Sentiment analysis, as a branch of unstructured data mining, has interested people greatly. By skimming through the summarized information about consumer reviews on the web, consumers would know which products to buy or not to buy; retailers/manufacturers would know opinions from people who bought their products or service. It would be beneficial to the sellers, who want to improve their products or service and achieve greater competitiveness.Automatically identifying subjective content on the web is an important prerequisite for sentiment analysis, which is separating opinion from facts. In English, researchers have done some work on subjectivity detection and some existing resources contain lists of subjective words and phrases. However, Chinese lacks systematic studies on automatic subjectivity detection. This paper tries to, form a part of speech perspective, generalize the intricacies of subjective expressions, such as word or phrase and summarize some rules for identifying Chinese subjective expressions automatically.This paper presents a method to measure the strength of subjectivity on a sentence-level, merely using patterns of word class combination. This idea makes the classification of subjective and objective sentences easy and flexible. By setting up different thresholds, the precision and recall can be changed as required. Both subjective and objective sentences achieve high precision and recall (both around 79%). The results are comparable to studies of English subjectivity detection, thus proves the method of measuring subjective strength on a sentence-level is feasible. The subjective patterns of word class combination proposed in this paper can be used for Chinese sentiment analysis. |