Interpreting Text-Based Similarity Model for Recommendation Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current recommendation systems, particularly those using machine learning models, face challenges in interpreting and explaining their predictions due to the complex nature of their algorithms, leading to unreliable recommendations, especially when usage data is unavailable and in domains like audio, video, and text content.
Innovation Solution
The development of an Interpreting Text-Based Similarity (ITBS) model that calculates gradients and feature vectors to identify word-pairs with similar semantic meanings, providing insights into how pre-trained language models generate recommendations, thereby improving model interpretability and user trust.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pre-trained language models are used to generate recommendations, then recommendation accuracy is improved, but model interpretability deteriorates
Solution Approach 1:
The patent introduces an Interpreting Text-Based Similarity (ITBS) model as an intermediary between the pre-trained language model and the user. This ITBS model calculates gradients and generates explanations that bridge the gap between the complex internal workings of the language model and human-understandable reasoning, thereby maintaining recommendation accuracy while improving interpretability
Solution Approach 2:
The patent replaces the opaque mechanical system of deep neural network predictions with an interpretable gradient-based explanation system. By using gradient calculations to generate text explanations, the system substitutes the uninterpretable internal state transitions with human-readable semantic relationships, resolving the contradiction between accuracy and interpretability
2Measurement precision
If complex machine learning algorithms are used, then prediction accuracy is improved, but explanation reliability deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the ITBS model continuously generates explanations based on gradient calculations and uses these explanations to refine its understanding of the prediction process. This feedback loop ensures that explanations remain reliable and accurate by constantly aligning them with the actual prediction outcomes of the pre-trained language model
Solution Approach 2:
The patent performs preliminary gradient calculations and explanation generation before final predictions are made. By pre-computing the explanatory components and validating them against known relationships, the system ensures that explanations are reliable and accurate prior to deployment, rather than attempting to explain complex predictions after they are made
Data Source
AI summary
The disclosure herein describes a system for interpreting text-based similarity between a seed item and a recommended item selected by a pre-trained language model from a plurality of candidate items based on semantic similarities between the seed item and the recommended item. The system analyzes similarity scores and contextual paragraph representations representing text-based descriptions of the seed item and recommended item to generate gradient maps and word scores representing the text-based descriptions. A model for interpreting text-based similarity utilizes the calculated gradients and word scores to match words from the seed item description with words in the recommended item description having similar semantic meaning. The word-pairs having the highest weight are identified by the system as the word-pairs having the greatest influence over the selection of the recommended item from the candidate items by the original pre-trained language model.