Discrete Choice Model Estimation Accuracy via Record Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing discrete choice models struggle to accurately estimate which option is chosen by a choosing subject from a plurality of options, especially when different subjects recognize options differently, leading to incorrect determination of option effects.
Innovation Solution
An information processing method that distinguishes records based on whether a choosing subject has recognized a specific option, setting different expected values for probability of belonging to a class, and learning a discrete choice model using these values and a mathematical formula to accurately estimate option choices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a discrete choice model is learned using all records without distinguishing subject recognition, then the model structure remains simple, but the measurement precision of option choice estimation deteriorates
Solution Approach 1:
The patent segments records into two types: records where the choosing subject recognized the first option and records where the choosing subject did not recognize the first option. This segmentation allows the model to handle different subject recognition states separately, improving measurement precision without requiring complete redesign of the model structure.
Solution Approach 2:
The patent applies different expected values to different segments of records. For records where the subject recognized the first option, a first expected value is used; for records where the subject did not recognize the first option, a second expected value is used. This local differentiation improves estimation accuracy for each subgroup while maintaining overall model simplicity.
2Reliability
If uniform expected values are used for all records, then the learning process is simple, but the reliability of class probability determination deteriorates
Solution Approach 1:
The patent divides records into segments based on subject recognition of the first option. This segmentation enables the assignment of different expected values to different segments, improving the reliability of class probability determination while keeping the learning process manageable through structured differentiation.
Solution Approach 2:
The patent changes the expected value parameter based on whether the choosing subject recognized the first option. By adjusting this single parameter across different record segments, the model achieves more reliable class probability determination without fundamentally changing the learning methodology.
Data Source
AI summary
An information processing method executes a process including: acquiring records that each represents an option; for each record of one or more records corresponding to a choosing subject that has recognized a first option, setting a first value that represents that a choosing subject belongs to a first class in which the first option has been recognized, as an expected value of a probability that a choosing subject belongs to the first class; for each record of remaining records, setting, as the expected value, a second value that represents that a choosing subject belongs to the first class based on a reference value of a probability that a choosing subject belongs to the first class and a mathematical formula that represents an effect of each option; and learning the reference value and a parameter in the mathematical formula based on each record of the acquired records and the expected value.