Navigate

Knowledge Graph Question Answering

What is Google’s Knowledge Graph Question Answer Feature?

Knowledge graph question answering (KGQA) is taking a lot of real estate in search engine result pages (SERPs).

Google’s Knowledge Graph question answering feature answers users’ queries without requiring them to click through to a website. Every search engine is hoping to return the best information based on the intent of the searcher. QA elements have a rich history in search engine result pages.

To be a trusted go-to source for answers, you need to be known online. Google understands query streams and uses them to identify topics and extract trusted data from the web to update ontologies. Google cards, knowledge graphs (KGs), and knowledge collections are a way for users to interact with Google. Like “people also ask” questions on search results, Knowledge Graph question answering keeps people on Google SERPs longer.

Table of Contents

Let’s first establish a foundational vocabulary.

What is the Difference Between Knowledge Panels and Knowledge Graphs?

Knowledge graphs may be sourced to provide richer knowledge panels in search results and return answers to queries.

It helps to view Knowledge Panels as a front-end manifestation of the Google Knowledge Graph. More data is behind what we see in panel graph data. Once you establish a Knowledge Graph entity, Google will rely on it and consider it a canonical source of information. The tech giant did not invent the KG as a supplemental to desktop user experiences; it was a response to the need for better mobile query answers. So many sites were (and still are) horrible on mobile devices. The GKG intends to provide accurate information to its user; its primary goal is not to drive traffic to your site.

Previously, Google doesn’t seem to rank web pages on the basis of accuracy. Today, its quality evaluators have more instructions on how to assess Experience, Expertise, Authoritativeness, and Trustworthiness (E-E-A-T). Answer accuracy is a trust factor, and its guidelines tell us that trust is the most important factor. In contrast, “accuracy” is a factor in which entities display in Knowledge Panels.

Knowledge panels are one rich results type in Google’s search results pages. They give searchers a vetted overview of information related to a given entity. If you are preparing content for AI Overview features, you’re doing the right steps for Knowledge Graph QA inclusion.

What is the Difference Between Knowledge Panels and Google Business Profiles?

Google Business Profiles (GBP) looks much the same as its knowledge panels. GBPs are unique to businesses that serve customers at a particular location or within a designated service area. GBP access lets business owners manage their digital presence on Google Maps and search. This is free. In contrast, your Google Knowledge Panel (GKP) is automatically generated by Google using information about your entity online. It has full control over its propagation and what it chooses to update within it.

What is the Difference Between Google’s Knowledge Panel and Knowledge Vault?

Think of the Google Knowledge Vault (GKV) as produced by an algorithm that generates a machine-readable encyclopedia.

Google only adds information to its GKV once it is assured that what it displays in Knowledge Panels is correct and useful. The GKV is solely based on machine learning and machine logic. Separate entities from multiple domains are moved into the Knowledge Vault only after Google’s global knowledge algorithm gains sufficient confidence in its understanding of the specified entity.

“…we introduce Knowledge Vault, a Web-scale probabilistic knowledge base that combines extractions from Web content (obtained via analysis of text, tabular data, page structure, and human annotations) with prior knowledge derived from existing knowledge repositories. We employ supervised machine learning methods for fusing these distinct information sources. The Knowledge Vault is substantially bigger than any previously published structured knowledge repository, and features a probabilistic inference system that computes calibrated probabilities of fact correctness.” – Knowledge Vault: A Web-Scale Approach to Probabilistic Knowledge Fusion [1]

Answering Complex Questions with Machine Learning

Google receives 93% of daily queries. Just how it traditionally functions as a search engine and ends up to your product or service. To improve on its question-answering capabilities, a Google patent states that: “Natural Language Processing (NLP) can involve answering natural language questions based on information contained within natural language documents.”

“Techniques described enable answering a natural language question using machine learning-based methods to gather and analyze evidence from web searches.” – [2]

 

Question answering over knowledge graphs (QA-KG) is a method that uses facts in a knowledge graph (KG) to answer natural language questions.

This helps users access knowledge in the KG more easily, even if they don’t know its data structures. However, before adding entities to its knowledge base, Google has to first algorithmically understand the question being asked. It seeks to understand the query intent that triggered the question.

