Neural Network Sub-models for Flexible Content Discovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional content discovery systems are limited in flexibility and effectiveness, primarily focusing on TV program-level interactions, failing to accommodate user preferences beyond specific titles and lacking in handling queries for entities like channels or moods.
Innovation Solution
A computer-implemented method using a trained neural network with sub-models for attribute types, allowing for the processing of user queries by combining relevancy parameters across different attribute types to provide ranked associations and recommendations for content items, enabling more flexible and accurate content discovery.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional search and recommendation services are used, then users can find TV programmes, but the system lacks flexibility to handle queries for entities like channels or moods
Solution Approach 1:
The patent segments the content discovery system into multiple independent sub-models, each dedicated to a specific attribute type (programme, channel, contributor, mood, tag). Each sub-model contains nodes representing attribute values and learned weights representing associations. This segmentation allows the system to handle different entity types flexibly while maintaining manageable complexity through modular architecture.
Solution Approach 2:
The patent creates a universal content discovery system that can handle multiple entity types (programmes, channels, contributors, moods, tags) through a common neural network framework. The same sub-model structure and association learning mechanism are applied across all entity types, enabling the system to provide ranked associations for any queried entity type rather than being limited to specific TV programme searches.
2Adaptability or versatility
If content discovery is limited to TV programme level, then the system is simpler, but it fails to accommodate user preferences beyond specific titles
Solution Approach 1:
The patent extends the content discovery system from a single dimension (TV programme titles) to multiple dimensions by introducing separate sub-models for different attribute types (programme, channel, contributor, mood, tag). Each sub-model adds a new dimension of user preference that can be independently queried and combined, allowing users to express preferences at multiple levels of abstraction rather than being constrained to programme titles only.
3Adaptability or versatility
If a comprehensive neural network model is trained for all attribute types, then query flexibility improves, but processing time and computational resources increase
Solution Approach 1:
The patent divides the comprehensive neural network into multiple smaller, specialized sub-models, each handling a specific attribute type. This segmentation reduces the computational burden on each individual model during training and inference, as each sub-model only needs to learn associations within its specific domain rather than processing all attribute types simultaneously.
Solution Approach 2:
The patent retrieves and combines only the specific sub-models needed for the current query rather than processing the entire comprehensive model. When a user queries for associations related to a specific entity type, the system selectively activates only the relevant sub-model(s), performing partial action that is sufficient for the current task while avoiding unnecessary computational overhead from other sub-models.
Data Source
AI summary
Systems and methods for presenting content to a user are described. A trained neural network is stored in memory, defining input nodes representing respective attribute values of attribute types, and weights embodying strengths of connections between the input nodes and hidden nodes, as trained for the particular user. Sub-models of the neural network are defined from sets of input nodes of the same attribute type and a corresponding hidden state matrix of trained weights. A request for content assets is processed using a retrieved first sub-model corresponding to a query attribute type and second sub-model corresponding to a target attribute type, to determine relevancy parameters for the user. Content assets are selected for presentation by a media receiver device, based on the determined relevancy parameters. Other embodiments are also described and claimed.