Sentiment analysis

Judging whether comments and mentions about a topic lean positive or negative, and how strongly, using software or manual review.

By Azande PorterUpdated 4 min read

What is sentiment analysis?

Sentiment analysis is the process of judging whether a piece of text, such as a comment, review, or mention, expresses a positive, negative, or neutral attitude, usually with software that scores each item and then adds the scores up.

In social media marketing it answers a question that counts can't: not how many people responded, but how they felt. For example, a launch post with 400 replies looks the same on an engagement dashboard whether those replies are excited or furious.

How does sentiment analysis work?

There are three broad approaches, often combined:

  • Manual coding. A person reads a sample and tags each item positive, negative, neutral, or mixed. Slow, but reliable on small volumes, and usually the standard the other methods are checked against.
  • Word lists. Software counts words associated with positive or negative feeling. Fast and transparent, but easily fooled by context.
  • Machine learning models. A model trained on labeled text scores each item. These handle phrasing better than word lists but are only as good as the data they were trained on.

Google Cloud's documentation gives a concrete example of model output. Its Natural Language API returns a score from -1.0 (negative) to 1.0 (positive) and a separate magnitude for the overall strength of emotion. A score near zero can mean a neutral text or a mixed one, and the magnitude tells them apart: low for truly neutral, high for strong feelings that cancel out. Google also notes that the threshold for "clearly positive" or "clearly negative" varies by use case, so you have to set and tune your own.

How is sentiment analysis used in social media marketing?

Typical uses:

  • Launch and announcement reactions. Was the response to a price change mostly questions, mostly complaints, or mostly praise?
  • Brand health over time. Is the tone of mentions shifting month to month, alongside brand awareness?
  • Triage. Flagging negative replies quickly is part of community management.
  • Campaign comparison. Which of two campaigns drew warmer responses, not just more of them?

At scale, it usually runs on top of social listening, which gathers mentions across platforms rather than just replies to your own posts. For example, if 300 replies to an announcement are coded and 90 read as negative, that is 30% negative. The useful next step is reading those 90 to find out why, which a score alone can't tell you.

What are the limits of sentiment analysis?

Language is harder than it looks. Sarcasm can read as praise, slang flips meanings, emoji carry tone that word lists miss, and one reply can praise the product and hate the price. Industry jargon causes errors too: "sick" and "killer" are compliments in some communities.

Sampling matters as much as scoring. The people who reply are not your whole audience, and a few loud accounts can swing a small sample. A reshare isn't agreement either, since shares can carry criticism in a quote post. Treat sentiment scores as a direction, check a sample by hand, and pair them with the rest of your social media analytics rather than reporting them as a precise measure of how people feel.

What is the difference between sentiment and engagement rate?

Engagement rate measures how many people responded; sentiment measures how they responded. A post can have a high engagement rate and negative sentiment, which is what a pile-on looks like in the numbers. The two together tell you more than either alone. For small accounts, reading the replies yourself often beats any tool, and the questions people ask there are some of the best material you have; turning customer questions into social posts shows how.

Platform details checked

How Distinctful helps

Distinctful does not offer sentiment analysis or social listening across the web. It works with your own workspace instead: each post's reported results compared with that account's usual, and Chat to help you read those results and plan the next post, with you making the call. See social media analytics.

Related terms

Go deeper

Read your results against your usual

Compare each post with what that account usually gets before you plan the next one.