Group-Aware Recommendation System Using Watched-With Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current recommender systems for TV viewing do not effectively account for group interactions and preferences when multiple users watch together, leading to inflexible and non-scalable recommendations, as they rely solely on individual profiles without considering the influence of group dynamics.
Innovation Solution
A recommendation system that records and utilizes 'watched with' information to generate recommendations for groups by creating a combined watching history, weighing watch events based on whether multiple users watched jointly, and applying this history to existing recommender systems using collaborative or content-based approaches.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If individual user profiles are used for recommendations, then the system is simple to implement, but it cannot account for group interactions and preferences when multiple users watch together
Solution Approach 1:
The patent merges individual user profiles with group context by creating composite group profiles that aggregate preferences from multiple individual users. The system combines data from individual profiles with group interaction data to generate recommendations that reflect both individual and collective preferences, resolving the contradiction between simplicity and adaptability to group dynamics.
Solution Approach 2:
The patent segments the recommendation system into multiple layers: individual user profiles, group profiles, and context-aware components. This segmentation allows the system to maintain simple individual profiles while adding group-level complexity only where needed, enabling group interaction accounting without completely redesigning the entire system.
2Measurement precision
If group profiles are created for every possible group composition, then recommendations can be accurate for specific groups, but the system becomes non-scalable with factorial growth in profiles needed
Solution Approach 1:
The patent creates universal group profiles that can serve multiple group compositions rather than creating separate profiles for each specific group. A single group profile can be applied to various combinations of users who share similar viewing patterns and preferences, eliminating the need for factorial growth in profile creation while maintaining recommendation accuracy for different group configurations.
Solution Approach 2:
The patent changes the parameters used to define groups from exact user composition to preference-based clustering. Instead of creating profiles for specific user combinations (N!/(k!(N-k)!) profiles), the system uses preference parameters and viewing behavior patterns to group users, allowing the same profile to apply across multiple user compositions with similar preferences.
3Adaptability or versatility
If individual profiles are used without group context, then data records are simple to manage, but recommendations do not reflect the influence of group dynamics on user preferences
Solution Approach 1:
The patent introduces group context data as an intermediary layer between individual profiles and recommendation generation. This intermediary captures group interaction patterns and preferences without requiring complex modifications to individual profile structures. The group context acts as a mediator that translates individual preferences into group-level recommendations, maintaining data simplicity while enabling group preference accounting.
Data Source
Figure 1a~1b
Figure 2
Figure 3
AI summary
System for outputting a choice recommendation to one or more users based on earlier choices made by them. A data record is made per choice per user, each data record including a representation of the user, a representation of the choice and a representation of any other user involved with that choice, such as other users that watched a program with the user. The data record may include a representation of the user, a representation of the choice and a representation of no more than one other user involved with that choice, where more records are made when more other users are involved. The data record may include a representation of an environment linked to the choice, e.g. of a localization and/or hardware and/or software linked to the choice. The recorded representation of other users involved with the choice, such as other users that watched a program with the user, is used later to compute the choice recommendation. A synthetic set of watching events for a detected group of users may be determined for this purpose, based on the identity of the members of the group and the representation of "watch with" users of the watching event.