# US 2022/0277147 A1 · Profile-based natural language message generation and selection > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2022/0277147 A1 - Document ID: US20220277147A1 - Status: Published application - Kind: A1 - Application number: 17/699,973 - Filed: 2022-03-21 - Published: 2022-09-01 - Earliest claimed: 2021-01-06 - Assignees: Capital One Services, LLC - Applicants: Capital One Services, LLC - Inventors: Huong NGUYEN; Isha Chaturvedi; Kalanand Mishra - Web page: https://patentgrep.com/patent/US20220277147A1 ## CPC - G06F 40/42: Data-driven translation - H04L 51/48: Message addressing, e.g. address format or anonymous messages, aliases - H04L 67/306: User profiles - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M) ## Abstract In some embodiments, text for user consumption may be generated based on an intended user action category and a user profile. In some embodiments, an action category, a plurality of text seeds, and a profile comprising feature values may be obtained. Context values may be generated based on the feature values, and text generation models may be obtained based on the text seeds. In some embodiments, messages may be generated using the text generation models based on the action category and the context values. Weights associated with the messages may be determined, and a first text message of the messages may be sent to an address associated with the profile based on the weights. Based on a reaction value obtained in response to the first message, a first expected allocation value may be updated based on the reaction value. ## Claims ### Claim at position 1 1-20. (canceled) ### Claim 21 A system for generating text using text seeds associated with a user that comprises a set of memory devices storing instructions and a set of processors configured to execute the instructions, the instructions comprising: obtaining a first text seed and a second text seed based on a user profile for a user; generating a first message using a first set of learning model parameters corresponding with the first text seed and a second message using a second set of learning model parameters corresponding with the second text seed; determining a first weight associated with the first message based on a first expected allocation value of the user profile, wherein the first expected allocation value is associated with the first text seed; determining a second weight associated with the second message based on a second expected allocation value of the user profile, wherein the second expected allocation value is associated with the second text seed; sending the first message to an address based on a comparison between the first weight and the second weight; and updating the first expected allocation value in response to a determination that a reaction value associated with the first message satisfies a threshold. ### Claim 22 (depends on 21) The system of claim 21, the instructions further comprising obtaining an action category based on the user profile, wherein: generating the first message comprises generating the first message based on the action category; and generating the second message comprises generating the second message based on the action category. ### Claim 23 (depends on 21) The system of claim 21, the instructions further comprising: using a machine learning model to determine a learning-predicted allocation value based on the first message; determining whether the learning-predicted allocation value satisfies a reward threshold; and updating the user profile based on a determination that the learning-predicted allocation value satisfies the reward threshold by updating a weight of the user profile associated with the first text seed. ### Claim 24 (depends on 21) The system of claim 21, further comprising: determining a plurality of possible categories based on a value of the user profile; determining a set of context values based on the value of the user profile; and selecting a category of the plurality of possible categories based on the set of context values, wherein generating the first message comprises generating the first message based on the category. ### Claim 25 (depends on 21) The system of claim 21, further comprising: obtaining a set of quantitative score changes of the user profile; and determining a category based on the set of quantitative score changes, wherein generating the first message comprises generating the first message based on the category. ### Claim 26 A method comprising: obtaining a first text seed and a second text seed based on a profile; generating a first message using a first set of learning model parameters corresponding with the first text seed and a second message using a second set of learning model parameters corresponding with the second text seed; determining a first weight associated with the first message based on a first allocation value of the profile, wherein the first allocation value is associated with the first text seed; determining a second weight associated with the second message based on a second allocation value of the profile, wherein the second allocation value is associated with the second text seed; sending the first message to an address based on a comparison between the first weight and the second weight; and updating the first allocation value in response to a determination that a reaction value associated with the first message satisfies a threshold. ### Claim 27 (depends on 26) The method of claim 26, further comprising: obtaining a set of electronic addresses associated with the profile; obtaining a plurality of categories; and selecting a first category of the plurality categories based on the set of electronic addresses, wherein generating the first message comprises generating the first message based on the first category. ### Claim 28 (depends on 26) The method of claim 26, further comprising determining whether a response message to the first message was received within a duration threshold, wherein updating the first allocation value comprises reducing the first allocation value in response to the response message not being received within the duration threshold. ### Claim 29 (depends on 26) The method of claim 26, further comprising: obtaining a response message corresponding