# US 2022/0045975 A1 · Communication content tailoring > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2022/0045975 A1 - Document ID: US20220045975A1 - Status: Published application - Kind: A1 - Application number: 16/987,024 - Filed: 2020-08-06 - Published: 2022-02-10 - Earliest claimed: 2020-08-06 - Assignees: International Business Machines Corporation - Applicants: International Business Machines Corporation - Inventors: Frederik Frank Flöther; Shikhar Kwatra; Patrick Lustenberger; Stefan Ravizza - Web page: https://patentgrep.com/patent/US20220045975A1 ## CPC - H04L 51/046: Interoperability with other network applications or services - G06F 40/253: Grammatical analysis; Style critique - G06F 40/35: Discourse or dialogue representation - G06F 40/56: Natural language generation - G06N 3/0475: Generative networks - G06N 3/09: Supervised learning - G06N 3/092: Reinforcement learning - G06N 20/00: Machine learning - G06N 20/20: Ensemble learning - H04L 51/063: Content adaptation, e.g. replacement of unsuitable content - H04L 51/216: Handling conversation history, e.g. grouping of messages in sessions or threads - G06N 3/045: Combinations of networks - G06N 5/02: Knowledge representation; Symbolic representation - H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M) - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A method for personalizing a message between a sender and a receiver is provided. The method comprises semantically analyzing a communication history to form a knowledge graph, deriving formality level values using a first trained ML model, analyzing parameter values of replies to determine receiver impact score, and training a second ML system to generate a model to predict the receiver impact score value. The method also comprises selecting a linguistic expression in a message being drafted, determining an expression intent, modifying the linguistic expression based on the formality level and the expression intent to generate a modified linguistic expression, and testing whether the modified linguistic expression has an increased likelihood of a higher receiver impact score. The method also comprises repeating selecting the linguistic expression, determining the expression intent, modifying the linguistic expression, and testing until a stop criterion is met. ## Claims ### Claim 1 A method for personalizing a message between a sender and a receiver, said method comprising: semantically analyzing a communication history between said sender and said receiver and forming a knowledge graph between a sender identifier identifying said sender and a receiver identifier identifying said receiver; deriving formality level values between said sender and said receiver from said knowledge graph using a first trained machine-learning model; analyzing parameter values of replies in said communication history to determine receiver impact score values; training a second machine-learning system to generate a model to predict said receiver impact score values based on said knowledge graph and said formality level; selecting a linguistic expression in a message being drafted; determining an expression intent of said selected linguistic expression; modifying said linguistic expression based on said formality level and said expression intent to generate a modified linguistic expression; testing whether said modified linguistic expression has an increased likelihood to lead to a higher receiver impact score value using a third trained machine-learning model; and repeating said selecting said linguistic expression, determining said expression intent, modifying said linguistic expression, and testing for said higher receiver impact score until a stop criterion is met. ### Claim 2 (depends on 1) The method according to claim 1 wherein said third machine-learning model is a reinforcement learning model. ### Claim 3 (depends on 2) The method according to claim 2 wherein said training of said third machine-learning model also comprises using a bi-directional transformer to predict said expression intent. ### Claim 4 (depends on 3) The method according to claim 3 wherein said training of said bi-directional transformer further comprises using said communication history, said knowledge graph, formality levels, and receiver impact score values as training data. ### Claim 5 (depends on 1) The method according to claim 1 wherein: said receiver identifier is a plurality of receiver identifiers used to identify a plurality of users, and said modified linguistic expression is built taking effects of all receiver identifiers into account. ### Claim 6 (depends on 1) The method according to claim 1 wherein said modification of said linguistic expression is influenced by outcomes of at least one selected out of said formality level analysis, a confidentiality level analysis of said message, a message topic analysis, and a tone analysis. ### Claim 7 (depends on 1) The method according to claim 1 wherein: said modification of said linguistic expression is performed by at least one selected from the group consisting of a replacement of a word, spinning of sentence structure, reordering of content, deletion of a word, re-shuffle an order of message blocks, use of synonyms, wording which was used by said receiver in past communication, adjust said style, and a GPT2 transformer transformation, and said GPT2 transformer transformation is inculcated with partial sentence creation to generate a segment of said paragraph in a personalized manner by synthesizing personalized text for said user. ### Claim 8 (depends on 1) The method according to claim 1 wherein said semantic analysis of said communication history comprises: identifying topics with a latent Dirichlet allocation (LDA) topic model. ### Claim 9 (depends on 1) The method according to claim 1 wherein said formality level is obtained with a bag-of-words model and said first trained machine-learning model is a Gaussian Naïve Bayes classifier. ### Claim 10 (depends on 1) The method according to claim 1 wherein said linguistic expression is further refined by predicting a mood or task of said receiver at a time at which said message is set to arrive using IoT sensors or calendar information. ### Claim 11 (depends on 1) The method according to claim 1 wherein said receiver impact score is influenced by at least one selected out of the group consisting of IoT sensor data, wearable system data, computer vision data, presence/absence of a reply, timing of a reply, timing of said reply in relation to other replies by said receiver or different recipients, timing of said reply in relation to time zones and calendars, presence/absence of out-of-office messages, length of said reply, content of said reply, a task requested by said sender being carried out by said receiver, and an emoji. ### Claim 12 (depends on 1) The method according to claim 1 wherein said message is a written message