Non-Time-Series ML for Real-Time User Intent Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional intent determining techniques fail to accurately predict user intent due to their inability to process time-related information effectively, while time-series-based machine learning models are slow and complex, making it challenging to train models with historical data properly weighted by temporal information in real-time.
Innovation Solution
A non-time-series-based machine learning system that determines user intent by using time-based feature encoding, which includes positional encoding and a decay function to weight events based on their temporal relevance, allowing for real-time intent prediction through a pre-trained machine learning model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If time-series-based machine learning models are used to incorporate time-related information, then prediction accuracy is improved, but system complexity and processing speed decrease
Solution Approach 1:
The patent segments the time-series data processing into two distinct phases: an offline training phase where temporal patterns are learned from historical data, and an online inference phase where the pre-trained model makes predictions. This segmentation allows the complex time-series processing to be performed once during training, while keeping the online prediction simple and fast.
Solution Approach 2:
The patent applies preliminary action by pre-training the machine learning model on historical data with temporal information before actual use. The model learns temporal patterns and relationships during this preliminary training phase, so that during real-time operation, it can quickly make predictions without reprocessing the complex time-series data.
2Measurement precision
If time-series-based machine learning models are used to incorporate time-related information, then prediction accuracy is improved, but processing speed decreases
Solution Approach 1:
The patent segments the computational workload between offline training (where time-series processing occurs) and online inference (where simple predictions are made). This allows the system to achieve both high accuracy through temporal pattern learning and fast real-time response through simplified prediction execution.
Solution Approach 2:
By performing the complex time-series analysis and model training as a preliminary offline action, the system prepares a pre-trained model that can execute quickly during real-time operation. The heavy computational burden is shifted to the training phase rather than the inference phase.
3Device complexity
If conventional intent determining techniques are used, then system simplicity is maintained, but temporal information processing capability is lost
Solution Approach 1:
The patent introduces an intermediary mechanism in the form of a pre-trained machine learning model that bridges the gap between simple data input and accurate temporal pattern recognition. The model acts as an intermediary that encapsulates temporal reasoning capabilities, allowing the system to maintain simplicity in its data processing pipeline while achieving sophisticated temporal analysis.
Data Source
AI summary
A method implemented via execution of computing instructions configured to run at one or more processors and stored at one or more non-transitory computer-readable media. The method can include receiving, via a computer network, an intent prediction request from a frontend system. The method further can include obtaining, from a database, one or more events in a lookback period associated with one or more items ordered by a user for the intent prediction request. The method also can include determining a time-based feature encoding for the one or more events for the user by: (a) determining a feature encoding for the one or more events; (b) determining a positional encoding for the one or more events; and (c) determining the time-based feature encoding based at least in part on the feature encoding, the positional encoding, and a decay function. The positional encoding can include one or more positional vectors associated with a temporal sequence of the one or more events. The method further can include determining, in real-time via a machine learning model, a user intent for the user based on the time-based feature encoding. Other embodiments are described.