Activity Ranking System Using Multi-Horizon List Combination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating ranked lists of activities based on user behavior struggle to accurately predict next activities due to the challenge of selecting an appropriate time horizon, which can lead to missing either short-term or long-term trends, affecting the accuracy of predictions.
Innovation Solution
A method that combines multiple lists with different time horizons to form a single, more accurate ranked list by updating scores and rankings based on recent activities, using a combination of temporal decay factors and sliding window procedures to capture both short-term and long-term trends, and providing this combined list to users via a computer interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Duration of action of stationary object
If a long time horizon is chosen for generating ranked lists, then long term trends are captured, but short but strong trends (micro-trends) are missed
Solution Approach 1:
The patent divides the single ranked list generation task into multiple segments, each corresponding to a different time horizon. Multiple lists are generated with different time horizons (e.g., short-term, medium-term, long-term), and each list captures trends appropriate to its time horizon. These segmented lists are then combined to provide comprehensive predictions that capture both micro-trends and long-term trends.
Solution Approach 2:
The patent adds a new dimension to the prediction system by introducing multiple time horizons as separate lists. Instead of using a single time horizon (one-dimensional approach), the system creates multiple lists along the time horizon dimension, each optimized for different temporal patterns. This multi-dimensional approach allows simultaneous capture of short-term and long-term trends without compromise.
2Measurement precision
If a short time horizon is chosen for generating ranked lists, then short term trends are captured, but long term trends are missed
Solution Approach 1:
The system segments the prediction task into multiple lists, each responsible for a specific time horizon. Short-term lists capture recent micro-trends, while long-term lists capture sustained trends. This segmentation allows each list to be optimized for its specific temporal range, improving overall prediction accuracy across all time scales.
Solution Approach 2:
By introducing multiple time horizon dimensions, the system can simultaneously maintain short-term and long-term perspectives. Each list represents a different slice along the time horizon dimension, allowing the system to capture trends at multiple temporal scales without sacrificing accuracy in any single dimension.
3Measurement precision
If multiple lists with different time horizons are combined, then prediction accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent merges multiple lists generated from different time horizons into a single combined ranked list. The merging process integrates the predictions from short-term, medium-term, and long-term lists using combination strategies (such as weighted averaging or voting mechanisms), producing a unified list that benefits from diverse temporal perspectives while presenting a single coherent output to users.
Solution Approach 2:
The system creates a multi-functional framework where a single prediction mechanism serves multiple time horizon requirements. The same underlying prediction algorithm is applied across different time horizons, and the results are universally combined into one list that satisfies diverse prediction needs, reducing the need for separate specialized systems for each time horizon.
Data Source
AI summary
There is provided a method of providing a ranked list of computer-implemented activities to a user. As a performed activity is received, a plurality of lists, which each of the lists defines a ranking of activities, is accessed. The rankings of different lists are calculated with respect to different time horizons. The positions in the lists are associated with scores which are updated in view of the received activity. Also, the rankings of the lists are updated in view of the received activity. The lists are then combined into a single list. The combination is based on the updated lists and the scores associated with the positions in the lists.