Khóa luận tốt nghiệp: Phân lớp quan điểm theo chủ đề dựa vào chuỗi con và cây con phụ thuộc trên miền tin tức tài chính

The rapid proliferation of user-generated content across digital platforms underscores the critical need for effective opinion analysis. Within this context, "phân lớp quan điểm" (opinion classification) emerges as a pivotal task in "khai phá quan điểm" (opinion mining), aiming to automatically categorize subjective expressions as positive or negative. While traditional bag-of-words models have seen application, they often fail to capture the nuanced "trật tự và quan hệ ngữ nghĩa" (word order and semantic relationships) crucial for accurate sentiment discernment. This thesis addresses this gap by proposing a machine learning-based approach to "phân lớp quan điểm theo chủ đề" (topic-based opinion classification) specifically within the "miền tin tức tài chính" (financial news domain). The objective is to leverage "chuỗi con" (substrings) and "cây con phụ thuộc" (dependency subtrees) as robust features to enhance classification performance.

Nghiên cứu sinh, kỹ sư và chuyên gia trong lĩnh vực Xử lý ngôn ngữ tự nhiên (NLP), khai phá dữ liệu, học máy, phân tích cảm xúc và tài chính.

The exponential growth of online textual data, particularly in specialized fields such as financial news, highlights the imperative for advanced "khai phá quan điểm" (opinion mining) techniques to accurately gauge public sentiment. This study tackles the challenging problem of "phân lớp quan điểm" (opinion classification), which involves automatically categorizing user opinions—be they positive or negative—related to specific topics or entities. Such capabilities are invaluable for consumers seeking informed product choices and for manufacturers desiring rapid feedback on their offerings, driving product development and marketing strategies. Recognizing the limitations of approaches that neglect the inherent syntactic and semantic structures of language, this research introduces a novel "phân lớp quan điểm theo chủ đề" (topic-based opinion classification) method. Our methodology employs a sophisticated machine learning framework, specifically utilizing a Support Vector Machine (SVM), which is empowered by a unique feature extraction strategy. This strategy focuses on deriving discriminative patterns from "chuỗi từ con" (substrings) and "cây con phụ thuộc" (dependency subtrees), meticulously capturing the intricate "mối quan hệ cú pháp" (syntactic relationships) between words within sentences. The efficacy of this proposed model was rigorously evaluated on a dataset comprising 312 comments spanning 180 articles from the "miền tin tức tài chính" (financial news domain). Experimental results demonstrated a promising accuracy of up to 72%, affirming the potential of structural linguistic features in enhancing sentiment analysis. This work contributes significantly to advancing robust opinion classification, offering a more nuanced and accurate understanding of sentiment in complex, domain-specific texts.