Getting started with the free-trial version of Patent Monitor

Artificial intelligence for patent classification made easy

Start

  1. Open Patent Monitor and log-in with your credentials

  1. Choose project (there’s only one – if you want to create more, please ask here)

You can access the detailed documentation at any time by clicking on the question mark in the toolbar on the left.

Importing Patent Documents

3. Click “Connector Management” on the project overview page

screen-shot-2018-11-06-at-19.47.24

4. Choose “IfiClaimsConnector” and click on “Create connector”

IFI CLAIMS provides patent data from over 90 countries. You can access more than 119m patent documents! In the free trial version only this one connector is supported – in the full version there are more!

5. Import patent documents that match your individual query

Add a name for the resulting patent collection and at least one of the following search criteria

  • a query (see searching guide here)
  • an applicant or assignee
  • publication dates (format: yyyymmdd)

Typical queries are (in any combination):

Phrases: "solar energy"
Boolean-OR: "solar energy" OR "wind power"
Proximity: "solar generation"~5
International classification: ic:A61K

Alternatively, you can enter a list of patent ids in the field “UCID List”. Please use the IFI Patent Number Converter to convert your specific format to IFI numbers and use the “Copy to Clipboard” functionality to enter the values.

You can import additional documents at any time by defining additional connectors. However, the absolute number of documents is reduced in the free-trial version.

6. Start importing

Click on  – the patent documents are now imported into your application directly from the patent database. This process takes place in the background, which means that you can continue with the next steps.

You can create and run many queries in parallel!

Creating the classification schema

7. Click on “Label system” on the project overview page to setup the categories

bildschirmfoto-2018-10-26-um-14.51.56

8. Add your individual target categories by clicking on “Create new label”

Training the system

9. Click on „Classification“ on the project overview page

screen-shot-2018-11-06-at-19.48.10

10. Select training data for the assignment of category labels

You will access the search page with some classification filters on the left panel:

  • Label: An overview on the assigned labels for the current search result
  • Confidence: Either “high”, “low” or empty
  • Status: Either “Unclassified”, “Approved” (i.e. manually labelled) or “Auto-classified”

Furthermore, there are filters on the patents metadata:

  • Applicant: Aggregation of patent applicants
  • IPC4: international patent classification on a 4-character-level
  • Source: the name of your collection(s)

Use the search functionalities by using filters and keywords to identify relevant patents for a specific category. There are now three options to label the resulting document set:

  • Click on the blue button “Label document(s)” without selecting particular documents: The assignment of a category to choose will be done on whole result set.
  • Mark the check box(es) left to the title information: The subsequent assignment via the blue button “Label document(s)” refers to the marked documents only.
  • Click on the small tick-mark below the title information of the document. The assignment refers only to that particular document.

11. Choose appropriate category label and confirm with “Save”

It is recommended to start with the provision of at least 10 training examples for each category. Additional assignments can be made at any time later in the classification process.

12. Getting an overview on the training documents

The facets on the left show the distribution of the training material. For the “Status” filter, click on the value “Approved” and afterwards open the filter “Label”. You will see the category labels together with their frequency of occurrence in the corpus.

13. Training the classifier using Machine Learning

Switch to “Classification Configuration” on the project overview page and click on the mortarboard icon . The “State” of the classifier will be changed to “TRAINING”. Click on  to update the current state. When “State” switches to “IDLE” and “Model” to “READY” the classification model has been trained and is ready to use.

New documents uploaded to the system must internally be pre-processed once before training can take place. Therefore, clicking the train button will consume more time after the initial setup. It will be much faster when re-training!

14. Evaluation of the classifier

Clicking on will show statistics of the model trained. It depicts an overview on the quality of the predictions in general (top) and for each category (bottom) by taking the measures in terms of precision, recall, f1-score and accuracy (common metrics in machine learning).

The values determined are very helpful in assessing the quality of the predictions. They also help in identifying categories that are underrepresented in the training set or to uncover inconsistencies in the training set and/or category system. Modify the trainings set to increase the values.

Automatic classification of patents

15. Switch to the “Classification” page on the project overview

The newly generated model can be used to automatically classify all previously unexamined documents. There are 3 options:

  • Click on the blue button “Classify document(s)” without selecting particular documents: The assignment of a category will be done on all documents with status “Unclassified”.
  • Mark the check box(es) left to the title information: The subsequent classification via the blue button “Classify document(s)” refers to the marked documents only.
  • Click on the “classify” icon below the title information of the document. The assignment refers only to that particular document.

The first option runs the automatic classification in the backend after some initial processing. Click on the filter “Status” to get the current progress of the classification process.

A confidence value (between 0 and 1) is assigned to each classified document, together with the predicted category. The label will appear in green color when the confidence value is high. If it’s lower, the color is red. Corresponding filters are available on the left panel of the application.

Click on the “Status” filter “Autoclassified”, select a category by using the “Label” filter and choose “Sort by confidence (descending)” and you will receive the most relevant patent documents for that particular category. You can combine these filters with additional ones (e.g. applicants, IPCs keywords, etc.).

16. Improving the classifier through active learning

Sorting the current selection by confidence “ascending” will show documents with predictions that come with a very low confidence. It’s worth to make corrections (or confirmations) to these particular predictions, because the system learns best from those examples where it is uncertain.

Simply click the check mark left or the cross right from the category label in order to confirm or remove the prediction. After some examples the system should be trained again (13), and the resulting performance metrics will increase (14). Use the resulting model to re-classify your patent collection (15).

CONGRATULATION !

You have just used artificial intelligence for your first automatic patent classification project