Vector Propagation Click Graph Model for Query Intent Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intelligent customer service systems face challenges in accurately understanding user intents due to their reliance on traditional natural language processing techniques, leading to poor categorization results, especially when users employ colloquial language or ask short questions, which increases with the complexity of queries in e-commerce applications.
Innovation Solution
The implementation of Vector Propagation On a Click Graph (VPCG) models, trained on user click data, to determine similarity scores between user queries and candidate topics, enabling more accurate prediction of user needs and intents by leveraging user click behavior to improve matching and ranking of relevant topics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional natural language processing techniques are used for intent understanding, then the system structure remains simple, but the categorization accuracy deteriorates when users employ colloquial language or ask short questions
Solution Approach 1:
The patent transforms the intent understanding task from traditional text-based NLP to a vector-based similarity matching task. User queries and candidate topics are converted into vector representations, and the system computes similarity scores between them. This parameter transformation from discrete text to continuous vectors enables more nuanced matching of colloquial and short queries, improving categorization accuracy while maintaining manageable system complexity through efficient vector operations
Solution Approach 2:
The patent replaces traditional mechanical NLP processing (parsing, keyword matching, rule-based classification) with a data-driven vector similarity approach. Instead of relying on hand-crafted linguistic rules and complex parsing mechanisms, the system uses vector representations and similarity computation to capture semantic relationships, thereby improving accuracy for colloquial language while avoiding the complexity of rule-based systems
2Measurement precision
If more sophisticated NLP techniques are implemented to handle query complexity, then categorization accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent pre-computes and stores vector representations of candidate topics in advance. When a user query arrives, the system only needs to convert the query to a vector and compute similarity scores against the pre-computed topic vectors, rather than performing complex NLP analysis in real-time. This preliminary preparation significantly reduces processing time while maintaining high intent understanding accuracy
Solution Approach 2:
The patent uses vector representations as simplified copies or approximations of the full semantic meaning of text. Instead of processing the complete linguistic structure of queries and topics, the system works with compact vector embeddings that capture essential semantic information, enabling fast similarity computation without sacrificing understanding accuracy
Data Source
AI summary
A query is received from a user at a data service engine. The query includes a string of characters. A number of candidate topics are identified by the data service engine based on the query. A similarity score is determined between the query and each of the plurality of candidate topics based on a Vector Propagation On a Click Graph (VPCG) model trained based on user click data. A number of candidate topics are ranked based on the similarity scores. One or more topics are selected from the ranked candidate topics. The selected topics are outputted via a user interface (UI).