Case Study: Major Telecom Company achieves real-time multi-lingual classification and 99.99% accuracy with Gathr

Real-time multi-lingual sentiment analysis for a top U.S. telecom provider

Major Telecom Company needed a real-time, multi-lingual text classification and sentiment-analysis system capable of ingesting and parsing very high volumes (250M / 15 TB records per day), storing and indexing petabytes of data, and meeting strict SLAs (5s cold-data query) and high accuracy targets (four nines). Gathr was engaged to provide the solution, leveraging the StreamAnalytix platform and best-of-breed streaming and search technologies to address these scale, latency and language challenges.

Gathr implemented a three-module architecture—an Analytics Module with a low-latency, matrix-decomposition text classifier and enhanced feature-level sentiment analysis; an Event Store/Indexer abstraction for scalable storage and indexing; and a Publish Module to deliver results to downstream systems—using technologies such as Apache Kafka, HBase, Elasticsearch, Apache Storm and R. The solution delivered rapid, accurate real-time categorization and multi-lingual sentiment insights, adjustable domain-specific classes, linear cluster scalability and the ability to add custom components, enabling the Major Telecom Company to process 250M (15 TB) records/day and meet the required low-latency/query SLA.