For ambiguous queries, semantic interpretation assists in answering complex questions and seeks to replicate human cognition. Web articles often fail to display a publication date or when it was last updated. In contrast, Google’s Knowledge Graph continuously updates. For example, I was about to quote an article for this writing but first researched and saw “This article is more than 3 years old.”

MarketWatch estimates that “the semantic knowledge base industry will be worth $33 billion by 2023, with year over year growth of 10% through the rest of the decade.” It’s January 18, 2023, Semantic Knowledge Graphing Market Size Related To Time And Cost Is Expected To Grow The Industry In Upcoming Years To 2029 article includes Semantic Search, Question and Answer Machine and Information Retrieval.

It is mind-splitting just how much of a rise in scientific innovation is dedicated to better KGs. Equally, digital marketers and SEOs benefit by quickly adapting.

KGs are generally seen as large-scale semantic networks that store facts as triples in the form of (subject entity, relation, object entity) or (subject entity, attribute, value). The edges in the graph represent the relations between these entities. Most KGs are built on top of different existing data sources to connect data. Until GPTChat emerged within GPT3, Google was unthreatened by other large-scale KGs, such as DBpedia, Freebase, and YAGO.

The push for more humanlike question answers

The competition is at an unparalleled scale between Google, OpenAI, Bing, and others to provide more humanlike answers to questions instead of just links to information. Google continuously uses and tests various large AI language models to improve its search engine and knowledge panels. We see stiff search engines competition to take a lead in providing the best search experiene. This has prompted Google to release its Search Generative Experience Search Lab as quickly as possible.

The term ‘knowledge graph’ has a vast relational family; it includes the fields of knowledge graphs, graph databases, knowledge vaults, knowledge panels, neural networks, machine learning, NLP, artificial intelligence, linked data, knowledge graph embedding, knowledge transfer, transfer learning, Knowledge Representation Learning (KRL), and more! Spending money on paid search and trivial site performance improvements pales in comparison to effectively filling question answer content gaps. The suggestions below come from my own experience.

The company’s data-driven systems are evaluated to establish trust in the scientific approach and its applications. Its Knowledge Graph (KG) Question Answering (QA) capabilities rely on complex data structures that are made accessible via natural-language interfaces.

How to Create Question Answer Content that Google Finds Helpful

The new SEO understands that Google is a type of answer engine and feeds it.

The more you publish verifying data, the more the tech giant can connect data. In this way, you facilitate the work of a search engine in understanding what the facts about your entity are. You provide help when you connect your own structured data across all the different third parties that talk about you. Google has no preference for whether structured data implementation is connected through a graph or node array versus having them as individual elements in their own blocks on the page.

  • FAQ content: Your company can create databases that are marked up with schema to assist Google in crawling and ingesting question-answer informational pages. Google may choose to source your website’s FAQ content; examples are: in “People also search for“, “People also ask”, “Others want to know”, or “Related searches”.
  • Website topic clusters: Information with a clear ontology can be used to denote topic expertise. Knowledge graphs organize entities using web data that Google trusts. You can be the primary source in different datasets. In this way, you are a data publisher. If you have claimed your knowledge panel, it may be a more reliable and quick way to trigger a knowledge panel update.
  • Accurate product database: As long as you do an impeccable job of keeping your product database updated, you are helping Google obtain high confidence and trust for your product facts. Google is more confident to show its users accurate and relevant information if your brand and products online are clear and consistent. Be consistent with everything when it comes to your online presence. Go by the same spelling, title, author bio, place of work, etc.
  • Upload image datasets: Images coming out of that particular database can be associated with your answers and populate your knowledge graph. The existence and accuracy of your Product QA datasets help ensure comparability.
  • Use FactClaim schema markup: Google’s search results are often drawn from its Knowledge Graph repository of billions of facts about people, places, and things. By including factual, statistical content that supports your opinion pieces, you show your awareness and knowledge of relevant fact-based sources.
  • Consistent Name, Address, Phone: There are more ways to manage your Google Business Profile moving into 2023. However, your NAP is foundational to how Google identifies your entity. It works best to have a stable address and use the one assigned in Google Maps. Knowledge graphs closely relate to Google Maps. It’s based on structured data, structured information in the form of NAP consistency: name, address, phone number, and how those make a difference in making sure that Google Maps updates. The same type of consistency supplies the GKG.
  • Automated Google Business Profile FAQ text responses: You can add automated FAQ responses directly in your Google Business Profile. It functions as an automated two-way conversation with question answering.
  • Incorporate an effective Google Post strategy: Google Scholar authors, notable brands, and US elected officials aren’t leveraging the opportunity to claim their Knowledge Panels. This in turn provides them access to Google Posts, which should be a part of your knowledge graph strategy for content.
  • Use audience data and market research: Initial market research provides audience data insights that can power innovative content campaigns and KG strategies. A knowledge base first classifies questions based on how “significant” they are in relation to people’s query intent.

