Trigger-Based Document Recommendation from Conversational Logs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional systems lack the capability to reliably identify portions of a conversation that trigger assistance, leading to either insufficient or irrelevant assistance being provided to agents, which wastes computing and human resources.
Innovation Solution
A contact center service employs domain-specific machine learning models to intelligently determine when and what assistance to provide by training ML models on conversational logs, using a trigger module to identify when assistance is needed and a retrieval module to retrieve relevant information, thereby optimizing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional systems provide assistance without intelligent identification, then assistance is provided to agents, but computing and human resources are wasted due to insufficient or irrelevant assistance
Solution Approach 1:
The system performs preliminary analysis of conversational logs during off-peak times to train ML models and identify assistance triggers in advance. This allows the system to have pre-computed assistance recommendations ready when needed, reducing real-time computing resource consumption while maintaining high reliability in assistance identification
Solution Approach 2:
The system uses domain-specific ML models that automatically learn from conversational data to identify when and what assistance is needed. The models self-optimize by continuously analyzing conversation patterns and assistance effectiveness, reducing the need for manual configuration and improving resource efficiency over time
2Loss of energy
If domain-specific ML models are used to intelligently determine assistance, then computing resources are optimized, but system complexity increases
Solution Approach 1:
The system is divided into distinct modular components: a trigger module that identifies when assistance is needed, a retrieval module that finds relevant assistance information, and domain-specific ML models that power each function. This segmentation allows each component to be optimized independently and simplifies deployment and maintenance while maintaining computing efficiency
Solution Approach 2:
The patent introduces intermediary components such as conversation log processors and assistance recommendation engines that mediate between raw conversation data and final assistance delivery. These intermediaries simplify the overall system architecture by handling complex ML model operations in isolated layers, reducing the apparent complexity for end users
3Productivity
If assistance is provided in real-time during conversations, then agent efficiency improves, but conversation duration may increase
Solution Approach 1:
The system provides assistance recommendations periodically at strategically determined points during conversations based on trigger events identified by ML models. Rather than continuously interrupting agents, the system analyzes conversation flow and provides assistance at optimal moments, maintaining agent efficiency while minimizing impact on conversation duration
Solution Approach 2:
The system provides partial assistance by offering targeted information only when specific triggers are detected, rather than providing comprehensive assistance throughout entire conversations. This selective approach improves agent efficiency at critical moments without unnecessarily extending conversation duration during routine interactions
Data Source
AI summary
Techniques for document recommendation based on conversational log for real time assistance are described. A first machine learning module identifies key phrases of a conversational log in real time. The first machine learning module executes multiple machine learning models trained to determine a probability that a portion of a conversation includes a key phrase. A second machine learning module identifies assistance pertaining to the identified key phrases of the conversational log. The second machine learning module executes a machine learning model trained to identify semantic similarity and word matching features of embedding representations of the key phrases and a knowledge base of assistance. The assistance is provided to a user during a conversation in real time.