Project Requirements

Guangzhou Pearl River Digital Group Co., Ltd. (“Pearl River Digital”) is seizing the opportunity of three-network convergence, accelerating HD interactive digital TV, and turning the home television into a multimedia information terminal to give users a new digital-media experience. With many new-media services launching and spreading, it needs a data-analytics platform that can collect viewing-behavior data from all two-way users and all services, and establish a complete analysis and mining mechanism to better understand user needs and promote business growth.

This project needs to address the following issues:

1) Data collection, storage, and forwarding. Use big-data technology to store and manage massive, multi-source, diverse data, support linear hardware expansion, and provide fast, real-time analytics that can quickly act on the business;

2) Personalized user recommendation. Go beyond the analytic and decision value of the data itself by building a big-data platform, integrating business capabilities, and providing users with fused, personalized content recommendations.

3) From content delivery to content production. Use big-data mining to know audience needs ahead of viewers and anticipate programs that will be popular. Audience tagging of metadata such as actors, plot, tone, and genre can also reveal preferences for analysis and observation, preparing subsequent film and television production and other content development.

Solution Design

Based on Sugon’s years of experience in broadcasting and on big-data analytics platforms built for other customers in the industry, Sugon designed the overall logical architecture shown below:

The viewing-behavior analytics platform is designed in four layers: data sources; a data-preprocessing layer (extraction, transformation, masking, loading, reduction, and so on); a big-data support platform (data storage and processing); and a business-application layer.

First, data from the sources pass through a unified extraction and transformation platform for extraction, format conversion, and masking; ETL tools then load the cleaned data into the big-data platform for storage. Because data from multiple sources are aggregated, volumes are typically very large, so the big-data platform must scale well.

Second, data loaded into the big-data platform are used for final analysis and mining. Query task flows and further application systems can also be designed for specific analysis and mining needs and for the customer’s business.

Finally, the big-data application layer uses modeling to further analyze and mine data that have been initially processed on the platform, and visualizes results with big-data visualization tools. Presentation can take many forms, including charts and documents that are easy to understand.

The project needs to integrate data from set-top boxes, the BOSS system, operations systems, the media-asset system, and other sources, build a viewing-behavior analytics platform, complete statistical analysis of viewing-related data, raise user satisfaction, and improve the precision of advertising.

Follow-on construction will, based on actual business needs, launch business-facing analytics applications such as a real-time ranking system, a personalized recommendation system, and a new-media index analytics system, and will provide a common analytics framework to move toward analytics as a service (big data 2.0).

Solution Value

The project uses Sugon’s self-developed XData big-data processing platform, proven over many years of practice, to process massive viewing behavior in a timely, efficient way. Main platform features include:

1) Real-time analysis of viewing hotspots and trends for tens of millions of users;

2) Timely capture and analysis of tens of billions of set-top-box logs, VOD system logs, and other information;

3) User-feature extraction and audience segmentation, user- and program-based collaborative filtering, and an intelligent recommendation system based on rich tags.