Alida, a firm that helps brands build engaged research communities to gather consumer feedback, has utilized machine learning to handle large volumes of responses from consumers. However, they discovered that traditional natural language processing models had difficulty understanding nuanced feedback found in open-ended survey responses.
This obstacle led them to leverage Anthropic’s Claude Instant model on Amazon Bedrock, a service that offers high-performing foundation models from leading AI companies. Amazon Bedrock allowed Alida to bring their service to market faster than if they had used other machine learning providers or vendors.
A key problem Alida faced was that while multiple-choice questions in surveys were easy to analyze, they lacked nuance and depth. Open-ended questions provided superior richness and context, but their complexity and nuance were hard for traditional natural language processing models to fully understand.
Through using Amazon Bedrock, Alida was able to bypass machine learning framework configurations and provisioning infrastructure. This gave them the opportunity to focus on prompt engineering, which is a more efficient approach to improving their machine learning capabilities.
Alida’s implementation of this system involved constructing topic and sentiment classification as a service, with survey-response analysis as its first application. This methodology minimized complexity and allowed for continual improvement of internal implementation details.
Through evaluating LLMs (Large Language Models) from various providers, Alida found that Anthropic’s Claude Instant provided the right balance between cost and performance. The use of “prompt chaining strategy” allowed for more detailed tracking and monitoring of accuracy and performance at each step.
The resulting solution was vastly superior to Alida’s existing NLP system. The new system was able to generate more comprehensive responses to pre-defined topics and assign sentiment. In contrast, the existing NLP system could only extract relevant keywords but failed to achieve comprehensive topic group assignment due to a lack of true comprehension.
Moreover, using in-context prompt engineering for LLMs significantly hastened the development phase since it eliminated the need to curate thousands of human-labeled data points to train traditional NLP models.
In conclusion, Alida’s integration of the Claude Instant model on Amazon Bedrock has given its clients richer insights more swiftly. This is a testament to the power of using advanced machine learning technologies in the analysis of complex open-ended survey responses.
