# US 2026/0289070 A1 · System

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0289070 A1
- Document ID: US20260289070A1
- Status: Published application
- Kind: A1
- Application number: 19/564,193
- Filed: 2026-03-12
- Published: 2026-09-24
- Earliest claimed: 2025-03-19
- Assignees: SoftBank Group Corp.
- Applicants: SoftBank Group Corp.
- Inventors: Hironobu KIKUCHI
- Web page: https://patentgrep.com/patent/US20260289070A1

## CPC

- G06F 40/151: Transformation
- G06F 16/334: Query execution (filtering based on additional data G06F16/335)
- G06F 40/186: Templates
- G06N 3/092: Reinforcement learning
- G06T 11/00: Two-dimensional [2D] image generation
- H04L 67/535: {Tracking the activity of the user (network monitoring arrangements H04L43/00; recording of computer activity G06F11/34)}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)

## Abstract

A system includes a processor that is configured to collect data including a user's browsing history, profile information, preferences, and emotional state, analyze the collected data to identify an interest of the user and the emotional state of the user, and generate a prompt sentence for instructing a generative AI model to generate specific information.

## Claims

### Claim 1

A system comprising:
circuitry configured to:
acquire, via a communication interface coupled to a packet-switched network, behavior log data of a user from a terminal device and from one or more external information sources, the behavior log data comprising interaction history records, attribute data, preference indicator data, and affective state data;
execute an analysis operation on the behavior log data using a machine learning model and a natural language processing pipeline to extract user characteristic data comprising a topic interest distribution, a preferred information quantity parameter, a style profile comprising writing style indicators, region-specific expression indicators, emphasis level indicators, and tense preference indicators;
construct a constrained instruction sequence based on the user characteristic data, the constrained instruction sequence comprising constraint condition parameters explicitly specifying a target topic field, an output length range, a language style directive, a region-specific expression directive, an emphasis level directive, and a tense specification directive;
input the constrained instruction sequence into a generative neural network model comprising a transformer-based architecture and acquire generated text data from the generative neural network model;
execute a post-processing operation on the generated text data, the post-processing operation comprising at least one of a style conversion operation, a region-specific expression insertion operation, and a tense adjustment operation, to produce output content data conforming to the style profile; and
transmit the output content data to the terminal device via the communication interface.

### Claim 2 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
compute the topic interest distribution by applying a topic modeling operation to the interaction history records to generate topic probability vectors, and derive the style profile by analyzing linguistic features of prior user-generated text data using the natural language processing pipeline to identify patterns in sentence length, vocabulary level, formality, and regional expression usage.

### Claim 3 (depends on 2)

The system according to claim 2, wherein the circuitry is further configured to:
select a prompt template from a plurality of prompt templates stored in an information storage device based on the topic interest distribution, and populate the selected prompt template with the constraint condition parameters derived from the user characteristic data to generate the constrained instruction sequence.

### Claim 4 (depends on 3)

The system according to claim 3, wherein the constraint condition parameters further comprise:
a tense control parameter specifying whether the generated text data is to be expressed in past tense, present tense, or future tense, and an emphasis control parameter specifying whether the generated text data is to use soft expressions or emphasized expressions, the tense control parameter and the emphasis control parameter derived from the tense preference indicators and the emphasis level indicators in the style profile.

### Claim 5 (depends on 1)

The system according to claim 1, wherein the post-processing operation comprises:
the style conversion operation configured to transform vocabulary and sentence structure of the generated text data to match the writing style indicators in the style profile, the region-specific expression insertion operation configured to replace generic expressions with region-specific equivalent expressions based on the region-specific expression indicators, and the tense adjustment operation configured to convert verb forms in the generated text data to conform to the tense specification directive.

### Claim 6 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
acquire evaluation data and interaction event data from the terminal device, compute a reward signal based on the evaluation data and engagement metrics derived from the interaction event data, and update the user characteristic data and generation condition parameters for the constrained instruction sequence using a reinforcement learning operation that adjusts weighting values of the constraint condition parameters based on the reward signal.

### Claim 7 (depends on 6)

The system according to claim 6, wherein the circuitry is further configured to:
compare the output content data with the constraint condition parameters to compute a conformance score, and when the conformance score is below a threshold value, modify the constrained instruction sequence by adjusting at least one of the output length range, the emphasis level directive, and the tense specification directive, and re-input the modified constrained instruction sequence into the generative neural network model to obtain refined generated text data.

### Claim 8 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
generate a plurality of constrained instruction sequences corresponding to a plurality of topic fields identified in the topic interest distribution, input each constrained instruction sequence into the generative neural network model to generate topic-specific text data for each topic field, and aggregate the topic-specific text data into a composite output content data structure for transmission to the terminal device.

