This paper introduces a novel classification system designed for Topic-Based Message Polarity classification, a crucial task in sentiment analysis for determining the positive or negative sentiment of a given message or topic. The system prioritizes the stronger sentiment when both positive and negative sentiments are expressed within the same context. The experimentation phase utilizes a Twitter dataset to assess the effectiveness of the proposed system. The classification model is trained using a Support Vector Machine (SVM) classifier, leveraging insights from a training dataset of tweets. The trained model is then applied to ascertain the sentiment of an anonymous tweet. Various feature sets, including Brown Dictionary Features, Semantic features, linguistic features, word embedding features, and Sentiment Lexicon features, are employed as inputs to the SVM classifier to capture patterns within the training dataset. Notably, this work introduces a set of new features based on word embeddings, supplementing the existing feature set. These novel features are incorporated into the experimentation, demonstrating their efficacy in improving sentiment prediction for topic-based messages. he results indicate that the proposed approach, particularly with the introduction of new features, achieves commendable performance in topic-based sentiment prediction. Comparative analysis against existing solutions in sentiment classification reveals that the proposed features enhance sentiment identification capacity significantly. Furthermore, the study identifies the superiority of the proposed features over existing ones, showcasing their ability to elevate sentiment classification accuracy. This research contributes valuable insights and advancements to the field of sentiment analysis, specifically in the domain of topic-based message polarity classification. The findings highlight the effectiveness of the proposed system and underscore the significance of incorporating innovative features, such as those based on word embeddings, to enhance sentiment prediction models.

Semantics, Twitter, Post, Word Embedding, Message, Classification