This paper presents a model that combines unsupervised and supervised techniques to learn word vectors that capture both semantic and sentiment information. The model uses document-level sentiment annotations, such as star ratings, to improve word representations. It is evaluated on sentiment and subjectivity corpora and outperforms previous methods. The model also introduces a large dataset of movie reviews for benchmarking. The model uses a probabilistic approach to learn word representations, incorporating both semantic and sentiment information. It is trained using alternating maximization and is shown to perform well in sentiment classification tasks. The model is compared with other vector space models and is found to be effective in capturing both semantic and sentiment relations. The model is also tested on subjectivity detection tasks and shows superior performance. The paper concludes that the model is a flexible and effective approach for sentiment analysis and retrieval.This paper presents a model that combines unsupervised and supervised techniques to learn word vectors that capture both semantic and sentiment information. The model uses document-level sentiment annotations, such as star ratings, to improve word representations. It is evaluated on sentiment and subjectivity corpora and outperforms previous methods. The model also introduces a large dataset of movie reviews for benchmarking. The model uses a probabilistic approach to learn word representations, incorporating both semantic and sentiment information. It is trained using alternating maximization and is shown to perform well in sentiment classification tasks. The model is compared with other vector space models and is found to be effective in capturing both semantic and sentiment relations. The model is also tested on subjectivity detection tasks and shows superior performance. The paper concludes that the model is a flexible and effective approach for sentiment analysis and retrieval.