Techniques for providing relevant search results for search queries
Inventors
So, Kelvin • AGARWAL, Alok K. • Zhu, Xiangxin • Yang, Yinfei • Hornkvist, John M. • Ouyang, Hua • Tirthapura, Srikanta N. • PI, Jiong Ming • Wu, Wentao • TAYADE, Aditya K. • SHAH, Pulah J. • PAREEK, Ayush • PUDER, Bradley • Xu, Liqin • NAMINI ASL, Nader • UNNAM, Sai Pavan Kumar • WANG, Shuona • ZHONG, Xiya • Wang, Xuan • PAN, Yinying
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12737425-B2
Publication Date
2026-09-15
Expiration Date
Abstract
One embodiment sets forth a method for providing relevant search results for search queries. According to some embodiments, the method can be implemented by a client computing device, and includes the steps of (1) receiving a query, wherein the query is associated with a user account, and the user account is associated with a user account vector, (2) generating a query vector based at least in part on the query, (3) generating an output vector based at least in part on the query vector and the user account vector, (4) obtaining, based at least in part on the query, a plurality of digital asset vectors, wherein each digital asset vector of the plurality of digital asset vectors corresponds to a respective digital asset, (5) comparing the output vector to the plurality of digital asset vectors to generate respective similarity scores for the plurality of digital asset vectors, (6) filtering the plurality of digital asset vectors in accordance with the similarity scores to establish a filtered plurality of digital asset vectors, and (7) displaying, in accordance with the filtered plurality of digital asset vectors, respective affordances for the respective digital assets that correspond to the filtered plurality of digital asset vectors.
Core Innovation
A client computing device provides relevant search results by receiving a query associated with a user account and a user account vector. The device generates a query vector based at least in part on the query, and generates an output vector based at least in part on the query vector and the user account vector. The device also obtains, based at least in part on the query, a plurality of digital asset vectors, where each digital asset vector corresponds to a respective digital asset.
The device compares the output vector to the plurality of digital asset vectors to generate respective similarity scores. The device filters the plurality of digital asset vectors in accordance with the similarity scores to establish a filtered plurality of digital asset vectors. The device then displays, in accordance with the filtered plurality of digital asset vectors, respective affordances for the digital assets corresponding to the filtered plurality of digital asset vectors.
In refinements described with respect to the same overall approach, the digital asset vectors are formed from digital asset metadata vector and digital asset content vector using transformer-based LLMs and machine learning models, respectively, and the resulting vectors are combined into a digital asset output vector. A user account vector is constructed from favorites, frequently accessed digital asset vectors, and query history vectors. The query vector is generated from query text and query statistics using a transformer-based LLM.
Further refinements include retrieving digital asset vectors using keyword-based searching via retrieval trees and keyword indexes combined with semantic-based searching using embeddings and vector indexes, and calibrating relevance scores across retrieval sources before combining or filtering into a plurality of digital asset vectors. For queries that include at least one question, the device provides a descriptive answer generated by at least one machine learning model using the query and the corresponding digital assets, and displays an affordance for the descriptive answer.
Claims Coverage
Independent claims: clm-00001 (method), clm-00012 (non-transitory computer readable storage medium), and clm-00023 (client computing device). Across these independent claims, the core coverage includes three major inventive features: user-account-based vector generation, vector similarity scoring and filtering, and displaying affordances for filtered digital assets, with dependent refinements such as metadata/content embedding construction, retrieval across keyword and semantic paths, relevance-score calibration, threshold similarity filtering, and question-to-descriptive-answer generation.
User-account-associated vector generation for query relevance
Receiving a query associated with a user account, and generating a query vector based at least in part on the query, and generating an output vector based at least in part on the query vector and the user account vector.
Digital asset vector similarity scoring and similarity-based filtering
Obtaining, based at least in part on the query, a plurality of digital asset vectors, comparing the output vector to the plurality of digital asset vectors to generate respective similarity scores for the plurality of digital asset vectors, and filtering the plurality of digital asset vectors in accordance with the similarity scores to establish a filtered plurality of digital asset vectors.
Displaying affordances for filtered digital assets
Displaying, in accordance with the filtered plurality of digital asset vectors, respective affordances for the respective digital assets that correspond to the filtered plurality of digital asset vectors.
Retrieval tree and keyword/semantic vector index combination for digital asset vectors
Generating retrieval trees from query keywords, performing keyword-based searching using a keyword index to obtain a first plurality of digital asset vectors, generating query embeddings, performing semantic-based searching using a vector index to obtain a second plurality of digital asset vectors, and generating the final plurality of digital asset vectors from the first plurality of digital asset vectors and the second plurality of digital asset vectors.
Metadata and content combined embedding construction for digital asset vectors
Obtaining a first vector using a transformer-based large language model from associated metadata and obtaining a second vector using a machine learning model from data content of the corresponding digital asset, and combining the first vector and the second vector to form a digital asset vector.
Calibrating and combining relevance scores across retrieval sources
Calibrating relevance scores of digital asset vectors from a second set to match those of a first set and then combining, via adjusting and/or filtering, the two sets into a single plurality of digital asset vectors.
Threshold similarity score filtering
Filtering the plurality of digital asset vectors by excluding those whose similarity scores do not meet a threshold similarity score.
Question detection and descriptive answer generation
Determining that a query contains at least one question, providing the query and corresponding digital assets to at least one machine learning model so it generates a descriptive answer, and displaying an affordance for that descriptive answer.
The independent claims cover a vector-based search that personalizes results using a user account vector to generate an output vector, compares it to digital asset vectors to compute similarity scores, filters the results based on those similarity scores, and displays affordances for the filtered assets; dependent claim refinements further specify user-account vector construction, digital-asset vector construction from metadata and content, combined keyword/semantic retrieval, relevance-score calibration across retrieval paths, optional threshold similarity filtering, and generating descriptive answers when the query includes questions.
Stated Advantages
Not explicitly described in patent.
Documented Applications
Not explicitly described in patent.
Interested in licensing this patent?