Sentiment analysis determines whether the analyzed text expresses a negative, positive, or neutral opinion. By analyzing the content of a text sample, it is possible to estimate the emotional state of the writer of the text and the effect that the writer wants to have on the readers. Sentiment analysis in Pega Platform combines the lexicon-based and machine learning-based approaches to predict the polarity of the analyzed text.
In Pega Platform, lexicons are lists of features that provide sentiment values for words, multiple sentiments within a phrase (for example, ridiculously awesome), negation words (for example, not and no), and stop words (for example, because, such, have). Use lexicons as semantic features for machine learning. Lexicons are defined for each supported language and stored as decision data records.
Text analyzers can contain algorithms that act on words, phrases, sentences, or the whole text. Pega Platform uses a maximum entropy algorithm to train sentiment analysis models. When the training is completed, you can upload the model as part of a text analyzer to perform sentiment analysis in your application to analyze the voice of customer materials, such as emails and chat messages. You can train custom sentiment analysis models in Prediction Studio.
Each sentence that undergoes sentiment analysis is assigned a sentiment score between -1 and 1. The individual scores of all sentences are used to calculate the overall sentiment of the text unit. You can specify the score range for the negative sentiment to decide how your text analyzer detects sentiment.
- Configuring sentiment analysis
Select the sentiment model and the lexicon to apply on the data that you want to analyze.