Analyzing the sentiment of user utterances in conversations with your bot

You can use sentiment analysis to determine the sentiments expressed in a user utterance. With the sentiment information you can manage conversation flow or perform post-call analysis. For example, if the user sentiment is negative you can create a flow to hand over a conversation to a human agent.

Amazon Lex integrates with Amazon Comprehend to detect user sentiment. The response from Amazon Comprehend indicates whether the overall sentiment of the text is positive, neutral, negative, or mixed. The response contains the most likely sentiment for the user utterance and the scores for each of the sentiment categories. The score represents the likelihood that the sentiment was correctly detected.

You enable sentiment analysis for a bot using the console or by using the Amazon Lex API. You enable sentiment analysis on an alias for the bot. On the Amazon Lex console:

If you are using the API, call the CreateBotAlias operation with the detectSentiment field set to true.

When sentiment analysis is enabled, the response from the RecognizeText and RecognizeUtterance operations return a field called sentimentResponse in the interpretations structure with other metadata. The sentimentResponse field has two fields, sentiment and sentimentScore, that contain the result of the sentiment analysis. If you are using a Lambda function, the sentimentResponse field is included in the event data sent to your function.

The following is an example of the sentimentResponse field returned as part of the RecognizeText or RecognizeUtterance response.

sentimentResponse {
    "sentimentScore": {
        "mixed": 0.030585512690246105,
        "positive": 0.94992071056365967,
        "neutral": 0.0141543131828308,
        "negative": 0.00893945890665054
    },
    "sentiment": "POSITIVE"
}

Amazon Lex calls Amazon Comprehend on your behalf to determine the sentiment in every utterance processed by the bot. By enabling sentiment analysis, you agree to the service terms and agreements for Amazon Comprehend. For more information about pricing for Amazon Comprehend, see Amazon Comprehend Pricing.

For more information about how Amazon Comprehend sentiment analysis works, see Determine the sentiment in the Amazon Comprehend Developer Guide.