2.2 Levels of Sentiment Analysis
In general sentiment is investigated in four levels
1. Doucment Level Sentiment Classification
2. Sentence Level Sentiment Classification
3. Aspect Level Sentiment Classification
4. Word Level Sentiment Classification
2.2.1 Document Level Sentiment Classification
At document level classification, sentiment of whole document is classified as positive or negative. e.g. in the reveiws of any product , sentiment analyzer identify review of each product as positive or negative. This type of sentiment analysis assumes that each document represent sentiment or opinion of single entity. Comparison or evaluation of multiple entitites is not possible in document level sentiment classification.
2.2.2 Sentence Level Sentiment Classification
At sentence level classification, sentiment of sentence is classified as positive, negative or neutral. Neutral means no openion. Sentence level sentiment classification is very close to subjective classification in which objective sentences (factual information ) are distinguished from subjective sentences (opinions).
Difference between document and sentence level classification is that doument indentify positive and negative opinion while in sentence level sentiment classification, each sentence is classified as objective and subjective and further subjective sentence is classified into positive, negative and neutral opinion. In document level, sentiment is classified as a whole while different angles of entities are not taken into account e.g. in the review of cell phone when the review is positive it means in general the writer likes the cell phone but if writer focus on different other aspects of cellphone there may be some negative opinion like “cell phone screen is small”. If each feature and aspect is taken into account then result will go into sentence level sentiment classification. Sentence level classification is used for more accurate analysis of an entity.
2.2.3 Aspect Level Sentiment Classification
Using document and sentence level classification, one cannot exactly analyze that what people liked or not liked. Aspect level analysis is one step further and it foucs on idea of opinion that includes sentiment i.e. positive or negative with target identifiers. The use of sentiment analyzer will be limited without target identifiers e.g. the sentence “Samsung is doing great job under poor economy” is evaluating two aspects i.e. samsung and economy. Sentence is positive for samsung while negative for economy so it is very important to identify the target objects.
2.2.4 Word Level Sentiment Classification
In word level sentiment classification, each word is evaluated .e.g. “the cellphone is good” sentence has “good” word that bear positive polarity. The difference between sentence and word level is how the total score is evaluated. e.g. “cellphone is good but condition is bad” in sentence level the polarity is neutral but in word level totald polarity is based on weight and frequency not polarity only