# US 11,907,657 B1 · Dynamically extracting n-grams for automated vocabulary updates > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,907,657 B1 - Document ID: US11907657B1 - Status: Granted - Kind: B1 - Application number: 18/217,523 - Filed: 2023-06-30 - Granted: 2024-02-20 - Earliest claimed: 2023-06-30 - Assignees: Intuit Inc. - Applicants: Intuit Inc. - Inventors: Byungkyu Kang; Shivakumara Narayanaswamy; Andrew Mattarella-Micke - Web page: https://patentgrep.com/patent/US11907657B1 ## CPC - G06F 40/279: Recognition of textual entities - G06F 40/242: Dictionaries - G06F 40/284: Lexical analysis, e.g. tokenisation or collocates - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract Systems and methods dynamically extracting n-grams for automated vocabulary updates. Text is received. An n-gram extracted from the text is matched to a canonical n-gram from a vocabulary to identify a tag for the text. An n-gram weight is computed for the n-gram extracted from the text. The n-gram weight may be computed by adjusting a term frequency of the n-gram. A relevancy score is computed for the tag using the n-gram weight and using an n-gram frequency of the canonical n-gram. The relevancy score is computed by dividing the n-gram weight by a value proportional to the n-gram frequency of the canonical n-gram. The relevancy score of the n-gram is presented. ## Claims ### Claim 1 A method, comprising: receiving text; matching an n-gram extracted from the text to a canonical n-gram from a vocabulary to identify a tag for the text; computing an n-gram weight for the n-gram extracted from the text, wherein the n-gram weight is computed by adjusting a term frequency of the n-gram; computing a relevancy score for the tag using the n-gram weight and using an n-gram frequency of the canonical n-gram, wherein the relevancy score is computed by dividing the n-gram weight by a value proportional to the n-gram frequency of the canonical n-gram; and presenting the relevancy score of the n-gram. ### Claim 2 (depends on 1) The method of claim 1, further comprising: ranking the n-gram within a set of n-grams extracted from the text using a maximum marginal relevance algorithm and using an embedding vector for the n-gram. ### Claim 3 (depends on 1) The method of claim 1, further comprising: grouping the tag with a set of tags corresponding to the canonical n-gram to combine a set of n-gram weights comprising the n-gram weight to form a combined weight, wherein the set of tags correspond to a set of n-grams mapped to the canonical n-gram; and computing the relevancy score using the combined weight for the n-gram weight. ### Claim 4 (depends on 1) The method of claim 1, further comprising: selecting a threshold number of n-grams using the relevancy score; and presenting the relevancy score identifying the n-gram as a keyword of the text. ### Claim 5 (depends on 1) The method of claim 1, wherein the vocabulary is periodically updated to include data from a recent update interval and remove data from a previous update interval, wherein the n-gram is identified with an n-gram tag of a set of tags that identify metadata of the text, wherein the metadata comprises n-gram identifiers, document identifiers, participant identifiers, and timestamp identifiers. ### Claim 6 (depends on 1) The method of claim 1, wherein the vocabulary is updated by: selecting a set of historical data comprising one or more conversations and search queries during a time window comprising a plurality of update intervals; extracting a set of n-grams from the set of historical data for the time window; computing n-gram statistics for the set of n-grams for the time window; calculating canonical statistics for a set of canonized n-grams generated from the set of n-grams for the time window; generating a mapping between the set of canonical n-grams and the set of n-grams; and presenting the canonical statistics with the set of canonized n-grams. ### Claim 7 (depends on 6) The method of claim 6, further comprising: computing the n-gram statistics comprising: a query number identifying a number of the search queries in the set of historical data, an n-gram number of a set of n-gram numbers identifying a number of times the n-gram appears in the conversations of the set of historical data, and a conversation number of a set of conversation numbers identify a number of conversations in which the n-gram of the n-gram number appears. ### Claim 8 (depends on 6) The method of claim 6, further comprising: calculating the canonical statistics comprising a canonical n-gram frequency of a set of canonical n-gram frequencies identifying a frequency of the canonical n-gram in each document of the set of historical data, wherein a document is one of a conversation and a search query. ### Claim 