Cloud Music Public Opinion Analysis Platform: Architecture and GPT-Based Implementation

This article introduces the construction of NetEase Cloud Music's public opinion analysis platform, addressing the limitations of generic public opinion analysis which can only perform sentiment or trend analysis on specific topics without deep information mining.

Data Characteristics

The platform leverages diverse data sources including external channels (social media, news, blogs) and internal sources (APP feedback, song comments, Qiyu customer service data). These data sources feature higher relevance, real-time feedback, and more structured information including user, device, and system data.

Analysis Capabilities

The platform provides multi-dimensional analysis including: clustering analysis (mapping feedback to product function trees), feedback type analysis (problem feedback, product suggestions, usage inquiries, complaints), summary extraction, and sentiment analysis (positive, negative, neutral).

Technical Architecture

The platform consists of: Data Ingestion layer (MQ/HTTP protocols), Processing layer (Adapter for standardization, Analyzer for multi-dimensional analysis), Data Management, Analysis & Visualization layer, and Monitoring & Alerting layer. Data is stored in Elasticsearch for real-time query and analysis.

Analysis Engine

The analysis engine uses GPT-based built-in analyzers, offering advantages in semantic understanding, no requirement for labeled training data, and ability to discover new patterns. Cost optimization strategies include caching analysis results and using text similarity algorithms to filter relevant clusters before GPT analysis, reducing token consumption.

Monitoring and Alerting

The platform implements intelligent monitoring with three components: data filtering (supporting all attributes), alerting conditions (dynamic thresholds based on historical trends), and alert receivers (IM, SMS, phone, email). Smart alerting automatically creates rules with dynamically calculated thresholds refreshed periodically.

Future Directions

Future development will leverage GPT to provide intelligent daily/weekly reports, automatically summarizing and analyzing periodic public opinion data to reduce manual analysis costs.