Overview
We’re excited to introduce SES 2.0, a major upgrade to our Semantic Essentiality Score (SES) feature.
This release replaces our legacy Latent Semantic Analysis (LSA) model with a context-aware semantic framework built on Word2Vec, TF-IDF, and Singular Value Decomposition (SVD).
The result: fewer false positives, less manual review, and greater confidence in SEP analysis.
Why did we make this change?
The legacy SES (LSA-based) system sometimes flagged patents as relevant simply because they shared vocabulary with a standard - even when describing different technologies.
SES 2.0 resolves this issue by understanding semantic context, not just word overlap - leading to higher accuracy and interpretability.
| Aspect | SES (Legacy) | SES 2.0 (Current) |
| Core Model | Latent Semantic Analysis (LSA) | Context-aware model using Word2Vec, TF-IDF, and SVD |
| Context Understanding | Based on word co-occurrence | Learns true contextual meaning of terms |
| Noise Handling | Sensitive to shared vocabulary (“noise”) | Removes “patent-ese” through SVD filtering |
| Accuracy | Moderate; keyword-based | High precision via semantic modeling |
| Scalability | Limited by computational constraints | Optimized for large patent pools |
| Manual Review Needed | High | Reduced by ~X% (validated value pending) |
| Confidence & Interpretability | Variable | Stable, explainable, and client-ready |
In short: SES 2.0 shifts from surface-level similarity to deep semantic understanding, fundamentally improving the reliability of patent-to-standard matching.
| Example | Legacy SES (LSA) | SES 2.0 (Word2Vec + TF-IDF + SVD) |
| Patent A – “Wireless communication system using adaptive modulation for 5G NR” | Flagged as relevant to 5G NR and LTE due to shared terms (“wireless,” “system,” “communication”) → False positive | Correctly identified as relevant only to 5G NR, recognizing contextual link between “adaptive modulation” and 5G standard → True positive |
| Patent B – “Method for transmitting control signals in IoT network” | Missed due to limited keyword overlap → False negative | Correctly matched to IoT-related standards, understanding “control signaling” semantics → True positive |
Outcome: SES 2.0 Significantly reduced false positives and improved true-positive accuracy in pilot validation tests.
Customer Benefits
- Higher Accuracy: Identify truly essential patents with fewer false positives.
- Less Manual Review: Cut review time by nearly half.
- Deeper Insights: Base decisions on semantic understanding, not surface terms.
- Scalability: Confidently process larger and more diverse patent datasets.
Summary
SES 2.0 marks a major leap forward in SEP analysis. By integrating context-aware semantic modeling, we enable a transition from keyword-based filtering to concept-level reasoning—helping users focus on what truly matters: patents that are genuinely essential to their technology standards.
If you have questions or want to see the new SES in action, feel free to reach out to our support team here or your account manager.