# US 2026/0094061 A1 · Identifying noise in verbal feedback using artificial text from non-textual parameters and transfer learning

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

- Number: US 2026/0094061 A1
- Document ID: US20260094061A1
- Status: Published application
- Kind: A1
- Application number: 19/259,250
- Filed: 2025-07-03
- Published: 2026-04-02
- Earliest claimed: 2020-08-17
- Assignees: Microsoft Technology Licensing, LLC
- Applicants: Microsoft Technology Licensing, LLC
- Inventors: Isha Aniruddha OKE; Logan M. RINGER; Vandana GUNUPUDI; Yong Hui MA; Prabagaran SANTHANAKRISHNAN; Mengtao SUN
- Web page: https://patentgrep.com/patent/US20260094061A1

## CPC

- G06F 40/30: Semantic analysis
- G06N 20/00: Machine learning
- G06F 16/36: Creation of semantic tools, e.g. ontology or thesauri
- G06F 18/24: Classification techniques
- G06F 40/237: Lexical tools
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06N 3/09: Supervised learning
- G06N 3/096: Transfer learning
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/20: Ensemble learning
- G06N 3/045: Combinations of networks
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Methods and systems are provided for classifying free-text content using machine learning. Free-text content (e.g., customer feedback) and parameter values organized according to a schema are received. A free-text corpus is generated, and an artificial-text corpus is generated by applying rules to the parameter values. The artificial-text corpus is generated by converting the parameter values into a finite set of words based on the rules and concatenating the words of the finite set of words into a fixed sequence wordlist. Feature vectors (e.g., sentence embeddings) based on the free-text corpus and the artificial-text corpus are combined and forwarded to a machine learning model for classification. The machine learning model may be trained with a bias towards a specified metric (e.g., precision, recall, F1 score). The model may be trained using transfer learning with training data from a different category of free-text content (e.g., a different category of customer feedback).

## Claims

### Claim at position 1

-20. (canceled)

### Claim 21

A system comprising:
a processor; and
a memory storing instructions that, when executed, perform operations comprising:
receiving free-text content corresponding to user satisfaction feedback with performance of software, wherein the free-text content is naturally spoken or written feedback submitted by a user;
in response to receiving the free-text content, retrieving parameter values associated with the free-text content, wherein the parameter values are organized in accordance with a schema and identifies a feature of the software;
generating an artificial-text corpus by applying a rule to at least one of the parameter values, wherein the rule specifies instructions for converting the parameter values to a set of words, the artificial-text corpus comprising the set of words;
generating sentence embeddings based on the free-text content and the artificial-text corpus;
generating a classification relating to the free-text content based on the sentence embeddings, the classification indicating whether the software is functioning as expected; and
based on determining the software is not functioning as expected, performing a corrective action for the software.

### Claim 22 (depends on 21)

The system of claim 21, wherein the free-text content and the parameter values are extracted from a data file generated in response to submission of the user satisfaction feedback.

### Claim 23 (depends on 21)

The system of claim 21, wherein the parameter values specify at least one of:
an application identifier;
a device running the software; or
an operating system version of the device running the software.

### Claim 24 (depends on 21)

The system of claim 21, wherein the parameter values are automatically retrieved from a data file in response to receiving the free-text content.

### Claim 25 (depends on 21)

The system of claim 21, the operations further comprising:
generating a free-text corpus based on the free-text content; and
generating free-text numerical feature vectors by vectorizing the free-text corpus.

### Claim 26 (depends on 21)

The system of claim 21, wherein generating the sentence embeddings comprises:
generate first embeddings based on the free-text content;
generate second embeddings based on the artificial-text corpus; and
generating the sentence embeddings by combining the first embeddings and the second embeddings.

### Claim 27 (depends on 21)

The system of claim 21, wherein generating the classification comprises:
providing, to a machine learning model, first feature data based on the free-text content and second feature data based on the artificial-text corpus.

### Claim 28 (depends on 27)

The system of claim 27, wherein the machine learning model comprises a supervised learning model.

### Claim 29 (depends on 27)

The system of claim 27, wherein each of the first feature data and the second feature data comprises at least one of: sentence embeddings or word embeddings.

### Claim 30 (depends on 27)

The system of claim 27, wherein the machine learning model is trained based on historical free-text data, corresponding historical parameter values, and corresponding historical classification results.

### Claim 31 (depends on 21)

The system of claim 21, wherein the corrective action for the software comprises updating the software.

### Claim 32 (depends on 21)

The system of claim 21, wherein the corrective action for the software comprises providing technical support for the software.

### Claim 33 (depends on 21)

The system of claim 21, wherein words of the set of words are concatenated into a fixed sequence wordlist.

### Claim 34 (depends on 21)

The system of claim 21, wherein the rule specifies at least one of:
converting a parameter value to a particular word; or
using a parameter value having a particular parameter name as a word.

### Claim 35 (depends on 21)

The system of claim 21, wherein the rule specifies at least one of:
converting numerals having a particular parameter name to a word; or
concatenating a parameter value with a corresponding parameter name to form a word.

### Claim 36

A method comprising:
receiving free-text content corresponding to user satisfaction feedback with performance of software, wherein the free-text content is naturally spoken or written feedback submitted by a user;
in response to receiving the free-text content, retrieving a parameter value associated with the free-text content, wherein the parameter value is organized in accordance with a schema and identifies a feature of the software;
generating an artificial-text corpus by applying a rule to the parameter value, wherein the rule specifies instructions for converting the parameter value to a set of words, the artificial-text corpus comprising the set of words;
generating sentence embeddings based on the free-text content and the artificial-text corpus;
generating a classification for the free-text content based on the sentence embeddings, the classification indicating whether the user satisfaction feedback is actionable; and
based on determining the user satisfaction feedback is actionable, performing a corrective action for the software.

### Claim 37 (depends on 36)

The method of claim 36, wherein determining the user satisfaction feedback is actionable comprises determining the software is not functioning as expected.

### Claim 38 (depends on 36)

The method of claim 36, wherein the rule specifies generating a word indicating whether the parameter value is null or non-null.

### Claim 39 (depends on 36)

The method of claim 36, wherein generating the classification comprises:
providing, to a machine learning model, a first feature vector based on the free-text content and a second feature vector based on the artificial-text corpus.

### Claim 40

A device comprising:
a processor; and
a memory storing instructions that, when executed, perform operations comprising:
receiving free-text content corresponding to user satisfaction feedback with performance of software, wherein the free-text content is naturally spoken or written feedback of a user;
in response to receiving the free-text content, retrieving parameter values associated with the free-text content, wherein the parameter values identify a feature of the software;
generating an artificial-text corpus by applying a rule to at least one of the parameter values, wherein the rule specifies instructions for converting the parameter values to a set of words;
generating sentence embeddings based on the free-text content and the artificial-text corpus;
generating a classification relating to the free-text content based on the sentence embeddings, the classification indicating the software is not functioning as expected; and
based on determining the software is not functioning as expected, performing a corrective action for the software.
