Sentiment Analysis
Last updated:
What sentiment analysis is
Sentiment analysis, also called opinion mining, assigns an attitude to a piece of text. The simplest version labels a message as positive, negative or neutral. More detailed versions score intensity, detect emotions such as anger or confusion, or measure sentiment per topic, for example positive about the product but negative about delivery. That last variant is known as aspect-based sentiment analysis.
How it works
Early tools counted positive and negative words from a dictionary. Machine learning models later learned from labelled examples. Today, language models judge tone in context, including negation such as “not bad at all”, although sarcasm, irony and mixed feelings remain hard.
The output is a probability, not a fact. It is most reliable when aggregated over many conversations rather than read message by message.
Example
A customer writes: “Third time asking. Where is my refund?” A sentiment model flags this as strongly negative. In a support team, that flag could move the conversation up a queue, alert a colleague or change how a reply is worded. Across many chats, a high share of negative messages about one topic can point to a recurring problem.
Why it matters for business chatbots
Sentiment data helps teams prioritize unhappy customers and see which topics cause frustration. It is a signal, not a verdict: a short, neutral-sounding message can still come from an annoyed customer. Many teams combine it with direct feedback, such as CSAT surveys, and with reading real transcripts.
Sentiment analysis and intoCHAT
intoCHAT offers sentiment analysis as an option, off by default and set per agent. Once it is on, each conversation is labelled overnight as positive, neutral or negative, together with up to two topics from a list you keep (up to 15 topics, suggested from recent conversations if you leave it empty). The analysis does not use your monthly messages. You can filter conversations by topic and sentiment, export both as CSV columns, and see the sentiment split and top topics in the Outcomes section of the dashboard. Reading the transcripts still shows best how visitors phrase their questions and where answers fall short.