Abstract: Typically, lexical semantics have focused on how to correlate the meaning of the lexical item with its syntax with respect to the language construction. Due to the hidden semantic structures in text content, Latent Semantic Analysis (LSA) as one of the topic modeling algorithms has shown an effectiveness to treat with the text noisiness, high dimensional issues semantically. It can distinguish the most informative and discriminative features from a collection of text. In this study, the issue of dimensional sparse of the short messages features was avoided through enriching the message’s text with linguistic semantics. The enriched features have been fetched to the LSA algorithm to produce the features vector transformation and enhancing the classification process. The experiments of the research have shown very promising results in comparison to the most popular machine learning methods. The classification performance in terms of the evaluation metrics has been discussed and compared against the results without the proposed enriching.