Harassment Detection Using Positive Utterance Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies for detecting harassment in workplaces often fail to provide clear and timely feedback to individuals making utterances that may be considered harassing, as they only focus on negative levels without considering positive utterances that can reduce harassment, leading to incomplete understanding and ineffective suppression.
Innovation Solution
A system that converts speech data into text, calculates both negative and positive levels in conversations, and outputs harassment information to the speaker, including the level of harassment, to raise awareness and facilitate suppression.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If only negative levels are considered for harassment detection, then the detection process is simple, but the reliability of harassment level determination is insufficient
Solution Approach 1:
The analysis process is segmented into two independent components: negative level calculation (detecting harassing utterances) and positive level calculation (detecting non-harassing utterances). This segmentation allows the system to consider both positive and negative aspects separately, improving the reliability of harassment level determination while maintaining a manageable analysis process through modular computation.
2Loss of information
If harassment detection is performed without considering positive utterances, then the processing speed is fast, but the understanding of conversation context is incomplete
Solution Approach 1:
The system performs preliminary classification of utterances into positive and negative categories during the conversation flow. By pre-identifying and categorizing utterances as they occur, the system maintains complete context understanding without excessive processing delays, as the classification is integrated into the natural conversation rhythm rather than requiring post-conversation analysis.
Solution Approach 2:
The system provides continuous feedback to speakers about their harassment levels based on real-time analysis of both positive and negative utterances. This feedback mechanism allows speakers to understand their impact on others during the conversation, enabling them to adjust their behavior promptly. The feedback loop includes displaying harassment levels and providing notifications when thresholds are exceeded, creating a dynamic system that adapts to conversation flow.
3Measurement precision
If only prohibited words are used for harassment detection, then the detection method is simple, but the detection precision is limited
Solution Approach 1:
The system transitions from a single-parameter detection method (prohibited word matching) to a multi-parameter evaluation system that considers both negative levels (harassing content) and positive levels (non-harassing content). This parameter change enables more precise harassment detection by evaluating the overall conversation context rather than relying solely on keyword presence, while the automated calculation methods keep the implementation complexity manageable.
Data Source
AI summary
A speech-to-text unit of a harassment information providing apparatus converts an utterance of a speaker into text using speech data of a conversation. A determination unit calculates a negative level and a positive level of the conversation by analyzing text data indicating the utterance converted into text, and determines the level of harassment of the conversation by using the negative level and the positive level. An output unit outputs harassment information including information indicating the level of harassment.