More on using structured data on your website:

Ryan Levering from Google who primarily works on structured data stated on Mastodon: “Whatever the graph looks like for the whole page is what we use, regardless of where it comes from. It gets mushed together and while do know where it came from, that’s not usually used. However, the caveat here is that when you do it in multiple blocks, there are sometimes conflict/duplication problems. Also, over time richer/correct semantics will favor more connected graphs. We still see cases where people throw unrelated markup about things (like related products) at the same top level as the main entity from different blocks on the page and that makes it mostly noise. So sometimes centralizing the logic makes it more consistent/correct.”

A goal of graphs is the ability to function as the ground truth of the terminology, logic, and correct answers.

Here is a quote directly from Google about how its Knowledge Graph works.

“Google’s search results sometimes show information that comes from our Knowledge Graph, our database of billions of facts about people, places, and things. The Knowledge Graph allows us to answer factual questions such as ‘How tall is the Eiffel Tower?’ or ‘Where were the 2016 Summer Olympics held.’ Our goal with the Knowledge Graph is for our systems to discover and surface publicly known, factual information when it’s determined to be useful.” – How Google’s Knowledge Graph Works

You can feed your Knowledge Graph with information that demonstrates relationships and concepts connected to each other. While huge investments in chatbot artificial intelligence are underway, we currently know it needs a domain model in order to understand and answer questions. Machine learning can generate a huge knowledge base of sentences and use cases, but a static chatbot has limitations.

Google collects information about a particular topic or subject to first establish confidence before a data Knowledge Graph entry is updated. Graphs help us answer data-related questions so that Google can store and retrieve information easily. It basically comes down to understanding questions, connecting the questions to your knowledge graph, and inference the answers.

Suggested steps for KG question answer optimization:

  1. Look for what, who, where, why, and also how publications that you control.
  2. Identify which internal QA data can be sourced externally.
  3. Learn where to find it.
  4. Learn how it is already being used, by whom, how it may be used, and why.
  5. Use graphs to identify how to provide more value by analyzing their clusters, cohorts, and groups.
  6. Set up alerts to help monitor QA data signals concerning context, group signals, and dynamics within and with your entity relationships.
  7. Schedule maintenance time to manage and feed your graph QA content.

Natural language processing and graph alignment management facilitate finding cases of conflicting entities or relationship definitions. Google’s panels, graphs, and vault are about entity resolution.

Before answering a question on a platform that you control, intelligently understand the question first. You should know the searcher’s intention and the key information needed for the question. Search engines extract key information by searching named entities that are useful for knowledge graph inclusion. To be trusted themselves, they are selective before inferring the answer on the KG.

How to Request a Google Knowledge Panel Update?

Google provides its claimed Knowledge Graph owners a way to request updates and report issues. It is easier once you’ve gained the ability to provide direct feedback. Its instant answers are regularly updated from crawling the web and user feedback.

