Text Sentiment Analysis

Text sentiment analysis is a natural language processing method that classifies or scores text by the positive, negative, or neutral opinion it expresses.

텍스트 감성분석은 텍스트가 표현하는 의견을 긍정, 부정, 중립으로 분류하거나 점수화하는 자연어처리 방법이다.

How do I measure whether text responses are positive or negative?

텍스트 응답이 긍정적인지 부정적인지 어떻게 측정하나요?

Sentiment analysis assigns each document a polarity label or score using a lexicon, a trained classifier, or a pretrained language model. It fits open-ended survey responses, reviews, social media posts, or news text where the tone of opinion is the variable of interest.

When to use Text Sentiment Analysis

• Open-ended survey answers or customer reviews in a text column
• Tracking tone of social media or news coverage over time
• Creating a sentiment score to use as a variable in later analysis

When to use something else

• The goal is to find what themes the texts discuss: use topic modeling such as LDA or BERTopic
• Categories are custom and not about opinion polarity: train a supervised text classifier

Assumptions and requirements

• The lexicon or model matches the language and domain of the text
• Validation against a human-coded sample to check accuracy
• Awareness that sarcasm, negation, and mixed sentiment reduce accuracy
• Basic text cleaning appropriate to the method

What the analysis reports

• Sentiment label or polarity score per document
• Distribution of sentiment across documents or groups
• Agreement with human coding (accuracy, F1, or kappa) on a validation sample
• Examples of high- and low-scoring texts

How to report it

Of [N] responses, [value]% were classified as positive; agreement with human coding on a validation sample was κ = [value].

Related methods

• Topic Modeling (LDA)
• Text Classification
• BERT Fine-Tuning

Run Text Sentiment Analysis on Quorant