Image Recommendation via User Collection Significance Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Selecting relevant content for user recommendations in content hosting services is challenging, particularly in digital art, due to the difficulty in predicting users' varied visual tastes.
Innovation Solution
A method and system for recommending images to users based on user-defined collections, where a seed image is used to identify candidate images through their shared collections, with significance scores determined by the number of shared collections and properties such as categories and tags, to rank and recommend images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If content hosting services recommend content based on user browsing history and general algorithms, then recommendations can be generated automatically, but the relevance to users' specific visual tastes deteriorates
Solution Approach 1:
The patent introduces user-defined collections as an intermediary between automatic algorithms and user visual tastes. These collections serve as a mediator that captures users' subjective aesthetic preferences in a structured format, allowing the system to leverage both automation and personalization. The collections act as a bridge that translates implicit user tastes into explicit recommendation criteria.
Solution Approach 2:
The patent requires users to pre-organize their favorite content into collections before the recommendation process. This preliminary action of curating collections captures users' visual tastes in advance, creating a personalized reference framework that the recommendation algorithm can then utilize. The system performs recommendation based on these pre-established user preferences rather than attempting to infer tastes in real-time.
2Measurement precision
If the system uses multiple properties (categories, tags, collections) to rank candidate images, then recommendation relevance improves, but system complexity increases
Solution Approach 1:
The patent segments the recommendation process into distinct functional modules: a candidate identification module that filters images based on collection membership, a significance score determination module that evaluates candidate relevance, and a ranking module that orders candidates. This segmentation allows each module to handle specific aspects of the complexity independently, making the overall system more manageable while achieving high recommendation precision.
Solution Approach 2:
The patent transforms multiple qualitative properties (collection membership, categories, tags) into a quantitative significance score that can be used for straightforward ranking. By converting diverse properties into a unified numerical metric, the system simplifies the comparison and selection process while retaining the richness of multiple evaluation dimensions.
3Measurement precision
If the system retrieves and processes multiple properties for each candidate image, then recommendation accuracy improves, but processing time increases
Solution Approach 1:
The patent performs preliminary filtering by collection membership before detailed property analysis. The candidate identification module quickly narrows down the search space by identifying only those images that belong to the user's seed collections, eliminating the need to process properties of irrelevant images. This preliminary action significantly reduces the number of images requiring full property evaluation.
Solution Approach 2:
The patent segments property retrieval into two stages: first retrieving collection membership information for all candidate images, then retrieving detailed properties (categories, tags) only for those that pass the collection filter. This segmented approach ensures that computationally expensive property processing is applied only to a small subset of relevant candidates rather than the entire image library.
Data Source
AI summary
Each image of a plurality of user-defined collections of images has properties associated therewith. A seed image is defined as currently viewed or selected image and the properties associated with the seed image are retrieved. The seed image is used to identify seed collections as the collections of images which have the seed image as one of their images. A plurality of candidate images are identified from the seed collections. For each candidate image, a significance score is determined as a function of either the number of seed collections to which the candidate image belongs or the ratio of the number of seed collections to the set of all collections to which the candidate image belongs. Each candidate image is ranked based on the significance scores and the seed image properties. At least one candidate image is recommended to the user based on the ranking of the candidate images.