WIPS PRISM : Detailed Features and User Guide
In our previous issue, we introduced PRISM as a deep-learning-based AI patent automatic classification solution. It revolutionizes analysis speed and accuracy by automating the technical classification stage, which typically accounts for 50-60% of the patent analysis workload.
In this issue, we will dive deeper into the specific features that make PRISM an essential tool for patent professionals.
 |
| gl.wipsprism.com> AI Training |
AI Training & Configuration
Establishing
the AI classification model through data learning.
This stage involves the initial setup where the AI learns from user-provided data. By uploading a population and performing manual classification for training, a complete classification model is generated.
Data Import: Upload up to 200,000 records via Excel or by syncing with WIPS Global "My Folders" (multiple folders supported). Additional data can be added even after a project is created.
Classification Support: Supports both manual and stepwise classification. It utilizes AI-generated summaries (sentence/phrase level) in addition to titles, abstracts, and claims.
Training & Automation: You can train the AI on a small subset of the population to automatically classify the remaining documents. Precision can be adjusted using similarity settings (30–90%).
 |
| gl.wipsprism.com> Noise Filter |
Noise Filtering via AI Training
Filtering
out irrelevant data before technical classification.
The goal is to create a "Noise Filter" by training the AI on "Valid Documents" so it can identify and remove dissimilar "Noise Documents."
The generated filter is applied during AI classification to automatically categorize documents as Valid (YES) or Noise (NO).
· By selecting the noise filter option and uploading valid
training cases, the AI generates a final filter optimized for your specific
project.
 |
| gl.wipsprism.com> AI Classification |
Real-World AI Classification
Applying the trained models to derive results (Technical Classification & Noise Filtering).
Automated Technical Classification: Apply a pre-created model to receive a classified list for each document. The system displays the Top 3 classification categories along with their Similarity Scores (%).
Adjustment & Expansion: Users can set and edit classification trees up to 3 levels deep. You can also integrate and re-classify new populations with existing data.
Recommended Workflow: To maximize data quality, we recommend a two-step process: first, use the Noise Filter to select valid patents, and second, apply the Detailed Technical Classification model.
Continuous Improvement: Once a model is created, it can continuously classify new patents. Performance can be further refined by feeding newly classified data back into the model for additional training.
 |
| gl.wipsprism.com> Clustering |
Clustering (Unsupervised Learning)
AI-driven
grouping of similar patents without prior training data.
This is ideal for exploring new technologies, establishing initial classification systems, or predicting a competitor’s new business directions.
Data Preparation: To avoid missing valid patents, it is best to collect data using broad search queries, even if they include some noise. Requires a minimum of 50 records.
Clustering Settings: Supports up to 3 levels of grouping. Like AI training, users can select the data scope (Title, Abstract, Claims, or AI Summaries).
Review & Model Conversion: After reviewing the clustering results, if the classification structure is satisfactory, it can be converted into an AI Classification Model (Supervised Learning) for future automation.
 |
| gl.wipsprism.com> Matrix |
Matrix
Strategic
insight through Objective (Why) vs. Solution (How) analysis.
The OS Matrix maps the same dataset across two axes—Object and Solution—to analyze distribution patterns and identify problem structures, technical trends, and opportunity areas.
White Space Analysis: By setting AI classification or clustering results on the X and Y axes, you can see where data is concentrated or sparse. A "White Space" (an object with no corresponding solution) indicates an untried technical area.
Portfolio Optimization: This method allows for the simultaneous identification of competitive structures and expansion potential.
Targeted Search: By running a matrix with data scoped to "Solution" and "Effect" fields, users can find specific technologies that achieve a desired effect or identify alternative effects for a single solution.
 |
| gl.wipsprism.com> Chart |
Multi-Perspective Charts
Visualizing
classification results through various lenses.
PRISM provides interactive charts to visualize results by year, country, assignee, and technical category.
Enhanced Accuracy: Data cleaning features for assignee names improve the precision of the analysis.
Interactive Filtering: Selecting a specific tech field or date range triggers real-time updates across all charts, providing a customized and intuitive analysis environment.
We have taken a detailed look at the core features of WIPS PRISM. At its heart, PRISM is about Automated Classification through AI Learning. Whether you are creating a 'Personal Assistant" by teaching the AI your manual classification logic (Supervised Learning) or letting the AI discover strategies within massive datasets (Clustering/Unsupervised Learning), WIPS PRISM is here to transform your patent workflow efficiency.
Boost your patent intelligence with WIPS PRISM today!
