The global innovation landscape is characterized by an ever-growing mountain of patents. While these documents represent a vast repository of technical knowledge and inventive potential, a critical chasm often exists between this potential and its realization in the market. Patents frequently remain as isolated islands of invention rather than being integrated into commercially viable solutions. In my research, I address this disconnect by developing a methodology that shifts the focus from the patent document itself to the application context. I propose that the key to unlocking patent value lies in identifying combinations of patents—portfolios—that are specifically tailored to meet the concrete, often latent, demands of real-world application scenarios. This perspective moves beyond purely technical linkage to a more holistic, demand-driven view of innovation assembly. This article details my framework for this task and demonstrates its application through a compelling case study in the field of humanoid robot technology.
The core of my approach is the integration of diverse data sources to construct a rich, multi-faceted understanding of both technological capability and market need. The process begins with gathering structured and unstructured data. I collect patent documents to represent codified technological solutions. To enrich this view with forward-looking and application-oriented knowledge, I also gather academic literature. Furthermore, to ground the analysis in market reality, I scrape commercial reports for industry insights and user reviews from platforms like e-commerce sites to capture authentic end-user feedback and expectations. A summary of the processed data for the humanoid robot case is presented below.
| Data Type | Source | Processed Volume |
|---|---|---|
| Scientific Papers | Web of Science | 11,659 |
| Technical Patents | incoPat Database | 14,910 |
| Commercial Reports | Open Web Platforms | 1,812 |
| Product Reviews | E-commerce Platforms | 36,892 |
The methodological framework proceeds in three interconnected stages: (1) Mining scenario-specific technical requirements, (2) Constructing a scenario-aware Technology-Efficacy Knowledge Graph, and (3) Matching and scoring patent combinations based on these scenarios.
Stage 1: Mining Scenario-Specific Technical Requirements
Not all user demands carry equal weight for guiding impactful innovation. I employ the Kano model to categorize requirements extracted from text, focusing particularly on *Expectation* and *Excitement* type demands, as these drive market satisfaction and differentiation. To extract these, I first use the LDA topic model on commercial reports to identify broad application scenarios for humanoid robots. For instance, distinct topics coalesce around keywords like “nursing,” “rehabilitation,” “housework,” and “assembly,” pointing to scenarios like Medical Rehabilitation, Home Service, and Industrial Automation.
Next, from user reviews, I extract sentences containing performance or function keywords. Using the KeyBERT model, I distill these sentences into concise requirement keywords. These keywords are then organized according to the previously identified LDA-based scenarios and classified by their Kano type. This process yields a structured map of what users fundamentally need, expect, and would be delighted by in each specific context for a humanoid robot.
| Scenario | Expectation Demands (Sample Keywords) | Excitement Demands (Sample Keywords) |
|---|---|---|
| Industrial Automation | precision, speed, durability, simplicity, cost | autonomous learning, friendly interaction, multifunction, remote control |
| Medical Rehabilitation | safety, accuracy, personalization, stability | emotion recognition, intelligent adaptation, multi-language |
| Home Service | ease of use, low cost, easy maintenance, safety | educational assistance, entertainment, smart learning |
| Public Service | multi-language, accurate information, emergency response | emotional interaction, smart navigation, personalized service |
Stage 2: Building a Scenario-Aware Technology-Efficacy Knowledge Graph
To effectively match patents to scenarios, I need a structured representation of technological knowledge that includes *how* a technology works and *what benefit* it provides. I construct a Technology-Efficacy Knowledge Graph (TEKG) from patent and literature data. Using a BERT-BiLSTM-CRF model, I extract key entities from patent texts such as `Technical_Solution`, `Application_Efficacy`, `Key_Component`, and crucially, `Application_Scenario`. The UIE (Universal Information Extraction) model is then used to establish relationships between these entities, for example, linking a patent to its claimed efficacy and its intended application scenario.
The performance of the entity extraction for the humanoid robot domain is summarized below. Structured entities like patent numbers are extracted with high precision, while more conceptual entities like application value show satisfactory performance for subsequent analysis.
| Extracted Entity | Precision | Recall | F1-Score |
|---|---|---|---|
| Patent Number | 98.69% | 96.32% | 97.44% |
| Technical Solution | 73.76% | 72.00% | 72.23% |
| Application Efficacy | 76.54% | 75.66% | 76.00% |
| Application Scenario | 75.88% | 75.31% | 75.57% |
This graph is “scenario-aware” because the `Application_Scenario` node acts as a critical hub. It allows me to query all technologies and their associated efficacies that are relevant to a specific context, such as “Emergency Response” for a humanoid robot in public service.
