Machine Learning Hyperlink Relevance Scoring for Message Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to accurately predict the relevance of hyperlinks to search queries, particularly in messaging platforms, as messages may contain irrelevant hyperlinks that do not align with user search queries.
Innovation Solution
A relevance prediction system utilizing a machine learning model to analyze hyperlinks within messages, considering factors like context, sentiment, and hyperlink types, to determine an overall relevance score for resources based on user queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If the search system returns all resources from the database, then the quantity of resources is maximized, but the relevance accuracy deteriorates
Solution Approach 1:
The system changes the parameter of resource selection from binary inclusion/exclusion to a spectrum based on predicted relevance scores. By transforming the selection criterion from simple quantity maximization to relevance-score-based filtering, the system achieves both adequate resource quantity and high relevance accuracy simultaneously.
Solution Approach 2:
The patent replaces traditional mechanical search mechanisms (keyword matching, simple filtering) with a machine learning-based relevance prediction system. This substitution enables the system to accurately assess relevance without sacrificing resource quantity, as the ML model can rank and select resources based on predicted relevance rather than simple presence/absence criteria.
2Device complexity
If the search system uses simple keyword matching, then the device complexity is minimized, but the relevance prediction accuracy deteriorates
Solution Approach 1:
The patent introduces a machine learning model as an intermediary component between the search query and resource selection. This intermediary layer processes complex patterns in messages, hyperlinks, and contexts to predict relevance, achieving high accuracy without requiring the entire search system to be overly complex. The ML model acts as a specialized module that handles the complexity of relevance prediction.
3Measurement precision
If the search system filters resources based on strict relevance criteria, then the relevance accuracy is maximized, but the quantity of returned resources deteriorates
Solution Approach 1:
The system employs dynamic relevance thresholding where the threshold for resource inclusion is adjusted based on the distribution of predicted relevance scores. Rather than using a fixed strict threshold that would eliminate too many resources, the system dynamically determines an optimal threshold that maintains both high relevance accuracy and adequate resource quantity, adapting to different search contexts and data distributions.
Data Source
AI summary
Systems and methods for predicting relevance of resources to search queries. In some aspects, the system may receive a search query requesting resources from a database and may identify resources, including messages, relating to the search query. The system may extract, from the messages, hyperlinks specifying locations within the database. The system may input, into a model, the search query and the hyperlinks to cause the model to generate predictions of relevance of the hyperlinks to the search query. The system may then determine an overall relevance score for the resources in relation to the search query based on the predictions of relevance.