TABLE OF CONTENTS

• Introduction to Sophia Search
• Semantic search only 
  • Results Review
• The Prefilters - Advanced and Premium Users
• Features Mode - Advanced and Premium User
  • Features extraction and selection
  • Understanding Credits in Features Mode
  • Results Review in Features Mode
  • One Click Prior Art Report
• Limitations and Considerations

Introduction to Sophia Search

Sophia Search is an AI-powered patent semantic search that uses natural language and advanced neural embeddings. 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 engine aims to:

• Provide better results: Our new embedding-based technology ensures significantly higher recall and precision.
• Explainability:  "Features" mode provides actionable, exportable information on the relevancy of each result.

This AI-powered search tool, with the subscription level Advanced or Premium, offers also:

• Features Mode: Extract features from your input text, and rely on AI to read patents and pinpoint which elements/features have been already disclosed into existing patents.
  
• Pre-filtering: with the ability to pre-filter using the Advanced search form, less time is spent sifting through irrelevant information.

For full details on how your information is processed, with or without third-party intervention, please refer to the disclaimer document: AI tools in Orbit Intelligence: Usage disclaimer

Semantic search only 

Sophia Search is found under the Menu tab, among the different searches in Orbit. 

Go to the main interface and type your text in the search box to access the semantic mode. This usage as described here is unlimited. 

A character counter appears in the bottom-right corner of the text area to help you stay within the optimal length.

• Optimal text size: between 500 and 3,000 characters.
• Minimum requirement: text must be at least 100 characters.
• Beyond 3,000 characters your remaining text will be truncated and ignored

Note: Non-English text is sent to a third-party service for automated translation. We kindly advise you to input English text, and preferably the most self-explicit about the invention (like an executive abstract).

Before running your search, set the maximum number of results at the bottom of the page. The default setting is 50 results.

Results Review

The hitlist displays the results along with a Semantic Match relevance column. Each label represents a range of relevance scores, indicating how closely the result matches your query:

• Perfect: 99-100% match
• Excellent: 97-98.99% match
• Good: 96-96.99% match 
• Moderate: 95-95.99% match
• Weak: 92-94.99% match
• 0-91.99% is considered as no match

Percentages here disclosed are inherent to embeddings technology, and cannot be comparable with the Relevancy score provided by other kind of search in Orbit.

Results are sorted by their matching score, calculated from embeddings generated from the user's input, and compared with the families' embeddings.

Explanation of a label is available by hovering this one, and view the numerical relevance score. 

Prefilters usage for Advanced and Premium Users

Prefilters let you narrow down your search before running it, to improve relevance and efficiency. They are set to limit the initial scope of the semantic search, whereas traditional filters will only filter the output of the semantic search

This opens the Advanced Search form, where you can refine your search by filtering by date range, jurisdiction, or classification. You can also use keyword suggestions or apply the corporate tree feature to target company-related results.

Once activated, the active Prefilters appear above the input field. You can remove them quickly by clicking the 'X' icon. 

Please note that when Prefilters are applied, the maximum number of results is limited to 50. Retrieving 1000 results is no more supported once activated.

Features Mode extraction - Advanced and Premium User

Sophia Search can extract the 'Features' described in your input. A feature refers to a distinct technical characteristic or component that contributes to the innovation, or a differentiation of the subject matter in the text. 

These Features act like clues, showing the connection between your query and the document. They can also help to fine-tune your results by simply deselecting any Features that aren’t relevant to your search.

⚠️ Features mode uses monthly credits (see section below). 1 credit consumed once you have requested the hitlist.

Features extraction and selection

Click the Extract features button at the bottom of the search window. 

A list of detected Features appears on the right side of the window. By default, all generated Features are selected. Review the list and deselect any irrelevant Features to fine-tune your search results. 

Possibly you can select 3 or 4 most relevant or technical distinctive features. The more features selected, the more processing and time it will take.

You can remove all extracted features at any time by clicking on Clear from the top of the list. This will switch back the current search to the standard Semantic mode (no features).

Understanding Credits in Features Mode

• Advanced and Premium users have a monthly credit allocation for using Features mode in Sophia Search. No credit allowed for Essential users.
• Each user receives a fixed number of credits per month, automatically refilled on the 1st of each month. Unused credits do not roll over to the next month.
• Credit allocation by license type:
  • Advanced: 20 credits
  • Premium: 40 credits
• One credit is consumed each time a search in Features mode is successfully executed : Search is launched with at least one feature selected and hitlist reached. No credit is consumed by just extracting features.
• The credit system does not applied to Prefilters usage or to Semantic Mode (No Features)

The number of remaining credits is displayed at the bottom right of the search interface.

Results Review in Features Mode

After a search with Features mode activated, the Features Match column is added to your hitlist.

Keep in mind that in Features mode, Features match ranking takes priority over Semantic Match ranking. In other words, the second (for instance) 'semantically' relevant result may appear indeed at the top of your hitlist as this result is disclosing more features than the most 'semantically' relevant result.

When you are hovering over a feature match, a tooltip displays its title and description for quick reference.

For more insights on a specific result, you can activate the tab Sophia Search in the right-handed pane from the hitlist, or you need to go to the Family View by clicking on the patent family of interest.

✅ Tip: Don’t forget to add the Sophia Search tab to your display for full functionality.

The Sophia Search tab provides detailed information about each feature and the snippets that justify its status.

For each feature, you will find:

• Number, title, and description for easy reference where the disclosure is deemed to be
• Status like disclosed or partially disclosed, displayed clearly after the feature title.

Under each feature disclosed, up to 3 snippets will be shown to substantiate this point:

• You’ll see a snippet number, the Publication stage, and the claim or paragraph number where the snippet was extracted from. Nota: FAMPAT, the publication stage shown is the one used for embedding computation.
• To copy a snippet, hover over it and click the “Copy to clipboard” button that appears to the top right of the snippet.
  

Additional tools in this tab:

• The menu on the right visually represents the status of all features and serves as a navigation panel to move between features easily.
• You can export all the content of this tab, by using the XLS button at the top right.
  

One Click report: your prior art report

From the hitlist, when you select up to 10 results, you can instantly generate a prior art report, ready-to-use Excel file. 

This export/report is available only when you are in review results from a Features mode search.

This Excel file always includes 3 sheets:

1. Info: When, what and which features have been used to generate this export
2. Table_Characteristics: For each feature selected, a table will show where a possible disclosure is currently available, among the different publications stages present your results
3. Table_Documents: a basic export where you will find commonly exported datalike title, abstract and permalinks from your seleted results in hitlist.

Limitations and Considerations

Please find in this section, frequent questions asked about Sophia Search, and other practical points on which we would like to draw your attention.

Execution Time
Searching with Features mode takes significantly longer than in Semantic mode only, and applying Prefilters may further increase execution time (depending of your keyword complexity if any).

Here some average execution times under normal conditions:

• Semantic mode (no Prefilters): 1–5 seconds
• Semantic mode (with Prefilters): 10–20 seconds
• Features mode (5 Features, 50 results max): 1–2 minutes

Input Text Size
Only the first 3,000 characters of the input text are used for the search. A character counter is available on the Sophia Search page.

Representative Member
One embedding is computed per FAMPAT family using a specific algorithm to select the “computational” member. This may differ from your representative member chosen as defined in your settings. In other words, search and feature-to-text matching may not align with the user’s representative member.

Combining Queries
When combining a Sophia Search query in Features mode with another query from the Search History, Feature information is lost, and the final result of your strategy will not display any content in the Sophia Search tab.