Template-Constrained Sentiment Models for New Feature Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional sentiment analysis models fail to capture nuanced details within documents and require retraining when new features are introduced, leading to inefficiencies and resource consumption.
Innovation Solution
A target-based sentiment model that is generative and constrained by a predetermined template, allowing it to identify features not explicitly trained for, without the need for retraining, using a text-to-text-transfer-transformation (T5) model for natural language generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional classification techniques are used to train sentiment models for specific features, then the models can optimize for those trained features, but the models require retraining when new features are introduced, leading to inefficiency and resource consumption
Solution Approach 1:
The patent segments the sentiment analysis task into multiple independent target-based models, each trained on specific target features. This segmentation allows the system to analyze different aspects of customer feedback separately while maintaining the ability to incorporate new targets without retraining the entire system, thus resolving the contradiction between adaptability and training efficiency.
Solution Approach 2:
The patent creates a universal sentiment analysis framework that can handle multiple different target types (product features, service aspects, etc.) through a common model architecture. The model is designed to be multi-functional, capable of analyzing various features without requiring separate specialized models, thereby improving both adaptability and training efficiency.
2Measurement precision
If conventional sentiment analysis models output sentiment of entire documents, then the models are simple to implement, but the results miss important details and nuances within the document
Solution Approach 1:
The patent divides the document-level sentiment analysis into fine-grained target-based sentiment analysis. Instead of analyzing the entire document as a single unit, the model segments the analysis by identifying specific targets (product features, service aspects) and their associated sentiments and reasons. This segmentation dramatically improves measurement precision by capturing nuanced details while maintaining manageable model complexity through focused target-based processing.
Data Source
AI summary
Systems and methods are directed to training and utilizing a generative language model that is constrained by a predetermined template that is used to train the generative language model. Once trained, customer data is accessed and transmitted to an evaluation component associated with the generative language model. The generative language model generates one or more sentences based on a feedback input of the plurality of feedback inputs, whereby the one or more sentences each include a sentiment, a target, and a reason for the sentiment in a format defined by the predetermined template. The evaluation component then identifies the sentiment, the target, and the reason from a sentence of the one or more sentences. A communication is then presented, on a device of a user, based on at least the sentiment and the reason identified from the sentence. The communication can be an alert or a report.