Probabilistic Location History Analysis for Consumer Behavior Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems face challenges in efficiently analyzing large datasets of consumer location histories to understand consumer behavior effectively, particularly due to issues with data-intensive tiles and devices that cause concurrency and memory problems, and the complexity of algorithms used.
Innovation Solution
The approach involves obtaining and analyzing time-stamped geolocations, using probabilistic models to infer reasons for user movement between geographic areas, and employing pre-computed tables and distributed caching to expedite analysis, along with data clustering algorithms like DBSCAN and Hidden Markov Models to extract meaningful patterns from location data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional algorithms are used to analyze consumer location histories, then analysis completeness can be maintained, but processing speed and memory efficiency deteriorate due to data-intensive operations and concurrency problems
Solution Approach 1:
The patent segments the location history data into discrete geographic areas and time intervals, creating a structured grid system. This segmentation allows the analysis to process data in manageable units rather than handling entire datasets at once, improving processing speed while reducing memory requirements. The geographic space is divided into tiles, and temporal data is segmented into sequential time steps, enabling efficient parallel processing.
Solution Approach 2:
The patent employs pre-computed tables that store previously calculated transition probabilities and geographic area characteristics. By performing computations in advance and storing results in lookup tables, the system avoids redundant calculations during actual analysis, significantly improving processing speed. The pre-computed tables contain aggregated statistics that can be quickly retrieved without re-processing raw location data.
2Measurement precision
If detailed consumer location data is collected, then behavioral insight accuracy improves, but data storage requirements and processing memory increase
Solution Approach 1:
The patent extracts only the essential features from detailed location data by identifying and focusing on transitions between geographic areas rather than processing every individual location point. The system extracts key behavioral patterns such as visit frequencies, dwell times, and transition sequences, discarding redundant granular location information. This extraction maintains behavioral insight accuracy while reducing data volume for storage and processing.
Solution Approach 2:
The patent transforms detailed location coordinates into discrete geographic area identifiers and temporal parameters. By changing the representation from continuous coordinate data to discrete area-time transitions, the system reduces data complexity while preserving essential behavioral information. The transformation involves parameterizing location data into categorical variables that capture behavioral patterns without requiring storage of precise coordinate information.
3Measurement precision
If complex probabilistic models are implemented to infer user behavior, then inference accuracy improves, but computational resource requirements increase
Solution Approach 1:
The patent pre-computes transition probabilities and stores them in lookup tables before actual inference operations. By calculating these probabilistic parameters in advance using batch processing, the system avoids performing complex probabilistic calculations during real-time inference. This preliminary computation reduces the computational energy required during actual behavior inference while maintaining high accuracy through the use of pre-analyzed probabilistic models.
Solution Approach 2:
The patent uses pre-computed tables that contain copied and stored probabilistic relationships derived from aggregated user behavior data. Instead of performing fresh probabilistic calculations for each inference query, the system retrieves pre-established probability distributions and transition models from stored tables. This copying approach maintains inference accuracy by using validated probabilistic models while dramatically reducing computational energy requirements.
Data Source
AI summary
Provided is a process of inferring a user's reason for movement between geolocations sensed by a mobile device of the user, the process including: obtaining a history of time-stamped geolocations of a user; selecting a plurality of geographic areas based on each of the selected geographic areas including at least one of the time-stamped geolocations; obtaining a probabilistic model specifying parameters comprising: a plurality of candidate user events, each candidate user event being an underlying potential reason why the user moved between geographic locations; probabilities of the user transitioning between each pair of the candidate user events; and probabilities of obtaining a geolocation reported by the computing devices associated with the user in each of the plurality of geographic areas following occurrence of each of the candidate user events; inferring, with one or more processors, one of the candidate user events.