### Claim 9 (depends on 1)

The system according to claim 1, wherein the generative neural network model comprises:
an embedding layer configured to convert input tokens into multidimensional tensor representations, a positional encoding module, a plurality of transformer layers each comprising multi-head self-attention and feed-forward sublayers with layer normalization, and a decoding stage configured to generate a probability distribution over a vocabulary and select output tokens using a sampling strategy with configurable temperature and top-k parameters.

### Claim 10 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
classify the affective state data into an affective state category using a neural network classifier, and include the affective state category as an additional constraint condition parameter in the constrained instruction sequence to condition the generative neural network model to generate text data with a tone corresponding to the affective state category.

### Claim 11 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
apply a quality validation operation to the generated text data by computing a relevance score based on semantic similarity between the generated text data and the topic interest distribution, and a style conformance score based on comparison of linguistic features of the generated text data against the style profile, and discard generated text data having a relevance score or style conformance score below respective threshold values.

### Claim 12 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
determine a delivery timing based on temporal patterns in the interaction history records indicating time periods of high user engagement, and schedule transmission of the output content data to the terminal device to coincide with the determined delivery timing.

### Claim 13 (depends on 6)

The system according to claim 6, wherein the circuitry is further configured to:
generate a plurality of variant constrained instruction sequences differing in at least one constraint condition parameter, input each variant into the generative neural network model to produce variant output content data, transmit different variants to the terminal device across successive sessions, and compute comparative engagement metrics for each variant to update the constraint condition parameters via the reinforcement learning operation.

### Claim 14 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
acquire contextual data from one or more external information processing apparatuses via the communication interface, the contextual data comprising at least one of trending topic data, temporal event data, and geographic relevance data, and incorporate the contextual data into the constrained instruction sequence as supplemental constraint condition parameters.

### Claim 15 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
assign the user to a user segment based on clustering the user characteristic data with characteristic data of other users using an unsupervised learning operation, and apply segment-level constraint condition parameters derived from aggregated characteristics of the user segment as default values in the constrained instruction sequence when individual user characteristic data is insufficient.

### Claim 16 (depends on 1)

The system according to claim 1, wherein the circuitry is further configured to:
generate supplemental visual content data by inputting a visual generation instruction derived from the output content data into an image generation neural network, and transmit the supplemental visual content data together with the output content data to the terminal device for combined presentation.

### Claim 17 (depends on 1)

The system according to claim 1, wherein the output content data comprises personalized news articles, recommendation summaries, or notification messages adapted to the user, the behavior log data comprises web browsing history and content viewing records, and the style profile governs tone, formality, and regional dialect of the output content data for a personalized information delivery service.

### Claim 18

A system comprising:
circuitry configured to:
acquire behavior log data of a user from a terminal device via a communication interface coupled to a packet-switched network;
execute an analysis operation on the behavior log data to extract user characteristic data comprising a topic interest distribution and a style profile;
construct a constrained instruction sequence comprising constraint condition parameters derived from the user characteristic data and input the constrained instruction sequence into a generative neural network model to generate text data;
execute a post-processing operation on the generated text data comprising at least one of a style conversion operation, a region-specific expression insertion operation, and a tense adjustment operation to produce output content data; and
transmit the output content data to the terminal device.

### Claim 19 (depends on 18)

The system according to claim 18, wherein the circuitry is further configured to:
acquire evaluation data from the terminal device and update the user characteristic data and generation condition parameters using a reinforcement learning operation based on the evaluation data.

### Claim 20

A method comprising:
acquiring, by circuitry via a communication interface coupled to a packet-switched network, behavior log data of a user from a terminal device and from one or more external information sources, the behavior log data comprising interaction history records, attribute data, preference indicator data, and affective state data;
executing an analysis operation on the behavior log data using a machine learning model and a natural language processing pipeline to extract user characteristic data comprising a topic interest distribution, a preferred information quantity parameter, a style profile comprising writing style indicators, region-specific expression indicators, emphasis level indicators, and tense preference indicators;
constructing a constrained instruction sequence based on the user characteristic data, the constrained instruction sequence comprising constraint condition parameters explicitly specifying a target topic field, an output length range, a language style directive, a region-specific expression directive, an emphasis level directive, and a tense specification directive;
inputting the constrained instruction sequence into a generative neural network model comprising a transformer-based architecture and acquiring generated text data from the generative neural network model;
executing a post-processing operation on the generated text data, the post-processing operation comprising at least one of a style conversion operation, a region-specific expression insertion operation, and a tense adjustment operation, to produce output content data conforming to the style profile; and
transmitting the output content data to the terminal device via the communication interface.