“We also know that entities whose information is included in knowledge panels (like prominent individuals or the creators of a television show) are self-authoritative, and we provide ways for these entities to provide direct feedback. Therefore, some of the information displayed may also come from verified entities who have suggested edits to facts on their own knowledge panels. – About knowledge panels

“We also receive factual information directly from content owners in various ways, including from those who suggest changes to knowledge panels they’ve claimed.” – How Google’s Knowledge Graph Works

 

Many consider the key benefits of gaining a semantic knowledge graph are that it provides brand clarity, data recovery, and sales experiences. But since so many people ask questions, it’s important to also consider its ability to integrate data and use it for providing answers. What not be the retailer that proves valuable in this way?

How Does Question Answering Information Retrieval Work?

Google pulls together questions cluster content from sources it can be sure of.

2023 is the era of improving your Knowledge Graphs strategy as more and more lead conversions take place directly on search engine result pages (SERPs). Google assesses what it can trust about your entity and chooses what will be included in your Knowledge Graph, Knowledge Panels, and Knowledge Vault. It knows about your target audience and customers; it seeks to align your strengths and knowledge across the web to provide the best answers. Audience research and SERP analysis can inform your approach to marketing.

When Google extracts QA entity information from Web pages, association scores involving those entities and their relationships with other entities are determined. It cares much about factual answers describing the properties of those entities. Once you’ve determined your best marketing strategy, it’s time to move it into marketing tactics, where you’ve taken specific marketing actions to improve your SERP outcomes. Both today and even more so in the future, understanding QA information retrieval and how to inform your KGs is a vital component of effective SEO.

We learn from Google patents how a natural language processing model may answer a natural language text question.

“A computing system includes a machine-learned natural language processing model that includes an encoder model trained to receive a natural language text body and output a knowledge graph and a programmer model trained to receive a natural language question and output a program. The computing system includes a computer-readable medium storing instructions that, when executed, cause the processor to perform operations. The operations include obtaining the natural language text body, inputting the natural language text body into the encoder model, receiving, as an output of the encoder model, the knowledge graph, obtaining the natural language question, inputting the natural language question into the programmer model, receiving the program as an output of the programmer model, and executing the program on the knowledge graph to produce an answer to the natural language question.” – Natural Language Processing With An N-Gram Machine, Patent No.: WO2019083519A1, Publication Date: May 2, 2019 [3]

 

Knowledge Graph Relevance Scoring

Combine machine language learning and data graphs to connect the context of audience question to your answers. Google KG relevance scoring is using pre-trained LM to score nodes on KGs conditioned for a question response. Google has a general framework for weighting information within its KGs. Its machine learning uses joint reasoning over text and KGs. In this way, it connects the context of questions with answer content by using LMs and graph neural networks.

Overall, Google KGs are more efficient and trusted than web pages. So where is this going?

Question Answering KGs Seek to Provide Verified Knowledge

Google Knowledge Graph Provides Direct Answers to Queries

The facts provided by the Google Knowledge Graph in response to a query are initially derived from other sources. (Until recently, this was largely from Wikipedia and Wikidata). Google works hard to trust any and all information populating its KGs. It must be challenging to satisfy queries accurately. For example, to answer “Who were the founders of Google?”, the Knowledge Graph needs to extract a triple (subject-predicate-object) here along the lines of “[Organization] founded by [Person(s)]”

Wikipedia and Wikidata provide precise information like that.

Aaron Bradly, Knowledge Graph Strategist at Electronic Arts, posed a fascinating question on Twitter a few years back. “To wit, a bigger underlying question is whether we should consider ‘facts’ provided by the Google Knowledge Graph to be factually correct (and whether Google itself considers Graph-provided ‘facts to be factually correct).”

One can quickly see why the “answers” and “facts” provided by the Knowledge Graph need to be trusted by users.

