Sentiment Analysis Using Adaptive Segmentation and Lexicon Mixing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current sentiment analysis technologies face challenges in accurately analyzing sentiments in contact center interactions due to noise in input data, informal and conversational speech, and the need for domain-specific models, which often require large amounts of manually labeled data and struggle with sentiment shifts and segmentation.
Innovation Solution
A method involving machine learning classifiers trained on a corpus of labeled contact center interactions, using techniques like neutral factoring, handling negations and intensifiers, and adaptive segmentation to improve sentiment analysis accuracy, which leverages both generic and domain-specific lexicons to generate models without extensive manual labeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If domain-specific sentiment analysis models are developed to improve accuracy in contact center interactions, then sentiment analysis accuracy is improved, but the amount of manually labeled data required increases
Solution Approach 1:
The system performs preliminary sentiment analysis using a generic lexicon before domain-specific adaptation. This preliminary action allows the system to establish baseline sentiment scores and then refine them with domain-specific adjustments, reducing the need for extensive manual labeling of domain-specific data while maintaining accuracy.
Solution Approach 2:
The system changes parameters by adjusting sentiment scores based on domain-specific context. Instead of requiring complete re-labeling of domain data, the system modifies existing sentiment parameters (scores and orientations) to account for domain-specific nuances, thereby improving accuracy without proportionally increasing manual labeling requirements.
2Measurement precision
If traditional sentiment analysis methods are used to process contact center interactions, then processing speed is maintained, but sentiment analysis accuracy deteriorates due to noise and informal speech
Solution Approach 1:
The system applies local quality by treating different portions of the text with different processing approaches. Specific tokens are identified as having positive or negative sentiment orientations, and these local sentiment indicators are aggregated to determine overall sentiment. This allows accurate sentiment detection in noisy, informal speech while maintaining efficient processing through targeted analysis rather than comprehensive processing.
3Measurement precision
If sentiment analysis is applied to entire text corpora without segmentation, then processing simplicity is maintained, but measurement precision deteriorates due to sentiment shifts within the text
Solution Approach 1:
The system segments the text corpus into individual tokens and evaluates sentiment at the token level rather than analyzing the entire corpus as a single unit. This segmentation allows the system to capture sentiment shifts that occur at different points in the text, improving measurement precision while managing complexity through systematic token-level processing.
Data Source
AI summary
Methods, systems, and computer program product for automatically performing sentiment analysis on texts, such as telephone call transcripts and electronic written communications. Disclosed techniques include, inter alia, lexicon training, handling of negations and shifters, pruning of lexicons, confidence calculation for token orientation, supervised customization, lexicon mixing, and adaptive segmentation.