Hierarchical Event Feature Prediction for Sparse Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for predicting events and controlling systems fail to fully exploit knowledge about hierarchical event features, leading to inefficiencies in situations with sparse or unseen data, particularly in large-scale applications like web search engines and online advertising.
Innovation Solution
A graphical data structure is created to represent hierarchical relationships between event features, allowing for online training and incremental learning, which enables the system to make improved predictions and control decisions by propagating knowledge through the structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing prediction systems are used without hierarchical feature representation, then the system complexity remains low, but the prediction quality deteriorates when data is sparse or unseen
Solution Approach 1:
The patent segments the feature space into hierarchical levels (e.g., URL prefix levels such as domain, subdomain, path components). Each level represents a granularity of feature detail, allowing the system to break down complex features into manageable hierarchical segments that can be processed independently and combined for predictions.
Solution Approach 2:
The patent adds a hierarchical dimension to the feature representation by organizing features into multiple levels of abstraction. This transforms the feature space from a flat structure to a multi-dimensional hierarchical structure, enabling the system to leverage relationships at different levels of granularity for improved prediction accuracy.
2Reliability
If hierarchical feature representation is implemented, then the system can handle sparse and unseen data better, but the device complexity increases
Solution Approach 1:
The patent pre-establishes the hierarchical feature structure and relationships before prediction tasks are performed. By organizing features into hierarchical levels in advance and pre-computing relationships between levels, the system is prepared to handle sparse and unseen data efficiently without requiring complex runtime computations.
Solution Approach 2:
The patent introduces intermediate hierarchical levels as mediators between raw features and final predictions. These intermediate representations (e.g., aggregated statistics at different URL prefix levels) serve as bridges that enable the system to infer properties of unseen features based on their hierarchical relationships with observed features.
3Adaptability or versatility
If online training with knowledge propagation is used, then the system adapts dynamically to hierarchical relationships, but the training complexity and time increase
Solution Approach 1:
The patent implements continuous online training that propagates knowledge through the hierarchical structure as data arrives. Instead of periodic batch training, the system continuously updates hierarchical feature representations and propagates learned patterns through the hierarchy, enabling dynamic adaptation without stopping operation.
Solution Approach 2:
The patent incorporates feedback mechanisms where prediction outcomes and observed interactions feed back into the hierarchical feature structure. This feedback drives iterative refinement of hierarchical representations and propagation rules, allowing the system to adapt to emerging patterns in the data over time.
4Productivity
If hierarchical feature clustering is applied to large-scale applications, then the system can process more features, but the computational resources and processing time increase
Solution Approach 1:
The patent merges features at the same hierarchical level and combines information across levels to create compact hierarchical representations. By merging redundant or related features into hierarchical groups, the system reduces the total number of features that need to be processed independently while preserving the essential information for predictions.
Solution Approach 2:
The patent creates universal hierarchical feature representations that serve multiple prediction tasks simultaneously. The same hierarchical structure and propagated knowledge can be applied to different types of predictions and applications, reducing the need for separate processing pipelines and lowering overall computational requirements.
Data Source
AI summary
Event prediction using hierarchical event features is described. In an embodiment a search engine monitors search results presented to users and whether users click on those search results. For example, features describing the search result events are universal resource locator prefix levels which are inherently hierarchically related. In an embodiment a graphical data structure is created and stored and used to represent the hierarchical relationships between features. An online training process is used in examples which enables knowledge to be propagated through the graphical data structure according to the hierarchical relations between features. In an example, the graphical data structure is used to predict whether a user will click on a search result and those predictions are used by the search engine to rank search results for future searches. In another example the events are advertisement impressions and the predictions are used by an online advertisement system.