Multiclass Classification Model for Search Result Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional machine-learning models used in online concierge systems are not trained to evaluate different types of relevance to search queries and users, limiting the quality of ads and recommendations provided to users.
Innovation Solution
A trained multiclass classification computer model is used to classify search query results into multiple relevance classes, allowing for more precise organization and recommendation of items to users based on their search queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine-learning models are used to score content relevance, then the system can provide basic search results, but the quality of ads and recommendations is limited because the models are not trained to evaluate different types of relevance
Solution Approach 1:
The patent segments the single relevance evaluation task into multiple distinct relevance types (e.g., exact match, substitute, complement, irrelevant). The multiclass classification model is trained to evaluate each item against multiple relevance classes simultaneously, allowing the system to provide differentiated recommendations based on specific relevance types rather than a single generic score.
2Ease of operation
If traditional machine-learning models are used, then the system structure remains simple, but the ability to provide useful content recommendations is limited
Solution Approach 1:
The patent changes the parameters of the machine-learning model by transitioning from traditional single-output relevance scoring to a multiclass classification framework with multiple output classes representing different relevance types. This parameter change enables the model to provide higher quality recommendations by distinguishing between different types of item-query relationships, despite the increased model complexity.
Data Source
AI summary
Classifying results of a user's search query using a trained classification model. In response to the search query, an online system retrieves a set of candidate search results, each candidate search result associated with a respective item of a plurality of items. The online system accesses the classification model that is trained to compute a probability of classification of each item into each class of a plurality of classes, each class associated with a type of relevance to the search query. The online system applies the classification model to generate, for each item, a classification score associated with each class. The online system classifies, based on the classification score, each item into a corresponding type of relevance to the search query. The online system selects, based on the classification of each item, a list of items for displaying at a user interface of a device associated with the user.