LinkedIn uses automated systems to recognize patterns on our messaging platform. These patterns help us make your professional communications more efficient and informed. For example, our systems do the following:
-
Look for specific strings of characters that indicate an emoji to render it as an image.
-
Look for text that indicates a web link (e.g. ends with ".com" or similar) to render a preview of the linked page.
-
When you start typing a name in the addressee field, attempt to anticipate who the recipient may be and provide auto-complete options.
-
Look for mentions of member or company names to add links to their profiles and LinkedIn Pages, respectively.
-
Check links shared in messages for malicious sites and look for blacklisted keywords to detect spam.
-
Suggest potentially relevant responses (e.g., messaging suggestions).
-
Look for viruses or other harmful code among attachments.
-
Look for certain characters (e.g. question mark at the end of a message) and context-specific keywords to propose relevant responses (smart replies).
-
Look for mentions of video chats, weekdays, or dates to see if our software can help you set up a meeting.
-
Detect and prevent likely harmful content.
Our systems sometimes use these insights together with contextual information. For instance:
-
If the sender is a recruiter, it's more likely that a message containing certain characters or words is about a job opportunity. Our systems may then propose messaging suggestions accordingly.
-
When a newly opened LinkedIn account sends frequent messages with certain words and links previously seen in other messages flagged as spam by members, our systems may detect these messages as spam.
Although recognizing text patterns in a message is often done at a very basic level (e.g. recognizing a string of characters that make an emoji), our systems sometimes use machine learning to develop and provide more complex functionality.
This essentially means that our analytical models and algorithms are improved over time based on members' usage. For instance, whether and how members use messaging suggestions or our messaging assistant helps refine the suggestions or other assistance and when they are presented. The refinement of our spam detection models as described above is another example of machine learning based on user feedback.
You can opt out of our automated scanning of your incoming private messages for content that violates our Professional Community Policies. Members in the European Union are opted-out by default due to privacy laws, and all other countries are opted-in by default. However, our automated systems will always provide basic defenses against spam, money fraud, viruses, malware, and malicious sites. The opt-out setting for harmful messages does not impact basic defenses.
Please note that with respect to our messaging assistant and similar bots, you have the choice to opt-in or not by adding, mentioning, or responding to such bots in a conversation.