Import and Analyze Knowledge Graph Results on a Large Scale

If analyzing SERPs is the first step in understanding your rankings on search engines, then analyzing the knowledge graph can be thought of as step zero.

SERP positions for a certain keyword show how each page is ranked in comparison to all other eligible pages. Knowledge graph scores on the other hand, show the ranks of the different meanings that a word can take for Google (a person, a city, a brand, etc.).


From Google’s documentation: This API is not suitable for use as a production-critical service. Your product should not form a critical dependence on this API.

It’s not clear whether this is from a technical reliability or a content correctness point of view, but it is what the docs mention. So please keep this in mind when using it.

Account Setup

In order to be able to send requests, you will need to create a project, set up billing, and activate the knowledge graph API for your project. You will then need to create credentials (API Key). Once you have that, you can use it as your key parameter when running requests, as shown below.

How to use Google’s Knowledge Graph API

What is “google”? Is it a search engine, a company, a brand, a very large number? What else is it?

And if it is all of those things, what is the relative ranking of each? What is the source of the information, its URL, images (if any)?

>>> google =  knowledge_graph(key=key, query='google')
>>> google
     query  resultScore                                    result.@type                  result.description
0   google       203191        ['Corporation', 'Organization', 'Thing']                  Technology company         Google
1   google        49462                            ['WebSite', 'Thing']                              Google         Search
2   google        19142                            ['WebSite', 'Thing']                                 nan          Gmail
3   google        13251                   ['Brand', 'WebSite', 'Thing']                             Website    Google Maps
4   google         7549     ['WebSite', 'SoftwareApplication', 'Thing']                             Website   Google Drive
5   google         6853                            ['WebSite', 'Thing']                             Website    Google Play
6   google         6543                ['SoftwareApplication', 'Thing']                         Web browser  Google Chrome
7   google         4312        ['Corporation', 'Organization', 'Thing']  Multinational conglomerate company  Alphabet Inc.
8   google         3395                ['SoftwareApplication', 'Thing']                                 nan Google Account
9   google         1306                                       ['Thing']                                 nan         Google

>>> google.columns
Index(['query', 'resultScore', '@type', 'result.@type', 'result.description',
   'result.image.contentUrl', 'result.image.url',
   'result.detailedDescription.url', 'result.detailedDescription.license',
   'result.url', '', 'result.@id', 'query_time'],

The above table is a sample response from the knowledge_graph() function. Many more columns are available as you can see in the second line above. We can see that “google” is a company, with a result score of 203,191 and it is a search engine/website with a result score of 49,462. It is then understood as an email application, a mapping application, and so on, as you can see in the column.

You can also see that we get the types under which this result falls, in the result.@type column. Multiple types show the type inheritance, and as you can also see, everything is a “Thing”. This is the top element in the type hierarchy under which everything belongs.

Like serp_goog() and serp_youtube(), this funcion works in the same manner, creating, sending, and aggregating the product of the arguments passed to it.

For example if you run

>>> knowledge_graph(key=key, query=['google', 'bing'], languages=['en', 'fr', 'de'])

The function will send 2 (queries) x 3 languages = 6 requests.

(google, en), (google, fr), (google, de) , (bing, en), (bing, fr), (bing, de)

This is actually the main value of having this function, because you usually want a large sample to evaluate certain keywords across languages or types.

Let’s check what “seo” and “search engine optimization” mean in different languages.

>>> seo = knowledge_graph(key=key, query=['seo', 'search engine optimization'], languages=['en', 'es', 'de'])
>>> seo
        query                           languages       resultScore                      result.@type                                          result.description
0       search engine optimization      de                     3587         Suchmaschinenoptimierung             ['Thing']                                         nan
1       search engine optimization      de                      321         Lokale Suchmaschinenoptimierung      ['Thing']                                         nan
2       search engine optimization      de                      252         Suchmaschinenmarketing               ['Thing']                                         nan
4       search engine optimization      en                    71756         Search engine optimization           ['Thing']                                         nan
5       search engine optimization      en                     5056         Search engine marketing              ['Thing']                                         nan
6       search engine optimization      en                      576         SEOP, Inc.                           ['Organization', 'Corporation', 'Thing']          Company
13      seo                             de                     3313         Seoul                                ['AdministrativeArea', 'Thing', 'City', 'Place']  Hauptstadt von Südkorea
14      seo                             de                     1509         Seo Yea-ji                           ['Thing', 'Person']                               Schauspielerin
15      seo                             de                      584         Suchmaschinenoptimierung             ['Thing']                                         nan
33      seo                             es                     1509         Seo Ye-ji                            ['Person', 'Thing']                               Actriz
34      seo                             es                      584         Posicionamiento en buscadores        ['Thing']                                         nan
35      seo                             es                      316         Jin                                  ['Person', 'Thing']                               Cantante
53      seo                             en                     8760         Search engine optimization           ['Thing']                                         nan
54      seo                             en                     3313         Seoul                                ['AdministrativeArea', 'Thing', 'City', 'Place']  Capital of South Korea
55      seo                             en                     1435         Sulli                                ['Thing', 'Person']                               South Korean actress
>>> seo.columns
Index(['query', 'languages', 'resultScore', '@type', '',
       'result.@type', 'result.@id', 'result.image.contentUrl',
       'result.image.url', 'result.detailedDescription.license',
       'result.detailedDescription.articleBody', 'result.description',
       'result.url', 'query_time'],

It’s interesting to see how the same word can mean different things in different contexts.

knowledge_graph(key, query=None, ids=None, languages=None, types=None, prefix=None, limit=None)[source]

Query Google’s Knowledge Graph with any combination of parameters.

Note that Google’s documentation states that “This API is not suitable for use as a production-critical service.” So please keep this in mind.

  • key (string) – Your Google developer key.

  • query (string) – A literal string to search for in the Knowledge Graph.

  • ids (string) – A list of entity IDs to search for in the Knowledge Graph.

  • languages (string) – The list of language codes (defined in ISO 639) to run the query with, for instance en.

  • types (string) – Restricts returned entities to those of the specified types. For example, you can specify Person (as defined in to restrict the results to entities representing people. If multiple types are specified, returned entities will contain one or more of these types.

  • prefix (boolean) – Enables prefix (initial substring) match against names and aliases of entities. For example, a prefix Jung will match entities and aliases such as Jung, Jungle, and Jung-ho Kang.

  • limit (number) – Limits the number of entities to be returned. Maximum is 500. Default is 20. Requests with high limits have a higher chance of timing out.