Posted On: Mar 9, 2022
Amazon Comprehend is a natural-language processing (NLP) service that uses machine learning to uncover insights in text data. Starting today, Comprehend now offers Targeted Sentiment, a new API that provides more granular sentiment insights by identifying the sentiment (positive, negative, neutral, or mixed) towards entities within text.
Businesses have access to vast amounts of social media posts, reviews, customer service calls/emails, and blogs which have information on how customers feel about their brands, products, and services. Understanding the voice of the customer is essential for these businesses to react quickly to customer feedback or to identify issues/trends with their offering. Historically, businesses have utilized manual processes to assess customer sentiment, but this practice is error prone and doesn’t scale. With Targeted Sentiment, enterprise customers and partners can pinpoint what their customer’s sentiment is expressed towards.
Customers currently use Comprehend’s overall sentiment API to identify the sentiment for an entire block of text. For statements like “The burger was delicious, but it was soggy” the overall sentiment output is “mixed” - which doesn’t provide enough details to influence business decisions. With Targeted Sentiment, the output will (i) identify the entities in the text, (ii) find the sentiment towards each entity mention, and (iii) group together multiple mentions of the same entity (i.e. co-reference; “burger” and “it” refer to same entity). With Targeted Sentiment, the output will show the “burger” (entity) is a food item (entity type) and is positive (sentiment), while “it” (entity referencing “burger”) is a food item (entity type) that is negative (sentiment).
Customers can use either sentiment or targeted sentiment depending on the type of insights they need. For example, let’s assume a restaurant owner wants to analyze the review, “The tacos were delicious and the staff was friendly.” The owner would use Sentiment API to know if the overall restaurant review was positive, negative, neutral, or mixed (in this example, the overall sentiment is positive). Alternatively, the owner would use Targeted Sentiment to know what in the restaurant was positive, negative, neutral, or mixed (in this example, “tacos” were positive and “staff” was positive). Customers can also use both APIs to first identify overall sentiment as a baseline, and then use targeted sentiment to drill into the sentiment details (e.g. specific entities).
To learn more and get started, visit the Amazon Comprehend product page or documentation page.