Recommendation System Diversity via Conditional Interest Scores
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommendation systems often return redundant results due to their methods of identifying data objects, which can obscure more dissimilar results of interest to the user, especially when the data set contains many similar objects.
Innovation Solution
The method involves calculating a conditional interest-likelihood score for each data object based on the assumption that if certain results are not selected, to identify a diverse set of results that maximize the likelihood of user interest, using a formula like ITOTAL=I(select R1)+I(select R2 given no selection of R1)+ . . . +I(select RK given no selection of any of R1 . . . RK-1), which considers the relatedness and marginal interest-likelihood scores to present a diverse result set.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional identification algorithms are used to select results, then the likelihood of user interest in individual results is maximized, but the results become redundant and similar
Solution Approach 1:
The patent changes the selection parameter from marginal interest-likelihood scores to conditional interest-likelihood scores. Instead of selecting results based on their individual interest scores, the system selects results based on the conditional probability of user interest given that previous results were not selected. This parameter transformation enables the system to maximize both reliability (user interest) and versatility (diversity) by considering the impact of each selection on the remaining pool of results.
2Quantity of substance
If multiple similar data objects are included in results, then the coverage of similar topics is improved, but the visibility of dissimilar results is reduced
Solution Approach 1:
The patent implements a feedback mechanism where the selection of each result updates the conditional probability calculations for remaining results. After selecting a result, the system recalculates interest-likelihood scores for remaining results based on the assumption that the selected result was not chosen, effectively feedback-adjusting the selection criteria. This ensures that dissimilar results maintain visibility while still covering important topics.
3Reliability
If results are selected based on highest interest scores, then the most interesting results are presented, but redundant similar results are selected instead of diverse alternatives
Solution Approach 1:
The patent makes the selection process dynamic by iteratively updating interest-likelihood scores based on previous selections. The conditional interest-likelihood score for each result changes dynamically as results are selected, reflecting the changing pool of available results. This dynamic adjustment ensures that the system adapts its selection criteria to prioritize diversity while maintaining high interest-likelihood throughout the selection process.
Data Source
AI summary
Methods, systems and apparatus, including computer program products, for providing a diversity of recommendations. According to one method, results are identified so as to increase the likelihood that at least one result will be of interest to a user. Following the identification of a first result, second and later results are identified based on an assumption that the previously identified results are not of interest to the user. The identification of diverse results can be based on formulas that approximate the probability or provide a likelihood score of a user selecting a given result, where a measured similarity between a given object and previously identified results tends to decrease the calculated probability approximation or likelihood score for that object.