User Intent Detection via Embedding Similarity for Tail Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing natural language understanding (NLU) systems struggle to accurately understand tail queries, which are longer and contain multi-word phrases with lower occurrence rates, leading to delayed processing times that are not viable for real-time applications like chat programs.
Innovation Solution
The proposed solution compares complex user input text to previous user inputs with known intent classifications and classifies the complex user input based on the closest-matching previous inputs, requiring fewer computing resources and enabling real-time intent detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If known natural language understanding approaches are used to process tail queries, then understanding accuracy may be maintained, but processing time increases significantly making it non-viable for real-time applications
Solution Approach 1:
The system segments the intent determination process into two distinct pathways: a fast pathway for common head queries using traditional NLU, and a slower pathway for complex tail queries using semantic similarity comparison. This segmentation allows each pathway to be optimized for its specific use case, maintaining accuracy while enabling real-time response for the majority of queries.
Solution Approach 2:
The system changes the processing parameters based on query characteristics. For tail queries, it transforms the approach from rule-based NLU processing to embedding-based semantic similarity processing. This parameter change allows the system to handle complex queries with lower occurrence rates by comparing them against a database of similar queries, achieving both accuracy and real-time performance.
2Measurement precision
If complex processing is performed on tail queries to achieve accurate intent understanding, then understanding quality improves, but computing resource consumption increases
Solution Approach 1:
The system performs preliminary action by pre-computing and storing embeddings for common query patterns and intents in a database. When a tail query arrives, the system retrieves and compares against these pre-computed embeddings rather than performing complex NLU processing from scratch. This preliminary preparation significantly reduces computing resource consumption while maintaining high intent understanding quality.
Solution Approach 2:
The system uses copying by creating embedding representations of queries and comparing these copies against a database of training phrase embeddings. Instead of performing complex symbolic NLU processing, the system works with continuous vector representations that can be efficiently compared using cosine similarity, reducing computational requirements while preserving semantic meaning.
Data Source
AI summary
A method of determining a user intent from a predefined set of user intents includes receiving user-generated text through an electronic user interface, such as a website, and generating first embeddings representative of the user-generated text. The method further includes calculating a respective individual intent score for each of a plurality of training phrases, each individual intent score calculated according to a similarity of the first embeddings to second embeddings, representative of a respective training phrase of the plurality of training phrases, wherein each training phrase is associated with an intent of a predefined set of user intents, outputting, to the user in response to the user-generated text, a plurality of user intents according to the respective individual intent scores, receiving, from the user, a selection of one of the plurality of user intents, and classifying, according to the selection, a user intent for the user-generated text.