthetic Minority Oversampling Technique) were utilized, improving the model’s ability to classify less frequent sentiment categories accurately. Additionally, the study underscored the importance of pre-trained models like PhoBERT and advanced feature extraction techniques such as TF-IDF to gain effectiveness.
The findings also notices the importance of sentiment analysis to transform customer feedback into valuable insights, enabling KFC and similar businesses to improve service quality, address customer concerns, and refine strategic decisions. Despite its successes, the study acknowledged limitations, including dataset specificity, computational complexity, and the exclusive reliance on textual feedback, which future research should aim to overcome.
Implications for KFC and the Fast-Food Industry
The results of this thesis have significant implications for KFC and the broader fast- food industry, particularly through the application of the best-performing model, PhoBERT, in a user-friendly interface. This interface would allow customers to input their comments or feedback, and the system would predict the sentiment label (positive, neutral, or negative) in real-time. Such a tool would enable businesses like KFC to process and analyze large volumes of customer feedback efficiently, identifying sentiment trends without the need for manual effort. Negative comments could be flagged instantly, allowing customer service teams to address concerns promptly and improve overall satisfaction. Additionally, insights from the sentiment predictions could guide strategic decisions, such as focusing on maintaining aspects
51
frequently praised by customers, like food quality, or addressing recurring issues, such as service quality. Beyond KFC, this approach offers scalability, making it applicable to other fast-food chains or restaurants, ultimately helping the industry enhance customer experience management and strengthen competitive positioning. By leveraging advanced sentiment analysis, KFC can transform customer feedback into actionable insights, fostering better customer relationships and driving business growth.
One of the key applications of this research lies in leveraging the best-performing model, PhoBERT, to develop a user-friendly interface for sentiment analysis. By integrating the PhoBERT model into a web or mobile platform, KFC and other fast- food chains can enable customers to input their comments or feedback directly. The system would then predict the sentiment label (positive, neutral, or negative) in real- time, providing immediate insights into customer opinions. Here is the interface of front-end web for user to type their comment. Then, the system will predict the label of the sentiment.
The interface of front-end web (1)
The interface of front-end web (2)
52 Limitations of the Study
This study, while offering valuable insights into sentiment analysis for customer feedback in the fast-food industry, has several limitations. First, this dataset is specific to KFC reviews in Hanoi, Vietnam, which may limit the generalizability of the findings to other regions or industries. Second, the study heavily relies on text-based feedback, neglecting other data forms such as images or videos that could show additional context for customer opinions. Lastly, the study focuses on sentiment prediction without delving deeply into the root causes of customer satisfaction or dissatisfaction, which could provide richer insights for business improvements.
Recommendations for Future Research
Future research should aim to solve the problems highlighted above to enhance the scope and applicability of sentiment analysis. Expanding the dataset to include reviews from other regions, industries, or even languages would help in creating more generalized models. Researchers could also explore multimodal sentiment analysis by incorporating data such as images, videos, or audio recordings alongside textual feedback for a more comprehensive understanding of customer sentiment. Simplifying and optimizing advanced models like PhoBERT to reduce computational requirements would make them more accessible for real-time and large-scale applications.
Additionally, future work could include aspect-level sentiment analysis combined with customer behavior prediction, providing actionable insights tailored to specific areas such as service quality, pricing, or menu preferences. Integrating these advancements could significantly contribute to more effective sentiment analysis applications in diverse contexts.
53
REFERENCES
[1] Marouane Birjali, Mohammed Kasri, Abderrahim Beni-Hssane, “A comprehensive survey on sentiment analysis: Approaches, challenges and trends”. Available at:
https://www.sciencedirect.com/science/article/abs/pii/S095070512100397X
[2] Mayur Wankhade, Annavarapu Chandra Sekhara Rao, “A survey on sentiment analysis methods, applications, and challenges”. Available at:
https://link.springer.com/article/10.1007/s10462-022-10144-1
[3] Erick Odhiambo Omuya, George Okeyo, Michael Kimwele, “Sentiment analysis on social media tweets using dimensionality reduction and natural language processing”. Available at:
https://onlinelibrary.wiley.com/doi/10.1002/eng2.12579
[4] “Understanding TF-IDF (Term Frequency-Inverse Document Frequency)”.
Available at:
https://www.geeksforgeeks.org/understanding-tf-idf-term-frequency-inverse-document- frequency/
[5] Nitesh V Chawla, Kevin Bowyer, “SMOTE: Synthetic Minority Over-sampling technique”. Available at:
https://www.researchgate.net/publication/220543125_SMOTE_Synthetic_Minority_Ov er-sampling_Technique
[6] Turintech, “What is imbalanced data and how to handle it?”
https://www.turintech.ai/what-is-imbalanced-data-and-how-to-handle-it/
[7] (2015) “Improving the User experience through pratical data analytics”. Available at:
https://www.sciencedirect.com/topics/computer-science/binary-logistic-
regression#:~:text=Binary%20Logistic%20Regression%20is%20defined,for%20regul ar%20linear%20regression%20models.
[8] Farrikhin, Keszya Wabang, Oky Dwi, “Application of the Naive Bayes classifier algorithm to classify community complaints”. Available at:
https://www.researchgate.net/publication/365099641_Application_of_The_Naive_Bay es_Classifier_Algorithm_to_Classify_Community_Complaints#pf4
[9] GreeksforGreeks, “Multinomial Naive Bayes”
https://www.geeksforgeeks.org/multinomial-naive-bayes/
[10] Dishant Salunke, “SVC-Support Vector Classifier”. Available at: