Feature Space of retrieved patents
Date: 2026-07-04
Stage: Representation
Objective
Define an appropriate representation for technical features extracted from retrieved patent publications so that they can be meaningfully compared across patents.
Approach
Patent technical features are heterogeneous in both content and level of abstraction. A single feature may describe a technical problem, a proposed solution, a component or configuration, an intended effect, or a combination of these aspects. Representing such features as undifferentiated text may cause semantically different technical meanings to be treated as similar simply because they share terminology.
Accordingly, each feature retrieved by LLM must divided into multiple technological aspects such as Problem, Solution, Component and Effect.
Observations
The abstract of patent WO2026126674A1:
An acquisition unit 15 acquires measured voltage and current of a secondary battery E1. A first SOC estimation unit 16 estimates a change in a first SOC at least up to the start of CV charging during CCCV charging of the secondary battery E1. A second SOC estimation unit 17 increases a second SOC for a CV charging section in correspondence with the pace of an increase in the current-time product of current curve characteristics, as a change in the second SOC from the SOC of the secondary battery E1 when CV charging is started to the SOC when a full charge is reached.
LLM has featured two notably features:
CV-phase SOC estimation correction
To address unnatural SOC display progression during constant voltage (CV) charging, the apparatus utilizes a pre-stored current curve characteristic to estimate a second SOC based on the rate of change of the current-time product, allowing for a natural transition to a 100% full-charge display.
Adaptive SOC increase rate control
To prevent premature 100% SOC display when initial SOC estimates are significantly higher than the true value, the system detects when the first SOC reaches a threshold before the CV phase and decelerates its increase rate, ensuring it reaches 100% only upon true full charge.
PSCE for each feature:
CV-phase SOC estimation correction
Adaptive SOC increase rate control
Observations
- LLM has decomposed the patent into two different correction mechanisms. Feature 1 addresses the issue of SOC progression during the CV phase, while Feature 2 addresses the issue of SOC estimation becoming too high before entering the CV phase. This distinction cannot be clearly identified from the patent abstract alone.
- The P-S-E aspects are well explained and show clear correspondence with one another. At first glance, the C-aspect appears to extract the components described in the S-aspect. However, apart from the component names themselves, it does not capture the relationships among those components. It is therefore still uncertain whether the C-aspect can serve as a representative feature or description under specific terminology.
Interpretation
In this experiment, the LLM was used not only to identify technical features from the patent disclosure, but also to decompose each feature into five parts: one overall feature description and four separate technological aspects, namely Problem, Solution, Component, and Effect.
The overall feature description can cover the complete P-S-C-E content in a more concise form than describing the four aspects separately. However, the P-S-E aspects allow readers to quickly approach the same technical feature from different perspectives. Therefore, the aspect-based representation still appears necessary.
Upcoming Issues
- The necessity of the C-aspect requires further study.
Next Step
Map the technological aspects into a technology space so that they can be meaningfully compared across patents.