Content Recommendation System Using Image and Text Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing content delivery systems struggle to personalize secondary content, such as advertisements and content recommendations, to individual consumers based on their unique watching histories, often providing generic content that may not interest specific viewers.
Innovation Solution
The system determines secondary content by analyzing a consumer's watching history through text and image data encoding, using similarity metrics to match content that shares similar text, images, or both, thereby increasing the likelihood of consumer interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If generic secondary content is delivered to consumers, then device complexity and processing requirements are reduced, but consumer engagement and interest in the content decrease
Solution Approach 1:
The system pre-processes and encodes text and image data from content items into consolidated feature sets before delivery. This preliminary encoding of content features enables rapid matching and personalization during content delivery without requiring complex real-time processing, thus resolving the contradiction between personalization capability and system complexity
Solution Approach 2:
The patent introduces consolidated feature sets as an intermediary representation layer between raw content data and the personalization algorithm. These feature sets act as a mediator that simplifies the matching process between consumer preferences and content recommendations, reducing the computational complexity while maintaining adaptability
2Productivity
If personalized secondary content is determined using watching history analysis, then consumer engagement increases, but processing time and computational resources increase
Solution Approach 1:
The system performs text encoding and image encoding of content features in advance, creating consolidated feature sets that can be quickly compared against consumer watching histories. This pre-processing eliminates the need for time-consuming real-time analysis during content delivery, thus maintaining high consumer engagement while minimizing determination time
Solution Approach 2:
The patent extracts only the essential text and image features from content items to create consolidated feature sets, rather than processing entire content items. This extraction of key features reduces the computational burden and processing time while still enabling effective personalization based on consumer watching histories
Data Source
AI summary
A request associated with a user for content recommendation may be determined. At least one content item indicative of a viewing history associated with the user may be determined. The at least one content item indicative of the viewing history may include content that the user has previously watched. Data associated with at least one image associated with each of a plurality of candidate content items may be determined. The plurality of candidate content items may include secondary content items that may be recommended to the user. Based on comparing data associated with at least one image associated with the at least one content item with the data associated with the at least one image associated with each candidate content item, at least one candidate content item may be determined. An indication of the at least one candidate content item may be sent, to a device associated with the user.