Stage 3: Matching & Scoring Patent Combinations for a Scenario
With a clear set of scenario demands and a knowledge graph of technological capabilities, the final step is to identify promising combinations (portfolios) of patents. I evaluate potential pairs of patents based on three critical dimensions:
- Technological Similarity ($S_T$): Ensures the patents are compatible and can be integrated within a common technical framework. It is calculated from the textual similarity of their technical descriptions and keywords.
$$ S_T = \alpha \cdot S_{text}(A_{text}, B_{text}) + \beta \cdot S_{kw}(A_{kw}, B_{kw}) $$ - Technological Complementarity ($C_T$): Measures how well the patents’ combined efficacies cover the required functions for the scenario. It is derived from the Jaccard similarity of their efficacy sets and the overlap of their technical solutions.
$$ C_T = \gamma \cdot C_E + \delta \cdot C_{AP} $$
$$ C_E = \frac{|c_{AB}(w_A \cap w_B)|}{|w_A \cup w_B|}, \quad C_{AP} = \frac{|s_A \cap s_B|}{|s_A \cup s_B|} $$ - Technological Uniqueness ($U_T$): Assesses the novelty and potential competitive advantage offered by the combination, based on the rarity of its IPC code combinations.
$$ U_T = \frac{n_{NCP}}{n_{CP}} $$
For a given target scenario and its associated demand keywords (e.g., “Emergency Response” requiring “real-time dispatch,” “hazard perception,” “all-terrain mobility”), I extract relevant patent nodes from the TEKG. I then compute a composite matching score for patent pairs by normalizing and summing their scores on $S_T$, $C_T$, and $U_T$. Pairs exceeding a high threshold (e.g., 0.85, validated by domain experts) are identified as high-potential humanoid robot patent portfolios for that scenario.
Empirical Application: Identifying Portfolios for a “Humanoid Robot” in Public Service
Applying this framework to the humanoid robot domain, I identified six primary application scenarios. Focusing on the “Public Service” scenario and its “Emergency Response” sub-demand, the matching algorithm evaluated numerous patent pairs. A snippet of the normalized scoring matrix is shown below, with high-scoring pairs highlighted.
| Patent | P5 | P10 | P12 | P26 | P62 | P82 | P113 | P177 | P279 |
|---|---|---|---|---|---|---|---|---|---|
| P5 | – | 0.131 | 0.222 | 0.104 | 0.016 | 0.719 | 0.604 | 0.276 | 0.979 |
| P10 | 0.131 | – | 0.864 | 0.422 | 0.402 | 0.808 | 0.272 | 0.240 | 0.772 |
| P12 | 0.222 | 0.864 | – | 0.664 | 0.450 | 0.033 | 0.680 | 0.630 | 0.620 |
| P62 | 0.016 | 0.402 | 0.450 | 0.477 | – | 0.811 | 0.142 | 0.973 | 0.260 |
Three high-potential portfolios were derived for the “Emergency Response” humanoid robot:
| Portfolio | Patents & Focus | Combined Value Proposition |
|---|---|---|
| 1 | P10 (Bipedal Jump Control) + P12 (Firefighting Robot Platform) | Creates a humanoid robot with exceptional obstacle-crossing mobility for firefighting, enabling rapid advancement in complex disaster sites where traditional robots fail. |
| 2 | P5 (Explosion-Proof Communication System) + P279 (Quadruped Emergency Handling) | Forms a collaborative system where a rugged, explosion-proof communication hub coordinates agile quadruped humanoid robots for reconnaissance and intervention in hazardous industrial accidents. |
| 3 | P62 (Pipe Inspection Robot) + P177 (Multi-Sensor Fusion for Navigation) | Enables an intelligent humanoid robot for underground utility emergency response, autonomously detecting leaks, cracks, or blockages using fused sensor data and executing precise repair actions. |
This empirical demonstration validates the methodology’s effectiveness. By starting from the concrete scenario need of “Emergency Response,” the system did not just find related patents; it identified synergistic *combinations* that together create a solution greater than the sum of its parts. For instance, Portfolio #2 effectively links secure communication (a systemic need) with agile physical intervention (a functional need), addressing the scenario’s requirements holistically.
Conclusion and Implications
In conclusion, my research presents a structured, scenario-semantic approach to bridge the gap between patent warehouses and market applications. By systematically integrating mined user demands with a semantically constructed Technology-Efficacy Knowledge Graph, I move beyond traditional patent analysis focused on citation or keyword co-occurrence. The tripartite scoring mechanism—balancing Similarity, Complementarity, and Uniqueness—provides a robust filter for identifying patent combinations with high commercialization potential tailored to specific contexts. The application to the humanoid robot field successfully identified actionable portfolios, proving the method’s practical utility. This approach offers a powerful tool for innovation managers, R&D strategists, and policymakers seeking to accelerate the translation of patented inventions into impactful, market-ready technological solutions for the humanoid robot industry and beyond. Future work will extend this framework to other domains and explore multi-modal data analysis, such as extracting information from patent figures, to further enrich the scenario-technology matching process.