情感分析序列化模型（演讲PPt）.ppt

Sequential Models for Sentiment Prediction Yi Mao Guy Lebanon (Proceedings of the ICML Workshop on Learning in Structured Output Spaces, Pittsburgh, PA, 2006.) Outline Motivation Introduction Problem Definition and Related Work The Authors Approach and Experiment Results Conclusion Motivation Many documents( reviews and blogs)are written to convey a particular opinion or sentiment. Predicting the document’s sentiment will: a)allow matching the sentiment, as well as the topic, with the user’s interests b)assist in document summarization and visualization Outline Motivation Introduction Problem Definition and Related Work The Authors Approach and Experiment Results Conclusion Introduction sentiments can be simply considered as positive ,neutral,negtive. it seems that sentiment prediction can be classified as a text categorization problem. but sentiment prediction is different, for: Sentiment prediction (1) sentiments are ordinal variables (2) several contradicting opinions might co-exist in one document (3) context plays a vital role in determining the sentiment. our method we use a sequential conditional model. enforce monotonicity constraints on the model’s parameters. so we treat the sentiment labels as ordinal variables Outline Motivation Introduction Problem Definition and Related Work The Authors Approach and Experiment Results Conclusion Local and Global Sentiments Global Sentiment the sentiment of the entire document Local Sentiment the sentiment associated with a particular part of the text Local and Global Sentiments global sentiment : a function of the local sentiment estimating the local sentiment is a key step in predicting the global sentiment Local Sentiment we view local sentiment as: a function on the words in a document (a piecewise constant function on sentences) taking values in a poset, (O,≤) taking context into account then the problem is: predicting a sequence of sentiments y = (y1, . . .