Sentiment-Aware Sentence Generation via Convex Combination Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems, including chatbots and human support agents, face challenges in automatically rephrasing sentences to change sentiment while retaining meaning, which is crucial for providing effective customer support and improving user experience.
Innovation Solution
A method using a machine learning model to generate candidate sentences word-by-word, combining a language model to predict next words and a sentiment model to evaluate sentiment scores, allowing for the selection of words that maximize probability and sentiment alignment, resulting in a replacement sentence with a specified sentiment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a language model is used to generate candidate sentences based on probability of word appearance, then the meaning retention is improved, but the sentiment control is insufficient
Solution Approach 1:
The patent combines a language model and a sentiment model into an integrated sentence generation system. The language model generates candidate sentences based on probability of word appearance to retain meaning, while the sentiment model evaluates and scores these candidates to ensure proper sentiment alignment. This merging of two specialized models resolves the contradiction by simultaneously achieving meaning retention and sentiment control.
Solution Approach 2:
The system implements feedback through the sentiment model that evaluates candidate sentences generated by the language model. The sentiment scores are fed back to select the most appropriate candidate that matches the desired sentiment while retaining the original meaning. This feedback mechanism enables the system to iteratively improve sentiment control without sacrificing meaning retention.
2Measurement precision
If multiple models are integrated to evaluate both probability and sentiment, then the sentiment accuracy is improved, but the system complexity increases
Solution Approach 1:
The patent segments the sentence generation task into two distinct evaluation dimensions: language probability (meaning retention) handled by the language model, and sentiment accuracy handled by the sentiment model. This segmentation allows each model to specialize in its strength while working together through a unified framework, achieving high sentiment accuracy without excessive overall system complexity.
3Manufacturing precision
If candidate sentences are evaluated using both probability and sentiment scores, then the quality of replacement sentences is improved, but the computational time increases
Solution Approach 1:
The system generates multiple candidate sentences beyond what would be strictly necessary (excessive action), then uses the sentiment model to evaluate and select the best ones. This approach ensures high quality replacement sentences by having more candidates to choose from, while the efficient sentiment scoring mechanism keeps computational time manageable by not requiring exhaustive evaluation of all possible sentences.
Data Source
AI summary
Certain aspects of the present disclosure provide techniques for generating a replacement sentence with the same or similar meaning but a different sentiment than an input sentence. The method generally includes receiving a request for a replacement sentence and iteratively determining a next word of the replacement sentence word-by-word based on an input sentence. Iteratively determining the next word generally includes evaluating a set of words of the input sentence using a language model configured to output candidate sentences and evaluating the candidate sentences using a sentiment model configured to output sentiment scores for the candidates sentences. Iteratively determining the next word further includes calculating convex combinations for the candidate sentences and selecting an ending word of one of the candidate sentences as the next word of the replacement sentence. The method further includes transmitting the replacement sentence in response to the request for the replacement sentence.