I. INTRODUCTION

Aspect-based sentiment analysis (ABSA) is a method deployed in natural language processing (NLP) and text mining to dissect sentiments conveyed regarding different aspects or qualities within text. ABSA provides a detailed understanding of sentiment in text. Rather than simply gauging an overall sentiment towards a document or sentence, ABSA gets into the sentiment surrounding individual aspects or features mentioned in the text. Additionally, businesses can identify where their products stand out and where they need improvement by analyzing sentiment for various aspects. For instance, if negative sentiments persistently dominate concerning the user interface of a tech product then redesigning it for better usability might be necessary for the company. Similarly, if a restaurant receives positive sentiment about its food quality but negative sentiment about its service, the management can concentrate on improving customer service training for staff. ABSA offers the capability to analyze customer opinions and sentiments towards various products or services available in the market. This enables companies to gain insights into how customers perceive specific aspects of their offerings, allowing them to identify emerging trends, preferences, and potential areas for improvement. Moreover, ABSA can be utilized to compare the sentiment towards different competitors in the market, enabling companies to pinpoint areas where they may have a competitive advantage or disadvantage. Furthermore, ABSA generalizes the process of analyzing customer feedback by automating the task. It can efficiently analyze feedback from diverse sources such as social media, online reviews, and customer surveys. By automatically categorizing feedback based on different aspects or features, companies can swiftly identify common pain points, issues, or areas of satisfaction among their customers.

After ABSA, the insights gained can be used to form future product updates or menu changes. For example, if there is regular demand or complaint from the customers for lack of a feature in a technological product, the company should consider including it in the next version. In addition, feedback on popular and unpopular menus can help restaurants improve them accordingly so as to meet customer demands more effectively. All in all, aspect-based sentiment analysis provides valuable clues about hidden opinions and emotions expressed in text thus enabling businesses to make data-driven choices while improving their products, services and general customer support.

II. RELATED WORK

Aspect based sentiment analysis focuses on understanding sentiments towards particular factors mentioned in text. Let us discuss some of the existing methodologies for achieving the same. Under supervised learning, we train a machine learning model on labeled data, where each review is annotated with the sentiment towards specific aspects. Techniques such as Support Vector Machines (SVM), Random Forests, and Naive Bayes classifiers are commonly used for sentiment classification. Deep learning models, particularly Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and more recently, models like BERT (Bidirectional Encoder Representations from Transformers), have shown significant advancements in sentiment analysis tasks. These models can capture complex relationships within text data and review classification tasks. Methods such as rule-based systems, sequence labeling using Conditional Random Fields (CRFs), and neural network-based approaches like BiLSTM-CRF (Bidirectional Long Short-Term Memory with Conditional Random Fields) are commonly used for aspect extraction. Lexicon-based methods leverage sentiment lexicons or dictionaries containing words annotated with their sentiment polarity (positive, negative, neutral). These methods calculate sentiment scores for individual aspects by aggregating the sentiment scores of the words associated with each aspect. Techniques like VADER (Valence Aware Dictionary and sEntiment Reasoner) and SentiWordNet are widely used for sentiment analysis. In certain domains like restaurants or tech products, domain-specific knowledge can be integrated into the sentiment analysis process. This includes creating domain-specific sentiment lexicons, incorporating domain-specific features into machine learning models, or fine-tuning pre-trained models on domain-specific data to improve performance.

III. CONCLUSION

One limitation was the potential for language and cultural variations in the reviews, which could affect the accuracy of sentiment analysis. Reviews often contain subjective opinions and noisy text, which can pose challenges for sentiment analysis. Additionally, as the volume of reviews grows, scalability and efficiency become critical concerns. One of the key challenges faced during the project was dealing with various delimiters used to separate different sentiments for different aspects within the reviews. This could include punctuation marks, emojis, or even specific phrases indicating sentiment. To address this, we implemented a preprocessing step to clean the text data before sentiment analysis. This preprocessing involved removing or standardizing delimiters to ensure consistency in sentiment analysis. Another challenge was determining and using appropriate keywords and filters to extract relevant information from the reviews. This involved identifying key terms or phrases indicative of sentiment towards specific aspects of the product or service being reviewed. To overcome this, we conducted thorough keyword analysis and refined our keyword filtering methods based on the context of the reviews and the domain of the project. Training the sentiment analysis model to appropriately handle variations of words commonly used in reviews, such as "gud" for "good" or "vry" for "very", presented another challenge. To address this, we employed techniques such as text normalization and expansion of the training dataset to include variations of commonly used words. Additionally, we leveraged techniques like stemming or lemmatization to map similar words to their base forms, thereby improving the model's ability to generalize and recognize sentiment.