Sentiment Classification Model Training Using Emoticon Labels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current social text sentiment classification models rely heavily on manually labeled samples, which are costly and limited in quantity, leading to poor performance and an inability to meet practical application requirements.
Innovation Solution
A classification model training method that combines a large quantity of weakly supervised samples, where emoticons in social texts are used as sentiment labels, with a small quantity of supervised samples, allowing for the training of a social text sentiment classification model without increasing manual labeling costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manually labeled samples are used for training, then model training can be performed, but the cost is high and the quantity is limited
Solution Approach 1:
The patent applies preliminary action by first training an initial classification model using a large quantity of weakly supervised samples (social texts with emoticons) before fine-tuning it with manually labeled samples. This preliminary training prepares the model to effectively utilize the limited manual labels, resolving the contradiction between limited sample quantity and model performance requirements
Solution Approach 2:
The patent introduces weakly supervised samples (social texts with emoticons) as an intermediary between abundant unlabeled data and limited manually labeled data. This intermediary training data allows the model to learn from large quantities of data without requiring expensive manual labeling, thereby increasing effective sample quantity while maintaining training quality
2Reliability
If manually labeled samples are increased to improve model performance, then model accuracy improves, but labeling costs increase
Solution Approach 1:
The model is preliminarily trained on weakly supervised data before fine-tuning with manual labels, which maximizes the utilization of each manual label and reduces the total number of manual labels needed to achieve target accuracy, thereby lowering labeling costs while maintaining model accuracy
Solution Approach 2:
The patent changes the training parameter from requiring only manually labeled data to using a combination of weakly supervised data (emoticon-labeled) and manually labeled data. This parameter change in training data composition allows achieving the same model accuracy with fewer manual labels, reducing labeling costs
3Measurement precision
If only manually labeled samples are used, then label quality is high, but the quantity is insufficient for training high-performance models
Solution Approach 1:
The patent segments the training data into two distinct segments: weakly supervised samples (social texts with emoticons) for initial model training and manually labeled samples for fine-tuning. This segmentation allows the model to first learn from abundant weakly labeled data and then refine its performance using high-quality manual labels, achieving both quantity and quality requirements
Solution Approach 2:
Weakly supervised samples serve as an intermediary training data that bridges the gap between abundant unlabeled social texts and limited manually labeled samples. This intermediary data enables the model to learn effective features from large quantities of data while maintaining label quality through the two-stage training approach
Data Source
AI summary
Embodiments of this application disclose a classification model training method, a classification method, a device, and a medium. An initial classification model is first trained by using a first sample set including a large quantity of first samples, to obtain a pre-trained model, each first sample including a social text and an emoticon label corresponding to the social text; and the pre-trained model is then trained by using a second sample set including a small quantity of second samples, to obtain a social text sentiment classification model that uses a social text as an input and use a sentiment class probability distribution corresponding to the social text as an output. In this method, the model is trained by combining a large quantity of weakly supervised samples with a small quantity of supervised samples, to ensure that the model obtained through training has better model performance without increasing manually labeled samples.