The paper presents a state-of-the-art sentiment analysis system designed to detect the sentiment of short informal textual messages, such as tweets and SMS, at both the message-level and term-level. The system leverages a supervised statistical text classification approach, incorporating various surface-form, semantic, and sentiment features. Key innovations include the use of novel high-coverage tweet-specific sentiment lexicons, which are automatically generated from tweets with sentiment-word hashtags and emoticons. To handle negated contexts, separate sentiment lexicons are created for words in affirmative and negated contexts. The system achieved first place in the SemEval-2013 shared task on sentiment analysis in Twitter, with F-scores of 69.02 and 88.93 for the message-level and term-level tasks, respectively. Post-competition improvements further boosted performance to 70.45 and 89.50. The system also performed well on the SemEval-2013 SMS test set and a corpus of movie review excerpts. Ablation experiments demonstrated that the use of automatically generated lexicons significantly improved performance, contributing up to 6.5 percentage points. The paper discusses the creation of these lexicons, their evaluation, and their impact on the system's performance.The paper presents a state-of-the-art sentiment analysis system designed to detect the sentiment of short informal textual messages, such as tweets and SMS, at both the message-level and term-level. The system leverages a supervised statistical text classification approach, incorporating various surface-form, semantic, and sentiment features. Key innovations include the use of novel high-coverage tweet-specific sentiment lexicons, which are automatically generated from tweets with sentiment-word hashtags and emoticons. To handle negated contexts, separate sentiment lexicons are created for words in affirmative and negated contexts. The system achieved first place in the SemEval-2013 shared task on sentiment analysis in Twitter, with F-scores of 69.02 and 88.93 for the message-level and term-level tasks, respectively. Post-competition improvements further boosted performance to 70.45 and 89.50. The system also performed well on the SemEval-2013 SMS test set and a corpus of movie review excerpts. Ablation experiments demonstrated that the use of automatically generated lexicons significantly improved performance, contributing up to 6.5 percentage points. The paper discusses the creation of these lexicons, their evaluation, and their impact on the system's performance.