Machine Learning Model for User Intent Inference in Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing information retrieval methods struggle to accurately interpret user intent from queries, leading to irrelevant search results, especially when queries are broad, imprecise, or use informal language.
Innovation Solution
A machine learning model is trained to infer user intent by processing user queries through a query model, feature model, and intent model, generating embeddings that suggest relevant documents, and presenting a limited set of relevant information to the user.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional information retrieval methods are used to process user queries, then the system is simple and fast to implement, but the search results are irrelevant when queries are broad or use informal language
Solution Approach 1:
The patent segments the query processing task into multiple specialized model portions: a query model portion for understanding the query, a document model portion for representing documents, and an intent model portion for inferring user intent. Each portion processes specific aspects of the retrieval task independently, then their outputs are combined to achieve accurate intent interpretation while maintaining manageable system complexity through modular design
Solution Approach 2:
The patent transforms the query and document representations into embedding vectors in a high-dimensional semantic space. By projecting queries, documents, and intent into this shared vector space, the system can measure semantic similarity and infer intent through geometric relationships in the embedding dimension, enabling accurate interpretation of broad or informal queries without requiring explicit rule-based logic
2Productivity
If traditional search methods are used, then the system requires minimal processing, but users must navigate through numerous irrelevant results and perform multiple queries
Solution Approach 1:
The patent performs preliminary intent inference by processing the user query through the trained model portions before retrieving documents. The query model and intent model generate embeddings that capture the user's underlying intent, allowing the system to pre-filter and rank documents that truly match the inferred intent, thereby reducing the number of irrelevant results users must navigate and minimizing the need for multiple follow-up queries
Solution Approach 2:
The patent uses interaction data from past query-document pairs to train and refine the model portions. By incorporating feedback from actual user interactions and retrieval outcomes, the system continuously improves its intent inference accuracy, leading to better-ranked results that reduce user time spent navigating irrelevant content and performing repeated queries
3Loss of information
If the system processes and presents all search results, then complete information is provided, but the user interface becomes cluttered and overwhelming
Solution Approach 1:
The patent extracts and presents only the most relevant information by using the intent model embeddings to identify and highlight key documents that best match the inferred user intent. Rather than displaying all search results, the system extracts the subset of documents with highest relevance scores and presents them prominently, maintaining information completeness for the user's actual needs while keeping the interface clean and usable
Solution Approach 2:
The patent applies different quality levels of information presentation to different parts of the results. High-priority documents that strongly match the inferred intent are presented with full detail and prominence, while lower-priority documents are summarized or hidden until needed. This local differentiation of information quality ensures complete information is available when needed while maintaining ease of operation through selective presentation
Data Source
AI summary
A method for providing document category recommendations may include training a machine learning algorithm; receiving, from a user, a selection of an anchor document; retrieving a user co-viewing sequence; generating, via the trained machine learning algorithm, a sequence embeddings set based on the anchor document and the user co-viewing sequence; comparing the generated embeddings set to respective embeddings for a plurality of candidate sets; determining the candidate set closest to the generated embeddings; and presenting at least one category from the closest candidate set.