Case Study:
Google’s understanding of the Software sector
March 2020
Executive Summary
Google uses Natural Language Processing (NLP) routines to convert its understanding of text into topics and concepts (Entities). We looked out how much understanding Google misses using this approach in the Software sector.
InLinks analyzed the page content of the top 10 Google results (in the US Market) ranking for the phrase Digital transformation and compared the Named Entities recognized by Google’s NLP API with the proprietory routines designed by Inlinks to uncover gaps in Google’s machine learning in the Software sector
The results showed that 12.3% of entities seen on the results in the Software sector SERPS (search engine Results pages) were positively identified by Google.
This compares to 17.5% average across all analyzed industry sectors.
How the Sector Compares
Different sectors tend to be analyzed with a different degree of accuracy by the search engines. This stems from two main challenges.
- the more demand there is by consumers within a given sector, the more the need for search engines to apply more sophisticated entity recognition to better answer user queries.
- the more sophisticated the industry is in creating search-friendly content, the more search engines can surface topics and concepts.
| Software Industry | Google Analysis | InLinks Analysis |
|---|---|---|
| Avg. Number of words per page | 3191 | |
| Avg. Number of Topics per page | 8 | 65.1 |
| Benchmark - Avg. nb of entities/page (all sectors) | 9.2 | 51.8 |
| Of which, Topic types are: | ||
| - Persons | 1 | 0 |
| - Organizations | 33 | 42 |
| - Cities & geo. areas | 11 | 12 |
| - Concepts | 23 | 355 |
| Semantic Density | 8.2 | |
How the research was conducted
Google’s Search API returned URLs from the following sites for the following sites competing for this phrase: "Digital transformation"
#en.wikipedia, #enterprisersproject, #sloanreview.mit.edu, #cio, #forbes, #i-scoop.eu, #infoworld, #redhat, #salesforce, #zdnet
The texts of each page are then sent to Google’s NLP API, in order to determine which entities are identified by the search engine. These are important for search since Google is then able to link these to its Knowledge Graph to feed its services including Google Discover, Google search, Voice Search and Google News. (Although, correct identification does not guarantee inclusion in these results)
Here is first of all the synthesis of the results returned by Google:
- 33 organizations, including Amazon (company) (detected 3 times) Forrester Research (2) Wikipedia (2)
- 23 concepts, including Uber (2) Chief information officer (2) DevOps (2)
- 10 geographical areas, including Japan (2) Trailhead (1) South Korea (1)
- 1 event, including World Economic Forum (1)
- 1 city, including Hong Kong (1)
- 1 person, including Andrew McAfee (1)
Errors in Google’s Detection Rate
Here are some errors in the categorization of entities:
- Taiwan was categorized as a concept instead of a geographical area
Most important entities (provided by InLinks), compared to those identified by Google:
- Digital transformation (seen 10 times) => detected 1 time by Google
- Information technology (9) => NOT detected by Google
- Technology (8) => NOT detected by Google
- Business (7) => NOT detected by Google
- Cloud computing (7) => detected by Google
- Company (6) => NOT detected by Google
- Disruptive innovation (6) => NOT detected by Google
- Management (6) => NOT detected by Google
- Employment (6) => NOT detected by Google
- Service (6) => NOT detected by Google
- Leadership (6) => NOT detected by Google
- Industry (5) => NOT detected by Google
- Application software (5) => NOT detected by Google
- Customer (5) => NOT detected by Google
- Sales (5) => NOT detected by Google
How can the Software Industry benefit from this report?
By understanding where Google is failing to recognize important concepts within the industry, there is an opportunity for companies in the sector to write clearer content that Google can better understand.
Another option is to explicitly state these concepts in Webpage schema for machine learning algorithms to take into account. This would require using Schema.org/WebPage and the "about" and mentions" properties for important concepts such as Information technology, Application software.
Internal linking of topics to pages dedicated to each important topic will also help to reduce cannibalisation of content in Google’s understanding of context within your content.
In Summary
Google's understanding of the Software market, based on its NLP algorithms remains limited at 12.3% for this industry. Businesses either need to improve their schema or make their content more understandable by Google to improve its level of understanding.