Vector Representation for Temporal Usage Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional technologies face challenges in uniquely labeling and modeling various types of POIs, such as clothing, daily necessities, and foods, and in representing user location patterns with limited labels, leading to information loss and inefficiencies in data utilization.
Innovation Solution
An information processing apparatus specifies vector representations for temporal usage patterns of areas and users using a three-layer neural network-based model, such as Skip-gram, to acquire and process location information history data, enabling unique and detailed modeling of usage patterns without the need for pre-collected POI information or manual labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional label transition models are used to represent staying locations by labels (e.g., residential districts, restaurants), then the data can be abstracted and meaning can be easily settled, but stores covering various categories cannot be uniquely labeled and POIs not registered in the data set cannot be labeled
Solution Approach 1:
The patent transforms the labeling system from discrete categorical labels to continuous vector representations. Each location is represented by a vector that captures its semantic meaning in a multi-dimensional space, allowing for nuanced differentiation of diverse POIs including stores covering various categories and unregistered locations, while maintaining the abstraction benefits of label-based models
2Ease of operation
If labels represent staying purposes (home, workplace, eating and drinking, entertainment), then the model is simple to operate, but a large part of information is lost because the number of labels is limited to several tens to several hundreds
Solution Approach 1:
The patent transitions from a low-dimensional label space (several tens to hundreds of categories) to a high-dimensional vector space where each dimension can capture a different aspect of location characteristics. This dimensional expansion allows the model to preserve fine-grained information about staying purposes and location features while maintaining operational simplicity through standardized vector processing
3Measurement precision
If vector representations are used to represent features of temporal usage patterns, then unique and detailed modeling of usage patterns is achieved, but the device complexity increases due to the need for neural network-based models
Solution Approach 1:
The patent uses pre-trained word embedding models (such as Word2Vec or GloVe) that have already learned semantic relationships from large corpora. These pre-trained models serve as reusable components that can be directly applied to location data without requiring training from scratch, thereby reducing the computational complexity and resource requirements while maintaining high precision in capturing temporal usage patterns
Data Source
AI summary
An information processing apparatus specifies a vector representation which represents features of respective temporal usage patterns of two or more areas. The information processing apparatus includes a processor. The processor acquires area-specific usage pattern data which indicates which of two or more patterns the temporal usage pattern of each area is. The processor specifies, on the basis of the area-specific usage pattern data, a vector representation which represents the feature of the temporal usage pattern of each area.