Secure Phrase Recommendation via Vector Decoupling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text recommendation systems lack accuracy in predicting user responses due to insufficient training on historical utterance data, and they face challenges in securely storing and retrieving user conversations for learning or suggestion purposes.
Innovation Solution
The system employs machine learning techniques to train models on historical utterance data, converting each utterance into a vector to represent intent and context. These vectors are used to match user inputs, allowing for secure storage of utterances by decoupling them from their vectors, which do not represent a security risk.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If historical utterance data is stored securely by decoupling utterances from vectors, then data security is improved, but system complexity increases due to the need to manage separate storage and matching mechanisms
Solution Approach 1:
The system divides the historical utterance data into two separate components: secure utterance storage and vector representation storage. This segmentation allows the utterances to be stored in a secure, access-controlled manner while vectors can be freely used for matching and recommendation operations, thereby improving data security without requiring complex encryption or access management for the entire system.
Solution Approach 2:
The patent introduces vectors as an intermediary representation that bridges the gap between secure utterance storage and recommendation functionality. Instead of directly accessing stored utterances for matching, the system converts both historical and new utterances into vectors, performs matching in the vector space, and only retrieves actual utterances when necessary. This intermediary mechanism enables secure storage while maintaining efficient recommendation capabilities.
2Measurement precision
If machine learning models are trained on historical utterance data, then recommendation accuracy is improved, but data processing time increases during the training phase
Solution Approach 1:
The system performs preliminary conversion of historical utterances into vector representations during the data preparation phase, before actual model training begins. By pre-processing and storing these vector representations, the system avoids the need to repeatedly process raw text data during training iterations, significantly reducing data processing time while maintaining the ability to train accurate models on the full historical dataset.
Data Source
AI summary
Described herein are systems, apparatus, methods and computer program products for machine learning intent classification. In various embodiments, historical utterances provided by users may be utilized for bot training. Context and personally identifiable information may be removed from the utterances. The utterances may be associated with vectors. The utterances and vectors may be used to determine recommendations.