Multimodal Pseudo-Labeling for User Personality Prediction Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for creating training datasets for predicting user personality are inefficient and economically unviable, as they rely on direct labeling, which lacks accuracy and efficiency, and augmented datasets are similar to established ones, failing to guarantee learning accuracy and efficiency.
Innovation Solution
A method using pseudo-labeling to create multimodal training datasets by extracting and integrating visual, voice, and text information from labeled and unlabeled datasets, measuring similarity, and applying pseudo-labels based on threshold values to enhance dataset creation efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If direct labeling method is used to create training datasets, then dataset accuracy is improved, but time consumption and cost increase significantly
Solution Approach 1:
The patent uses pseudo-labeling to create copies of labeled data patterns and applies them to unlabeled data. The model generates predicted labels (pseudo-labels) for unlabeled samples, effectively copying the labeling pattern from a small set of manually labeled data to a large set of unlabeled data, thereby reducing time consumption while maintaining acceptable accuracy
Solution Approach 2:
The patent performs preliminary labeling on a subset of data to create initial labeled samples, then uses these to train a model that automatically labels the remaining unlabeled data. This preliminary action on a small portion enables subsequent automated processing of the entire dataset, significantly reducing overall time consumption
2Quantity of substance
If data augmentation is used to increase dataset size, then quantity of training data is improved, but learning accuracy degrades due to similarity with existing datasets
Solution Approach 1:
The patent employs self-training where the model learns from initially labeled data, generates pseudo-labels for unlabeled data, and then re-trains on the expanded dataset. This self-service mechanism allows the system to automatically expand its training data without external intervention, increasing dataset size while maintaining diversity and accuracy
Solution Approach 2:
The patent changes the parameter of data labeling status from labeled to unlabeled and back through pseudo-labeling. By iteratively changing the labeling parameter and re-training the model, the system expands the effective training dataset size while maintaining or improving learning accuracy through diverse pseudo-labeled samples
3Productivity
If pseudo-labeling is applied to all unlabeled data, then dataset creation efficiency is improved, but reliability decreases due to error propagation
Solution Approach 1:
The patent applies partial pseudo-labeling where only high-confidence predictions are used as pseudo-labels, or pseudo-labeling is applied to only a portion of the unlabeled data in each iteration. This partial action approach maintains high reliability by being selective, while still achieving improved dataset creation efficiency through automated processing
Solution Approach 2:
The patent implements iterative feedback loops where the model generates pseudo-labels, re-trains on the expanded dataset, and evaluates performance. Based on feedback from performance evaluation, the system adjusts the pseudo-labeling threshold or selectsivity, thereby maintaining reliability while achieving efficiency improvements
Data Source
AI summary
There is provided a method for creating multimodal training datasets for predicting characteristics of a user by using pseudo-labeling. According to an embodiment, the method may acquire a labelled dataset in which an image of a user is labelled with personality information and may extract a multimodal feature vector from the image of the acquired labelled dataset, may acquire an un-labelled dataset in which an image of a user is not labelled with personality information and may extract a multimodal feature vector from the image of the acquired un-labelled dataset, may measure a similarity between the extracted multimodal feature vector of the labelled dataset and the multimodal feature vector of the un-labelled dataset, and may label the un-labelled dataset based on the measured similarity. Accordingly, by creating multimodal training datasets for predicting a user personality by using pseudo-labeling, training datasets may be obtained rapidly, economically and effectively.