or a voice message. ### Claim 13 (depends on 1) The method according to claim 1 wherein: said analysis reveals a lack of said communication history, and a secondary communication history replaces said communication history, wherein said secondary communication history is between said receiver and a secondary sender. ### Claim 14 A message personalizing system for personalizing a message between a sender and a receiver, said message personalizing system comprising: a memory; and a processor in communication with the memory, the processor being configured to perform operations comprising: semantically analyzing a communication history between said sender and said receiver and adapted for forming a knowledge graph between a sender identifier identifying said sender and a receiver identifier identifying said receiver; deriving from said knowledge graph formality level values between said sender and said receiver using a first trained machine-learning model; analyzing parameter values of replies in said communication history to determine receiver impact score values; training a second machine-learning system to generate a model to predict said receiver impact score values based on said knowledge graph and said formality level; selecting a linguistic expression in a message being drafted; determining an expression intent of said linguistic expression; modifying said linguistic expression based on said formality level and said expression intent to generate a modified linguistic expression; a testing if said modified linguistic expression has an increased likelihood of a higher receiver impact score value using a third trained machine-learning model; and repeating said selecting said linguistic expression, determining said expression intent, modifying said linguistic expression, and testing for said higher receiver impact score until a stop criterion is met. ### Claim 15 (depends on 14) The message personalizing system according to claim 14 wherein said third machine-learning model is a reinforcement learning model. ### Claim 16 (depends on 15) The message personalizing system according to claim 15 wherein said training of said third machine-learning model comprises: using a bi-directional transformer to predict said expression intent. ### Claim 17 (depends on 16) The message personalizing system according to claim 16 wherein said training of said bi-directional transformer comprises: using said communication history, said knowledge graph, said formality levels, and said receiver impact score values as training data. ### Claim 18 (depends on 14) The message personalizing system according to claim 14 wherein: said receiver identifier is a plurality of receiver identifiers used to identify a plurality of users, and said modification means generates said modified linguistic expression based on said formality level and said expression intent between said sender and said plurality of users. ### Claim 19 (depends on 14) The message personalizing system according to claim 14 wherein said modification of said linguistic expression is influenced by outcomes of at least one selected from the group consisting of said formality level analysis, a confidentiality level analysis of said message, a message topic analysis, and a tone analysis. ### Claim 20 (depends on 14) The message personalizing system according to claim 14 wherein: said modification of said linguistic expression is performed by at least one selected from the group consisting of a replacement of a word, spinning of sentence structure, reordering of content, deletion of a word, re-shuffle an order of message blocks, use of synonyms, wording which was used by said receiver in past communication, adjustment of style, and a GPT2 transformer transformation, and said GPT2 transformer transformation is inculcated with partial sentence creation to generate a segment of said paragraph in a personalized manner by synthesizing personalized text for said user. ### Claim 21 (depends on 14) The message personalizing system according to claim 14 wherein said semantic analysis of said communication history comprises identifying topics with a latent Dirichlet allocation (LDA) topic model. ### Claim 22 (depends on 14) The message personalizing system according to claim 14 wherein said formality level is obtained with a bag-of-words model and said first trained machine-learning model is a Gaussian Naïve Bayes classifier. ### Claim 23 (depends on 14) The message personalizing system according to claim 14 wherein said linguistic expression is further refined by predicting a mood or a task of said receiver at a time at which said message is set to arrive using IoT sensors or calendar information. ### Claim 24 (depends on 14) The message personalizing system according to claim 14 wherein said receiver impact score is influenced by at least one selected from the group consisting of IoT sensor data, wearable system data, computer vision data, presence/absence of a reply, timing of a reply, timing of a reply in relation to other replies by said receiver or different recipients, timing of a reply in relation to time zones and calendars, presence/absence of out-of-office messages, length of said reply, content of said reply, a task requested by said sender being carried out by said receiver, and an emoji. ### Claim 25 A computer program product for personalizing a message between a sender and a receiver, said computer program product comprising a computer readable storage medium having program instructions embodied therewith, said program instructions being executable by one or more computing systems or controllers to cause said one or more computing systems to: analyze, semantically, a communication history between said sender and said receiver to generate an analysis wherein said analysis is adapted for forming a knowledge graph between a sender identifier identifying said sender and a receiver identifier identifying said receiver, derive from said knowledge graph formality level values between said sender and said receiver using a first trained machine-learning model, analyze parameter values of replies in said communication history to determine receiver impact score values, train a second machine-learning system to generate a model to predict said receiver impact score value based on said knowledge graph and said formality level, select a linguistic expression in a message being drafted, determine an expression intent of said linguistic expression, generate a modified linguistic expression by modifying said linguistic expression based on said formality level and said expression intent to, test whether said modified linguistic expression has an increased likelihood of a higher receiver impact score value by using a third trained machine-learning model, and repeat said program instruction of said selection, said determination, said generation, and said test until a stop criterion is met.