Working with Concepts

This guide explains how you can use concepts to both analyze and filter by the brands, products, topics and well known people that are mentioned in content.

Concepts and types

Concepts are entities stored with their name and type. They are extracted from the articles that are shared or engaged with on LinkedIn.

There is a vast number of concepts that can be recognized. To help you work with concepts they are organised into the following types:

Type Description Examples
company A company mentioned in the content. IBM Starbucks Siemens
product

A product mentioned in the content.

Note that the definition of a product is broad, including software products, publications, services and television programs.

Skype Bloomberg_Businessweek IOS
city A city mentioned in the content. Chicago New_York_City Sydney
country A country mentioned in the content. China Italy United_States
location A location mentioned in the content aside from a city or country. For example a US state. Asia Yosemite_National_Park Minnesota
politician A politician mentioned in the content. J._Edgar_Hoover Winston_Churchill Benjamin_Franklin
entertainer An artist, entertainer, musician, or performer mentioned in the content. John_Milton Michael_Jackson Tricky_Stewart
sportsperson An athlete, coach or other sportsperson mentioned in the content. Tom_Brady Viswanathan_Anand Dale_Earnhardt
person A well-known person mentioned in the content (not classified as one of the types above). Richard_Branson Liza_Minnelli Blaise_Pascal
education An educational institution or organization, such as a university or college. Stanford_University University_of_London Syracuse_University
government agency A government agency or department mentioned in the content. United_States_Department_of_the_Navy NASA Food_and_Drug_Administration
organization An organization mentioned in the content (not classified as one of the types above). International_Organization_for_Standardization World_Economic_Forum Institution_of_Civil_Engineers
misc A concept not classified as one of the types above, such as an idea, theory, medical term. Spanish_language Dog Cardiovascular_disease

note icon


Currently concepts are only extracted from content written in English.

Analysis and filtering using concepts

There are three targets made available for filtering and analyzing concepts:

  • li.all.concepts.names - the names of concepts that have been extracted.
  • li.all.concepts.types - the types of concepts that have been extracted.
  • li.all.concepts.type_names - the name and type of concepts that have been extracted.

Each can be used for analysis or as part of query filters.

Analyzing concepts

Concepts are great for helping you to explore content that is popular on LinkedIn. They take you beyond trying to simply digest page titles and urls, and give you a view into the content of popular pages.

For most use cases the li.all.concepts.type_names target is most useful for analysis as it includes both the name and type of the concepts.

For example, the following analysis task analyzes the top concepts mentioned in content relating to machine learning:

{
  "subscription_id": "e9dde04774540ac119c2317a4d15a8b3a1350937",
  "name": "Machine learning concepts",
  "type": "analysis",
  "parameters": {
    "filter": "li.all.articles.title contains_any \"ai, artifical intelligence, ml, machine learning\"",
    "parameters": {
      "analysis_type": "freqDist",
      "parameters": {
        "target": "li.all.concepts.type_names",
        "threshold": 50
      }
    }
  }
}

Running the analysis gives results such as the following:

Type|Name Type Concept name Unique authors Interactions
unknown|Deep_learning unknown Deep_learning 62,100 95,700
person|Elon_Musk person Elon_Musk 50,000 59,400
organization|World_Economic_Forum organization World_Economic_Forum 43,500 50,200
education|Massachusetts_Institute_of_Technology education Massachusetts_Institute_of_Technology 24,300 37,200

The result tells us that 24,300 members engaged with content that mentioned MIT in the analysis period.

Filtering by concept mentions

You can also use concepts in your query filters. This enables you to filter not just on titles and summaries of content, but also on concepts mentioned in the body of content.

Taking the filter used in the example above for machine learning:

li.all.articles.title contains_any "ai, artifical intelligence, ml, machine learning"

This filter could be extended to look for concepts in the body of content:

li.all.articles.title contains_any "ai, artifical intelligence, ml, machine learning"
OR li.all.concepts.type_names in "unkown|Deep_learning,unknown|Artificial_intelligence,unknown|Machine_learning"

This broadens the filter to include more content and is a great technique for improving your filters.

Filtering by concept type

Note that filtering by a concept type will not only return concepts of that type.

Take this example task:

{
  "subscription_id": "e9dde04774540ac119c2317a4d15a8b3a1350937",
  "name": "Machine learning concepts",
  "type": "analysis",
  "parameters": {
    "filter": "li.all.concepts.types == \"company\"",
    "parameters": {
      "analysis_type": "freqDist",
      "parameters": {
        "target": "li.all.concepts.type_names",
        "threshold": 50
      }
    }
  }
}

This request is asking to analyze all concepts in content where at least one concept of type company is mentioned. It is not asking to analyze only concepts of type company.

Therefore the top entities that are mentioned in content alongside at least one company concept will be returned.

If you want to obtain a list of concepts of a certain type then:

  • use the li.all.concepts.type_names target for your analysis
  • in post-processing split the type and names like in the table of results above
  • filter to only the results that match the type you need