Group Recommender Mutual Information Maximization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing group recommendation systems struggle to provide effective item recommendations for ephemeral groups with sparse interaction histories, as they often ignore individual user preferences or assume uniform likelihoods across different groups, leading to degenerate solutions.

Innovation Solution

The proposed solution involves training neural networks to maximize user-group mutual information and contextually weight user preferences, creating a latent user-group embedding space that captures shared preferences and social associations, thereby providing robust item recommendations for ephemeral groups without requiring external side information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional group recommendation systems use heuristic pre-defined aggregation or data-driven strategies, then they can handle persistent groups with substantial activity history, but they fail to provide effective recommendations for ephemeral groups with sparse interaction histories

Engineering Contradiction:

Improverecommendation accuracyVSAvoidapplicability to ephemeral groups

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent changes the fundamental parameters of the recommendation approach by switching from aggregation-based methods to mutual information maximization. This involves transforming the objective function from aggregating individual preferences to directly optimizing the information shared between user preferences and group recommendations, enabling effective handling of sparse data in ephemeral groups

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces the mechanical aggregation process (combining individual ratings or preferences through predefined rules) with an information-theoretic approach. Instead of mechanically averaging or weighting individual preferences, the system uses mutual information to capture the underlying preference structure, substituting a data-driven information maximization mechanism for traditional aggregation mechanics

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Ease of manufacture

If group recommendation systems aggregate individual preferences using uniform weighting, then the implementation is simple, but they ignore the contextual importance of different group members and produce degenerate solutions

Engineering Contradiction:

Improveimplementation simplicityVSAvoidpreference capture accuracy

Core Design Contradiction:

Ease of manufactureVSManufacturing precision

Solution Approach 1:

The patent applies local quality by making the weighting of individual preferences context-dependent rather than uniform. Each group member's preferences are weighted according to their specific contribution to the group's preference structure, as measured by mutual information. This allows the system to adaptively emphasize or de-emphasize individual preferences based on their local importance in each specific group context

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent introduces dynamics by making the preference aggregation weights adaptive rather than static. The mutual information maximization process dynamically determines the appropriate weight for each user's preferences based on the observed data, allowing the system to flexibly adjust to different group compositions and interaction patterns rather than applying fixed uniform weights

Inventive Principle:

Principle #15Dynamics

3Reliability

If external side information such as social networks or user personality traits is used to improve group recommendations, then recommendation quality may improve, but the system complexity increases and side information may be unavailable

Engineering Contradiction:

Improverecommendation qualityVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies self-service by enabling the system to extract and utilize all necessary information from the available interaction data itself, without requiring external side information. The mutual information maximization approach allows the system to self-determine the appropriate user preferences and group dynamics from the observed user-item interactions, making the system self-sufficient and avoiding dependency on external data sources

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS11443346B2Group item recommendations for ephemeral groups based on mutual information maximization

Publication Date: 2022.09.13 VISA INTERNATIONAL SERVICE ASSOCIATION

AI summary

A computer-implemented method is disclosed for training neural networks of a group recommender to provide item recommendations for ephemeral groups having group interaction sparsity. A preference encoder and aggregator generate user and group preference embeddings from user-item interactions, wherein the preference embeddings form a latent user-group latent embedding space. The neural preference encoder and the aggregator are trained by regularizing the latent user-group embedding space to overcome the group interaction sparsity by: i) maximizing user-group mutual information (MI) between the group embeddings and the user embeddings so that the group embeddings encode shared group member preferences, while regularizing the user embeddings to capture user social associations, and ii) contextually identifying informative group members and regularizing the corresponding group embeddings using a contextually weighted user loss value to contextually weight users' personal preferences in proportion to their user-group MI to reflect personal preferences of the identified informative group members.