In this paper, we proposed a new mining approach to process unstructured data with complex structure
in Social Networking Service(SNS). The proposed method is to extract meaningful emotional information
more effectively by analyzing the texts posted on the SNS. In general, SentiwordNet method, which is the
main purpose of forming the relationship between existing vocabularies, has a problem that is difficult to
apply to korean sentences of implicit expression and Unique character structure. To improve this, applying
the Levenshtein distance algorithm, Korean words are efficiently searched and the sentiment word dictionary
is constructed by assigning weights to the result. Sentence analysis was processed in realtime to distinguish
general, positive, negative sentences and the results were expressed as a percentage. According to the
result of the experiment, we proved that is more effective for classifying emotions in sentences than Navie
Baysian. Hadoop system was used for efficient with handle web-scale data.