TABLE OF CONTENTS
- Introduction to Sophia Search
- Sophia Search for everyone: new semantic search with local embeddings
- Features extractions mode, disclosure and pre-filter
Introduction to Sophia Search
Sophia search combines a new generation of natural language search based on embeddings, with an AI-powered extraction of features disclosed in user's input and patent fulltext.
This new search aims to:
- provide better results: Our new embedding-based technology ensures significantly higher recall and precision.
- increase efficiency: With more accurate results and the ability to pre-filter using the Advanced search form, less time is spent sifting through irrelevant information.
- Explainability: "Features" mode provides actionable, exportable information on the relevancy of each result.
Sophia Search for everyone: new semantic search with local embeddings
Sophia search is available from the same place as the previous Semantic search - from the left-hand pane Menu. Here is the main interface:
Enter your text - a character counter on the bottom-right corner of the text area is available and allows you to target the optimal text size for search (between 500 and 3,000 characters - no search allowed below 100 characters).
You can put your mouse over the little i at the bottom right, and read more about the optimal text length, where we process your input and possible third-party translation for non-English text:
Before running your search, please define a results limit
Once you have ran the search, the well-known hitlist is displayed
Results are sorted by the matching score, based on embeddings created from the user's input, and compared with the families' embeddings.
[BETA]: 100% score is a perfect match between input text and patent - almost impossible as we have a limit of 3,000 characters in input. A score above 98% is outstanding. Between 98% and 95% is excellent. This scale is purely informative, and may evolve with the time to better reflect differences between families on a larger range of percentage.
Features extractions mode, disclosure and pre-filter
Sophia search can extract the "features" described in your input and compare them with up to 100 patent families - the most relevant of the search here above.
You will mainly use the 3 following buttons:
Features extraction and selection
When you request to extract the features, Orbit Intelligence sends your text to a third-party AI provider (please refer to our disclaimer here) , and in return, Orbit displays up to 5 features detected as:
We kindly advise to review each feature extracted, and to tick the most novel or inventive elements.
Prefilters
Sophia search allows you to add some criteria before running the search. You can recognise the advanced search form, and take advantage of all the possibilities offered by our advanced search: suggestions of keywords, classifications, corporate tree, etc....
This will narrow down the search results, like within an assignee portfolio, to identify a probable prior-art already publicly disclosed.
Hitlist features disclosure
[BETA] The following display is only available with the "abstract view" currently. Please change your setting as described in this article.
When your search includes features extraction, the hitlist will be slightly modified by the add of "blue balls" under each result, and a new dedicated tab Features:
When you put your mouse over each blue ball, a tooltip appears and describes if the related feature is disclosed, partially or not disclosed, by this patent family.
To make your in-depth review of each feature extracted, please refer to our Features tab. If you do not have this tab, please make sure to add it to your right-hand pane selection, and you will see a hitlist generated from a Sophia search.
For each partially or completely disclosed feature, this tab will show you up to 3 snippets / extracts from the English text of this family, with a reference to the paragraph or claim.
One click report from hitlist
Was this article helpful?
That’s Great!
Thank you for your feedback
Sorry! We couldn't be helpful
Thank you for your feedback
Feedback sent
We appreciate your effort and will try to fix the article