Utterance Ranking for Conversational Bot Response Coherence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
AI-based conversational bots struggle to provide coherent responses when users make multiple statements at once, as they typically default to responding to the most recent utterance, leading to context-inappropriate responses.
Innovation Solution
A method and computing platform that applies an utterance ranking algorithm to reorder and prioritize utterances in a sequence, using a data model updated with an observation history to generate coherent responses, incorporating annotated conversations and machine learning classifiers for context understanding.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the bot responds to the most recent utterance by default, then the response generation is simple and fast, but the response may be context-inappropriate when multiple utterances are made simultaneously
Solution Approach 1:
The system performs preliminary analysis by ranking utterances before generating a response. The utterance ranking module evaluates multiple utterances in advance and determines their relative importance, allowing the bot to select the most relevant utterance for response generation rather than defaulting to the most recent one.
Solution Approach 2:
The system dynamically adjusts the processing approach based on the number of utterances detected. When multiple utterances are present, the system activates the ranking mechanism; when a single utterance is detected, the system can proceed with standard processing, making the behavior adaptive to the input conditions.
2Reliability
If the bot processes all utterances in a sequence, then context understanding is improved, but the processing complexity and time increase
Solution Approach 1:
The system changes the parameter of utterance ordering by applying a ranking function that reorders utterances based on their importance scores. This transformation converts a simple sequential processing task into a prioritized processing sequence, improving context understanding while managing complexity through systematic reordering rather than complex analysis of all possible combinations.
Solution Approach 2:
The processing system is segmented into distinct modules: utterance detection, ranking, and response generation. This modular approach allows each component to handle its specific task independently, reducing overall system complexity while maintaining comprehensive processing of multiple utterances.
3Measurement precision
If the bot uses an utterance ranking algorithm, then response accuracy is improved, but the processing time and computational resources increase
Solution Approach 1:
The system applies partial action by ranking only the necessary utterances that require disambiguation. When multiple utterances are detected, the ranking algorithm processes them to determine importance, but the response generation focuses only on the highest-priority utterance, avoiding unnecessary processing of lower-priority content while still improving accuracy.
Data Source
AI summary
A conversational bot system uses a set of conversations that have been annotated to identify speech acts, wherein a speech act is a labeled grouping of utterances. To facilitate processing, a data model associated with a multi-turn conversation is received. The data model comprises an observation history. Upon receipt of query that includes a sequence of at least two or more utterances, an utterance ranking algorithm is applied. The algorithm selectively reorders the utterances in the sequence into a ranked order of importance that reflects a lowest to highest priority of response. In response to applying the utterance ranking algorithm, the data model is then updated to reflect the ranked order. In one embodiment, updating the data model positions the highest priority utterance as a most recent utterance in the observation history. The updated data model is then used to attempt to generate a coherent response to the query.