Dynamic Content Recommendation via Message Data Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Content serving platforms face challenges in accurately recommending content items to users due to a lack of information about user engagement with certain content items, especially when content providers do not include pixels to track user interactions, limiting the platform's ability to train user engagement models and resulting in irrelevant content being served to users.
Innovation Solution
The platform utilizes a two-model approach, where a first model serves exploration traffic to gather engagement feedback for new content items and a second model serves non-exploration traffic once adequately trained, leveraging content item information extracted from message data to rank and promote content items based on popularity and attributes, allowing for dynamic content item recommendations across a broader audience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the platform relies on pixel tracking from content providers to train user engagement models, then measurement precision of user engagement is improved, but device complexity and ease of operation deteriorate because content providers must implement and maintain tracking pixels
Solution Approach 1:
The platform performs self-service by extracting content item information directly from message data without requiring content providers to implement tracking pixels. The system autonomously analyzes message content, attachments, and metadata to identify content items and infer engagement signals, eliminating the need for external tracking infrastructure
Solution Approach 2:
Message data serves as an intermediary source that bridges the gap between content providers and the platform. Instead of directly tracking user interactions through pixels, the system uses message data (emails, text messages, social media messages) containing content item information as a proxy to infer user engagement and train models
2Adaptability or versatility
If the platform serves exploration traffic with new content items lacking engagement data, then adaptability of content recommendations is improved, but manufacturing precision of recommendation accuracy deteriorates
Solution Approach 1:
The system performs preliminary action by extracting and analyzing content item information from message data before serving recommendations. By pre-processing message data to identify content items, their attributes, and implicit engagement signals, the system prepares training data in advance, enabling accurate model training even for new content items without historical interaction data
Solution Approach 2:
The system uses partial action by leveraging only the portion of message data that contains content item information and engagement signals, rather than requiring complete traditional tracking data. This allows the system to work with incomplete but sufficient information to generate accurate recommendations for new content
3Loss of information
If the platform extracts content item information from message data, then loss of information is reduced by capturing engagement signals without pixels, but measurement precision of traditional engagement metrics deteriorates due to indirect measurement
Solution Approach 1:
Message data acts as an intermediary that indirectly captures engagement information. The system extracts content item information from messages (sender, recipient, timing, content) and uses this intermediary data to infer engagement metrics, achieving information capture without direct pixel tracking
Solution Approach 2:
The system replaces the mechanical pixel tracking system with an information extraction approach. Instead of using visual pixels that track user actions on webpages, the system uses natural language processing and data extraction from message content to infer engagement, substituting a different technological mechanism
Data Source
AI summary
One or more computing devices, systems, and/or methods for generating dynamic content item recommendations are provided. Content item information, extracted from message data, is aggregated to calculate popularity and attributes of content items. The content items are ranked based upon the popularity and attributes to generate a ranked list of content items. Exploration traffic is served utilizing a set of eligible content items selected from the ranked list of content items. An eligible content item is promoted for participation in auctions for serving non-exploration traffic based upon the eligible content item being served a threshold number of times.