Zignal Labs Performs Next-Level Sentiment Analysis Using Amazon SageMaker and Amazon EC2

• 分析与建模 - 机器学习
• 平台即服务 (PaaS) - 数据管理平台

• 销售与市场营销
• 商业运营

• 质量预测分析
• 实时定位系统 (RTLS)

• 数据科学服务
• 云规划/设计/实施服务

Zignal Labs, a company that helps its customers measure brand impact, mitigate reputation risks, and inform data-driven communications strategies, wanted to take existing sentiment classification techniques to the next level with a focus on reputation polarity. The company wanted to offer a solution that identifies the actual positive or negative impact of online content on a brand. Zignal Labs was all too familiar with the limitations of third-party sentiment-analysis solutions, having experimented with many of them itself. Some of these tools presented problems around scalability, and some weren't well suited to all the different media sources Zignal needed to track.

Zignal Labs is a company that offers solutions that analyze the entire digital media landscape to deliver instant insights for the company’s Fortune 1000 customers. The company is based in San Francisco, California, and employs 100 people. Zignal Labs helps its customers measure brand impact, mitigate reputation risks, and inform data-driven communications strategies. The company has been using Amazon Web Services (AWS) since its founding in 2011.

Zignal Labs used AWS to build a sentiment-analysis pipeline that could better understand the nuances of brand mentions across the entire digital landscape. The Zignal Labs pipeline does this with a machine-learning solution based on Amazon SageMaker, a fully managed platform that enables developers and data scientists to quickly and easily build, train, and deploy machine- learning models at any scale. It also uses Amazon EC2 C5 instances, featuring Intel Xeon Scalable (Skylake) processors. In addition to Amazon SageMaker and Amazon EC2 C5 instances, Zignal Labs utilizes a distributed streaming architecture, including Spark, Storm, and Elasticsearch, to ingest more than three billion documents per month. The collected articles, tweets, blog posts, reddit posts, broadcast television programs, and comment threads are analyzed in Amazon SageMaker using machine-learning models that are retrained daily with inputs that include label data from “Human Intelligence Tasks” performed by workers from the Amazon Mechanical Turk (Amazon MTurk) marketplace.

• The new Zignal Labs sentiment pipeline is delivering results that show at least 30 percent improvement in precision compared to prior methods, helping the company win and retain customers.
• The solution was built on AWS, increased accuracy is being delivered at much lower cost than would be possible using third-party sentiment-analysis solutions.
• Building the new sentiment pipeline on AWS reduced the cost of both its initial development and ongoing operations by 90 percent.

• Improved precision of sentiment analysis by 30%
• Reduced development and operations costs by 90%