Sticker Recommendation via Dialogue Act Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for recommending stickers in social network services and instant messengers often fail to accurately consider the context and emotional state of a dialogue, leading to inappropriate emoticon suggestions, especially when keywords do not match the intended emotional state or context.
Innovation Solution
A method that uses dialogue act analysis to understand the context of a conversation by classifying dialogue acts and generating dialogue situation information, allowing for the recommendation of appropriate stickers based on the analysis of multiple utterances, rather than just relying on keyword matching.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If keyword matching is used for emoticon recommendation, then the system is simple and fast, but the recommendation accuracy deteriorates when keywords do not match the intended emotional state
Solution Approach 1:
The patent segments the dialogue analysis into multiple components: surface analysis (morphological and syntactic), semantic analysis (meaning extraction), and pragmatic analysis (dialogue act classification). This segmentation allows the system to process dialogue at different levels of abstraction, improving recommendation accuracy without creating a monolithic complex system.
Solution Approach 2:
The patent introduces dialogue act classification as an intermediary layer between keyword matching and emoticon recommendation. The dialogue act classifier analyzes the function and intent of utterances, serving as a mediator that bridges simple keyword detection and complex emotional state understanding, thereby improving accuracy while maintaining manageable system complexity.
2Loss of information
If only single utterance analysis is performed, then the processing speed is fast, but the context understanding deteriorates
Solution Approach 1:
The patent performs preliminary surface analysis and semantic analysis on individual utterances before conducting dialogue act classification. This preliminary processing prepares the data in advance, allowing the system to efficiently handle multiple utterances in sequence without excessive processing time while maintaining context understanding.
Solution Approach 2:
The patent transitions from analyzing single utterances in isolation to analyzing sequences of utterances by adding the temporal dimension of dialogue flow. By classifying dialogue acts across multiple utterances, the system captures contextual information while maintaining efficient processing through structured analysis frameworks.
3Measurement precision
If dialogue act analysis of multiple utterances is performed, then the context understanding is improved, but the computational complexity increases
Solution Approach 1:
The patent segments the dialogue act analysis into manageable components: surface analysis, semantic analysis, and pragmatic analysis. Each segment handles specific aspects of the dialogue, reducing the complexity of any single analysis module while collectively achieving high context understanding accuracy through their integration.
4Reliability
If simple keyword matching is used, then the implementation is easy, but the emotional state recognition deteriorates when keywords are used ironically
Solution Approach 1:
The patent introduces dialogue act classification as an intermediary that analyzes the functional intent and contextual meaning of utterances. This intermediary layer detects ironic usage by examining whether the dialogue act matches the literal keyword meaning, thereby improving emotional state recognition reliability while maintaining reasonable implementation ease through modular design.
Data Source
AI summary
Provided is a method of recommending a sticker through a dialog act analysis. The method includes: by a server, performing a surface analysis on the last utterance between the first user terminal and the second user the terminal; performing a dialog act analysis on the last utterance using a result of the surface analysis; extracting a dialog context factor including a surface analysis result and a dialog act analysis result on a certain number of continuous utterances including the last utterance between the first user terminal and the second user terminal; selecting a sticker to be recommended to the first user using the dialog context factor; and providing the selected sticker for the first user terminal.