to the first message, wherein the response message comprising an identifier associated with the first message; and determining the reaction value based on the response message. ### Claim 30 (depends on 26) The method of claim 26, further comprising: labeling the profile with a category based on a set of feature values of the profile and a set of feature value intervals associated with the category; and determining the first set of learning model parameters based on the category. ### Claim 31 (depends on 26) The method of claim 26, further comprising: obtaining a category based on the profile; and updating a plurality of text seeds based on the category, wherein obtaining the first text seed and the second text seed comprises obtaining the first and second text seeds from the plurality of text seeds. ### Claim 32 (depends on 26) The method of claim 26, wherein determining the first weight comprises determining an output of a function, wherein: the output is positively correlated with the first allocation value; and the output is negatively correlated with a count of a set of previous messages associated with the first allocation value; and determining the first weight comprises determining the first weight based on the output. ### Claim 33 A tangible, non-transitory, machine-readable media storing instructions that, when executed by one or more processors, effectuating operations comprising: obtaining a first seed and a second seed based on a profile; generating a first message using a first set of parameters corresponding with the first seed and a second message using a second set of parameters corresponding with the second seed; determining a first weight associated with the first message based on a first allocation value of the profile, wherein the first allocation value is associated with the first seed; determining a second weight associated with the second message based on a second allocation value of the profile, wherein the second allocation value is associated with the second seed; sending the first message to an address based on the first weight and the second weight; and updating the first allocation value in response to a determination that a reaction value associated with the first message satisfies a threshold. ### Claim 34 (depends on 33) The media of claim 33, the operations further comprising: obtaining a category and a text generation model; determining a set of context values based on the profile; and determining a first set of neural network layers of the text generation model based on the category; and determining a second set of neural network layers of the text generation model based on the set of context values, wherein: a first output of the first set of neural network layers is provided as an input to the second set of neural network layers to generate a second output; the first set of parameters comprises the first and second sets of neural network layers; and generating the first message comprises generating the first message based on the second output. ### Claim 35 (depends on 33) The media of claim 33, the operations further comprising: obtaining a candidate category of a set of categories; determining whether the profile satisfies a profile condition associated with the candidate category; and in response to a determination that the profile satisfies the profile condition, selecting the candidate category as a selected category, wherein generating the first message comprises generating the first message based on the selected category. ### Claim 36 (depends on 33) The media of claim 33, further comprising: obtaining a category; and obtaining a set of tokens based on the category, wherein the first message comprises generating the first message based on the set of tokens. ### Claim 37 (depends on 36) The media of claim 36, wherein generating the first message comprises: determining a set of attention weights corresponding to the set of tokens using a neural network of a text generation model; updating a portion of the set of attention weights to generate a set of masked attention weights, wherein the set of masked attention weights reduces weights corresponding to the set of tokens having a position value exceeding a threshold range; and determining a subsequent token using a sequence of neural network layers of the text generation model based on the set of masked attention weights and the set of tokens, wherein generating the first message comprises generating the first message based on the subsequent token. ### Claim 38 (depends on 33) The media of claim 33, further comprising: obtaining a category; generating a sequence of tokens comprising a set of tokens and a masked token based on the category; and using a set of neural network layers of a text generation model to replace the masked token with a predicted token based on an output of the text generation model, wherein generating the first message comprises generating the first message based on the predicted token. ### Claim 39 (depends on 33) The media of claim 33, the operations further comprising: obtaining a category, wherein: generating the first message comprises generating the first message using a first text generation model based on the category; and the first text generation model is configured with the first set of parameters; determining whether the reaction value indicates that a target action associated with the category was satisfied; in response to a determination that the reaction value indicates that the target action associated with the action category was not satisfied, generating a second message using a second text generation model different from the first text generation model; and sending the second message to the address. ### Claim 40 (depends on 33) The media of claim 33, wherein: the profile is associated with a category; the category is associated with a category allocation value; and the category allocation value is associated with the first seed; and determining the first weight comprises using the category allocation value to determine the first weight by: determining whether an allocation value field of the profile is filled; and in response to a determination that the allocation value field is not filled, using the category allocation value to determine the first weight.