Text Query Embedding via Visual Projection for Content Recommendation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current recommender systems face challenges in effectively processing text-based queries, as they are typically designed to handle non-text content items, leading to indirect and less relevant content recommendations when users communicate via text.
Innovation Solution
The system generates n-grams from text queries and uses a trained machine learning model to produce embedding vectors that project these n-grams into a content item embedding space, allowing for the identification of nearest content items that are most relevant to the user's request, thereby providing more accurate recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If an embedding vector generator is trained to accept image as input type, then it can generate meaningful embedding vectors for image content, but it produces meaningless embedding vectors when text is provided as input
Solution Approach 1:
The patent introduces an intermediary text processing component that converts text queries into visual representations (images or visual features) that can be processed by the image-trained embedding vector generator. This intermediary transformation allows the system to maintain the specialized image processing capability while gaining text input capability through conversion to the native image format.
Solution Approach 2:
The patent replaces the traditional text-processing mechanism with an image-based mechanism. Instead of using a text embedding model, the system substitutes text input with visual representation, allowing the image-trained embedding generator to process text queries by converting them into visual form first.
2Ease of operation
If recommender services perform indirect mapping of text-based queries to non-text content items, then they can respond to text queries, but the content recommendations become less relevant to the user's request
Solution Approach 1:
The patent replaces the indirect text-to-text mapping mechanism with a direct visual-to-visual matching mechanism. By converting text queries into image representations and comparing them directly with image content items in the corpus, the system establishes a direct visual similarity relationship that preserves query intent and improves recommendation relevance.
Solution Approach 2:
The patent changes the parameter space from text-based semantic similarity to image-based visual similarity. By transforming the query into the same parameter space (image embeddings) as the content items, the system enables direct comparison using image similarity metrics, which better captures the visual nature of the content being recommended.
Data Source
AI summary
Systems and methods for recommending content to an online service user are presented. In response to a request from a user, a set of n-grams of the request are generated, with each n-gram comprising one or more terms from the request and each n-gram of the set of n-grams being unique. Embedding vectors projecting the n-grams into a content item embedding space are generated, and the embedding vectors are combined into a representative embedding vector for the request. The nearest content items are identified according to a distance measure between a projection of the representative embedding vector and embedding vectors of content items of a corpus of content items in the content item embedding space. At least some of the nearest content items are returned as recommended content in response to the request from the user.