Deep Learning Content Ranking for Short or Typo-Prone Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional retrieval systems struggle to accurately rank content items when query inputs are limited to a few characters or contain typographical errors, leading to retrieval of irrelevant items and inefficient use of computing resources.
Innovation Solution
A deep machine learning model is employed to generate ranking scores for content items based on query and content features, predicting probabilities of user-item interactions such as playback and information retrieval, using modified DLRM to handle both categorical and continuous features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional retrieval systems rank items based on query text matching, then relevant items can be identified, but the system performs poorly when query input is limited to a few characters or contains typographical errors
Solution Approach 1:
The system pre-trains deep machine learning models using large datasets of user interactions, queries, and content items before deployment. This preliminary training enables the model to learn patterns and relationships that allow it to make accurate predictions even with limited or erroneous query input, resolving the contradiction between ranking accuracy and query flexibility.
Solution Approach 2:
The patent introduces deep machine learning models as an intermediary between the query input and the ranking process. Instead of directly matching query text to content items, the model processes the query through learned representations and predicts user interaction probabilities, enabling accurate ranking even when the query is incomplete or contains errors.
2Productivity
If the ranking algorithm relies heavily on query text, then it can identify relevant items, but it retrieves irrelevant items when query input is limited or contains errors
Solution Approach 1:
The system transforms the ranking approach from direct text matching to using deep learning models that process multiple features including user profiles, content attributes, and interaction histories. This parameter transformation allows the system to maintain high retrieval efficiency while improving reliability by considering broader contextual factors beyond just query text.
Solution Approach 2:
The patent replaces traditional mechanical text-matching algorithms with deep machine learning models that use neural networks to process and understand query intent. This substitution enables the system to handle limited or erroneous queries more effectively by learning from patterns in large datasets rather than relying on exact text matches.
3Reliability
If users submit multiple queries to locate desired items, then they can find relevant content, but computing resources are consumed inefficiently
Solution Approach 1:
The system performs preliminary training of deep learning models offline using extensive datasets, so that when users submit queries, the models are already equipped with learned patterns and knowledge. This preliminary action enables the system to accurately rank items in a single query pass, reducing the need for multiple queries and thereby conserving computing resources.
Data Source
AI summary
Disclosed herein are system, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for ranking a plurality of content items for presentation to a user. An embodiment generates a ranking score for each content item by: providing input to a deep machine learning (ML) model, the input including at least one or more query features and one or more content item features, determining, by the deep ML model and based at least on the input, a first probability of a first type of interaction between the user and the content item and a second probability of a second type of interaction between the user and the content item, and calculating the ranking score for the content item based at least on the first and second probabilities. An embodiment ranks the content items for presentation based on the ranking score associated with each content item.