Transformer Neural Network Personalized Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems fail to provide personalized product recommendations, as they do not adequately account for individual user preferences, leading to suboptimal user interactions.
Innovation Solution
A transformer neural network is trained using user activity data, singular value decomposition, word2vec neural networks, text-to-text transferring transformers, and highway networks to generate personalized item recommendations by creating user, item, and action vectors, which are then used to determine probabilities of user actions on recommended items.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional recommendation algorithms (SVD, bandit algorithms) are used, then the system can provide general recommendations, but the recommendations are not personalized enough on a per user basis
Solution Approach 1:
The patent segments the recommendation problem into multiple vector representations: user vectors capturing individual user preferences, item vectors representing product characteristics, and action vectors encoding user behaviors. This segmentation allows the system to personalize recommendations by combining these distinct vector types in a transformer neural network, directly addressing the lack of personalization in traditional algorithms.
Solution Approach 2:
The patent introduces a new dimensional framework by using transformer neural networks to process and integrate multiple vector types (user, item, action) that traditional algorithms do not combine. This dimensional expansion from simple item-user matching to multi-vector integration enables deeper personalization while improving the reliability of predicted user actions through more comprehensive analysis.
2Measurement precision
If advanced neural network techniques (transformer, word2vec, highway networks) are used to improve personalization, then the accuracy of recommendations increases, but the system complexity increases
Solution Approach 1:
The patent merges multiple neural network components (transformer architecture, word2vec embeddings, highway networks) into a unified recommendation system. By combining these techniques, the system achieves high recommendation accuracy through integrated processing of user, item, and action vectors, while the modular architecture allows for manageable complexity through systematic integration of proven components.
Solution Approach 2:
The patent uses vector representations as intermediaries between raw user activity data and final recommendations. These vectors serve as mediators that transform complex user behavior patterns into structured formats that the transformer neural network can process efficiently, reducing the computational complexity burden while maintaining high recommendation accuracy.
3Adaptability or versatility
If multiple vector types (user vectors, item vectors, action vectors) are generated and integrated, then personalized recommendations improve, but the computational processing requirements increase
Solution Approach 1:
The patent performs preliminary processing by generating user vectors, item vectors, and action vectors from raw data before feeding them into the transformer neural network. This pre-computation of vector representations organizes the data in advance, enabling the main recommendation engine to process personalized recommendations more efficiently and reducing the computational energy required during actual recommendation generation.
Data Source
AI summary
Systems, devices, and techniques are disclosed for recommendations using a transformer neural network. User activity data including items and actions associated with users and a catalog including descriptions of the items may be received. User vectors for the users, item vectors for the items and action vectors the actions may be generated by applying singular vector decomposition to the user activity data. Sequence vectors may be generated based on item vectors and the action vectors. Transformer vectors may be generated by applying a text-to-text transferring transformer to descriptions of the items. Similarity vectors may be generated based on the transformer vectors. Merged vectors may be generated by merging the sequence vector, transformer vector, and similarity vector for items. A set of probabilities may be determined by inputting the user vector for the user, merged vectors for the items, and sequence vectors for the actions to a transformer neural network.