9 (depends on 6) The method of claim 6, further comprising: generating a table comprising rows for the canonical n-grams and columns for the canonical statistics. ### Claim 10 (depends on 6) The method of claim 6, further comprising: canonizing the set of normalized n-grams using the n-gram statistics to form a set of canonized n-grams. ### Claim 11 A system comprising: a processor; an application executing the processor and configured for: receiving text; matching an n-gram extracted from the text to a canonical n-gram from a vocabulary to identify a tag for the text; computing an n-gram weight for the n-gram extracted from the text, wherein the n-gram weight is computed by adjusting a term frequency of the n-gram; computing a relevancy score for the tag using the n-gram weight and using an n-gram frequency of the canonical n-gram, wherein the relevancy score is computed by dividing the n-gram weight by a value proportional to the n-gram frequency of the canonical n-gram; and presenting the relevancy score of the n-gram. ### Claim 12 (depends on 11) The system of claim 11, wherein the application is further configured for: ranking the n-gram within a set of n-grams extracted from the text using a maximum marginal relevance algorithm and using an embedding vector for the n-gram. ### Claim 13 (depends on 11) The system of claim 11, wherein the application is further configured for: grouping the tag with a set of tags corresponding to the canonical n-gram to combine a set of n-gram weights comprising the n-gram weight to form a combined weight, wherein the set of tags correspond to a set of n-grams mapped to the canonical n-gram; and computing the relevancy score using the combined weight for the n-gram weight. ### Claim 14 (depends on 11) The system of claim 11, wherein the application is further configured for: selecting a threshold number of n-grams using the relevancy score; and presenting the relevancy score identifying the n-gram as a keyword of the text. ### Claim 15 (depends on 11) The system of claim 11, wherein the vocabulary is periodically updated to include a recent update interval and remove a previous update interval, wherein the n-gram is identified with an n-gram tag of a set of tags that identify metadata of the text, wherein the metadata comprises n-gram identifiers, document identifiers, participant identifiers, and timestamp identifiers. ### Claim 16 (depends on 11) The system of claim 11, wherein the vocabulary is updated by: selecting a set of historical data comprising one or more conversations and search queries during a time window comprising a plurality of update intervals; extracting a set of n-grams from the set of historical data for the time window; computing n-gram statistics for the set of n-grams for the time window; calculating canonical statistics for a set of canonized n-grams generated from the set of n-grams for the time window; generating a mapping between the set of canonical n-grams and the set of n-grams; and presenting the canonical statistics with the set of canonized n-grams. ### Claim 17 (depends on 16) The system of claim 16, wherein the application is further configured for: computing the n-gram statistics comprising: a query number identifying a number of the search queries in the set of historical data, an n-gram number of a set of n-gram numbers identifying a number of times the n-gram appears in the conversations of the set of historical data, and a conversation number of a set of conversation numbers identify a number of conversations in which the n-gram of the n-gram number appears. ### Claim 18 (depends on 16) The system of claim 16, wherein the application is further configured for: calculating the canonical statistics comprising a canonical n-gram frequency of a set of canonical n-gram frequencies identifying a frequency of the canonical n-gram in each document of the set of historical data, wherein a document is one of a conversation and a search query. ### Claim 19 (depends on 16) The system of claim 16, wherein the application is further configured for: generating a table comprising rows for the canonical n-grams and columns for the canonical statistics. ### Claim 20 A method, comprising: selecting a set of historical data comprising one or more conversations and search queries during a time window comprising a plurality of update intervals; extracting a set of n-grams from the set of historical data for the time window; computing n-gram statistics for the set of n-grams for the time window; calculating canonical statistics for a set of canonized n-grams generated from the set of n-grams for the time window; generating a mapping between the set of canonical n-grams and the set of n-grams; and presenting the canonical statistics with the set of canonized n-grams.