Conversation Topic Mining Using Cross-Conversation Pruning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to effectively mine topics from nuanced conversations between customers and agents in real-time, which are challenging due to their multi-party nature and ill-defined turns, hindering insights into customer issues and agent performance.
Innovation Solution
A computer-implemented method using a topic mining algorithm to identify candidate topics from conversations, involving utterance tagging, keyword extraction, and graph-based ranking to determine importance, followed by pruning and grouping topics based on semantic similarity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated topic mining is implemented from conversation data, then productivity and efficiency are improved, but the complexity of processing multi-party conversations with ill-defined turns increases device complexity
Solution Approach 1:
The conversation processing is segmented into distinct stages: utterance identification, part-of-speech tagging, keyword extraction, keyphrase identification, and topic selection. Each stage handles a specific aspect of the complex processing task, breaking down the overall complexity into manageable components that can be processed sequentially
Solution Approach 2:
The system introduces intermediate processing layers including part-of-speech tags as intermediaries between raw text and keywords, and keywords as intermediaries between text and final topics. These intermediaries simplify the transformation process by creating structured representations at each stage, making the overall complex task more tractable
2Measurement precision
If manual analysis of conversations is performed to identify topics, then measurement precision is improved, but loss of time increases significantly
Solution Approach 1:
The system enables self-service automated topic mining where the conversation data is processed through the algorithmic pipeline without requiring manual intervention. The system automatically performs utterance segmentation, POS tagging, keyword extraction, and topic identification, eliminating the need for human analysts while maintaining consistent and repeatable results
Solution Approach 2:
The manual mechanical process of human analysis is replaced with an automated computational system using natural language processing algorithms. The system substitutes human cognitive processes with machine-based text analysis, including automated POS tagging using predefined tag sets and algorithmic keyword extraction based on frequency and positional criteria
3Quantity of substance
If comprehensive keyword extraction is performed from all utterances, then quantity of information is improved, but loss of time increases due to processing volume
Solution Approach 1:
The system performs partial keyword extraction by focusing on specific parts of speech (nouns, verbs, adjectives) rather than processing all words equally. It also applies positional filtering to capture keywords from significant locations in utterances, extracting sufficient information without the need to analyze every single word in detail
Solution Approach 2:
The system changes parameters by filtering keywords based on part-of-speech categories and positional criteria. It transforms the raw text into structured data with assigned POS tags, then applies parameter-based filtering to select relevant keywords, thereby reducing the volume of data that requires further processing while maintaining topic identification quality
Data Source
AI summary
A method for mining topics discussed in conversations that includes: receiving conversation data; and using a topic mining algorithm to mine topics from the conversation data. The topic mining algorithm includes identifying candidate topics in each of the conversations. The topic mining algorithm further includes identifying the topics of the conversations by: compiling a list of the candidate topics; pruning the list of candidate topics by discarding certain of the candidate topics per a cross-conversation factor that factors usage across all conversations; and identifying the candidate topics remaining on the pruned list of candidate topics as the topics. The topic mining algorithm further includes determining topic groups by grouping the topics according to a degree of semantic similarity between the topics; and associating a list of utterances with the topic groups.