US 12,386,872 B2
Language detection of user input text for online gaming
Nikolaus Sonntag, Foster City, CA (US); Aswath Manoharan, Sunnyvale, CA (US); I-Wu Lu, San Bruno, CA (US); Eric Holmdahl, San Francisco, CA (US); and Madhok Shivaratre, San Mateo, CA (US)
Assigned to Roblox Corporation, San Mateo, CA (US)
Filed by Roblox Corporation, San Mateo, CA (US)
Filed on Apr. 30, 2024, as Appl. No. 18/651,020.
Application 18/651,020 is a continuation of application No. 17/967,420, filed on Oct. 17, 2022, granted, now 11,989,215.
Application 17/967,420 is a continuation of application No. 16/858,467, filed on Apr. 24, 2020, granted, now 11,475,054.
Prior Publication US 2024/0330337 A1, Oct. 3, 2024
This patent is subject to a terminal disclaimer.
Int. Cl. G06F 16/00 (2019.01); G06F 16/3329 (2025.01); G06F 16/3332 (2025.01); G06F 16/335 (2019.01); G06F 40/263 (2020.01); G06F 40/56 (2020.01); G06N 20/00 (2019.01)
CPC G06F 16/3334 (2019.01) [G06F 16/3329 (2019.01); G06F 16/335 (2019.01); G06F 40/263 (2020.01); G06F 40/56 (2020.01); G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
parsing a user query into n-grams;
identifying, with a machine-learning model, candidate languages for the user query and respective confidence scores for the candidate languages based on the n-grams from the user query, wherein the machine learning model is trained based on at least one multilingual text corpus and game-related data;
identifying one or more response matches to the user query in language-specific game databases and one or more respective match scores for the one or more response matches, wherein each response match is in a respective language of the candidate languages;
determining a weighted score for each of the one or more response matches by weighting the one or more respective match scores based on a respective confidence score for the respective language of a respective response match; and
providing a response of search results including game information associated with particular response matches, based, at least in part, on the respective weighted score.