Content Recommendation Embeddings for Cold-Start Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current recommendation methods, particularly those based on ID features and shallow neural networks, struggle with cold start scenarios and fail to effectively model complex and diverse content, leading to poor performance and inefficiencies in content recommendation systems.
Innovation Solution
A hierarchical machine learning model comprising a first and second machine learning model is used to determine content item embedding representations and recommend content items based on historical user behavior, utilizing language models to enhance text understanding and reduce input sequence length.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ID-based embedding methods are used for content recommendation, then the model size remains small and computation is efficient, but the performance in cold start scenarios deteriorates and complex content modeling capability is insufficient
Solution Approach 1:
The patent segments the content representation into two parts: ID-based embedding (for efficient computation and known content) and text-based embedding (for semantic understanding and cold start). This segmentation allows the system to leverage both approaches without fully committing to the complexity of pure text models.
Solution Approach 2:
The patent merges ID-based embedding and text-based embedding into a unified content representation. The final embedding is computed as a combination of the ID embedding and the text embedding, allowing the system to benefit from both the efficiency of ID methods and the semantic power of language models.
2Productivity
If shallow neural networks with ID features are used, then computational efficiency is maintained, but the ability to model diverse and complex content deteriorates
Solution Approach 1:
The patent introduces text description as an intermediary that bridges the gap between simple ID features and complex content understanding. The text description is processed by a language model to generate semantic embeddings, which then serve as a mediator to enhance the basic ID-based representation without requiring the full complexity of end-to-end text processing.
3Measurement precision
If language models are applied to process content descriptions, then text understanding capability is improved, but input sequence length and computational complexity increase
Solution Approach 1:
The patent applies partial action by using language models only for processing the text description portion of content, rather than processing entire long sequences. The system selectively applies the language model to extract semantic features from descriptions, achieving sufficient text understanding without processing unnecessary tokens that would increase computation time.
Data Source
AI summary
A method, an apparatus, a device, a storage medium, and a program product for content recommendation are provided. The method includes: obtaining a content item sequence associated with historical behavior data of a target user, the content item sequence including a plurality of content items for which the target user sequentially performs conversion behavior; determining, by using a first machine learning model and respectively based on a first prompt element and description information of each of the plurality of content items, a plurality of content item embedding representations respectively corresponding to the plurality of content items, the first prompt element indicating extraction of a corresponding content item embedding representation from the description information of each content item; and determining, by using a second machine learning model and based on at least the plurality of content item embedding representations, a recommended content item to be recommended to the target user.