Bradley goes on to say, “So the Graph needs to lean on the trustworthiness of its sources in determining what assertions to make. So much so that Google has mulled over methods of improving on how they determine the trustworthiness of a source. Ultimately the provided assertion is ‘from somewhere’. And this becomes problematic when the payload of a response (especially voice) doesn’t include provenance information. Both knowledge aggregators (here Google) and knowledge users (here searchers) need to work at improving how we process these questions and answers.” [4]

Larry Page and Sergey Brin, Google’s founders, resurfaced after their 2019 departure to review Google’s artificial intelligence product strategy. They approved plans and pitched ideas to add new chatbot features to Google’s search engine. Google’s January 2023 massive staff layoffs follow its renewed commitment to put A.I. front and center in its plans. [5]

You can use the Google Knowledge Graph Search API to search or look up entities in Google Knowledge Graph. Google Cloud offers the following schema markup code example: [6]

{
  "@context": {
    "@vocab": "http://schema.org/"
  },
  "@type": "ItemList",
  "itemListElement": [
    {
      "result": {
        "@id": "c-07xuup16g",
        "name": "Stanford University",
        "description": "Private university in Stanford, California",
        "detailedDescription": {
          "articleBody": "Stanford University, officially Leland Stanford Junior University, is a private research university in Stanford, California. The campus occupies 8,180 acres, among the largest in the United States, and enrolls over 17,000 students. ",
          "url": "https://en.wikipedia.org/wiki/Stanford_University",
          "license": "https://en.wikipedia.org/wiki/Wikipedia:Text_of_Creative_Commons_Attribution-ShareAlike_3.0_Unported_License"
        },
        "url": "http://www.stanford.edu/",
        "image": {
          "contentUrl": "https://encrypted-tbn1.gstatic.com/images?q=tbn:ANd9GcTfPPf-ker0y_892m1wu8-U89furQgQ67foDFncY3r9sREpeWxV",
          "url": "https://es.wikipedia.org/wiki/Archivo:Logo_of_Stanford_University.png"
        },
        "identifier": [
          {
            "@type": "PropertyValue",
            "propertyID": "googleKgMID",
            "value": "/m/06pwq"
          },
          {
            "@type": "PropertyValue",
            "propertyID": "googlePlaceID",
            "value": "ChIJneqLZyq7j4ARf2j8RBrwzSk"
          },
          {
            "@type": "PropertyValue",
            "propertyID": "wikidataQID",
            "value": "Q41506"
          }
        ],
        "@type": [
          "Place",
          "Organization",
          "MovieTheater",
          "Corporation",
          "EducationalOrganization",
          "Thing",
          "CollegeOrUniversity"
        ]
      }
    }
  ]
}

We find that implementing schema markup is extremely helpful. If you are in double, read our pros and cons of adding structured data markup article.

Google Creates a New Question-Answering System

Conversational AI and Question-Answering systems (QASs) for knowledge graphs (KGs) are both emerging research areas

Bard is currently an experimental conversational AI service. The power source behind it is LaMDA. We’ve learned that its construction is based on large language models and draws information from the web. Sundar Pichai, CEO of Google and Alphabet, says “it’s a launchpad for curiosity and can help simplify complex topics.”

“Open-domain long-form question answering (LFQA) is a fundamental challenge in natural language processing (NLP) that involves retrieving documents relevant to a given question and using them to generate an elaborate paragraph-length answer. While there has been remarkable recent progress in factoid open-domain question answering (QA), where a short phrase or entity is enough to answer a question, much less work has been done in the area of long-form question answering.” – Progress and Challenges in Long-Form Open-Domain Question Answering – Google AI Blog [7]

Given Microsoft’s rapidly increasing involvement in OpenAI, their Knowledge Graph becomes more important to entities. The good thing is, when you optimize for a Google KG, it’s almost certain that you’ll gain a Bing KG with a little extra effort. Google Bard is the tech giant’s version of ChatGPT and we anticipate that its AI-generated answers are likely to require a strong knowledge graph presence.

If you sell products, be aware that your well-built product page entities should include answers that the KG will likely use.

Your efforts to manage a strong entity presence will be even more important in the future.

Taking your Semantic Search and GKG Forward

If this article raises your semantic search and graph technology awareness and now you are eager to respond to such opportunities, call Jeannie Hill at 651-206-2410.

Boost your personal or business knowledge graph by gaining our Queries Entities Audit

 

References:

[1] https://research.google/pubs/pub45634/

[2] https://patents.google.com/patent/WO2014008272A1/en

[3] https://patentscope.wipo.int/search/en/detail.jsf?docId=WO2019083519

[4] https://mobile.twitter.com/aaranged/status/1108444732282163200

[5] https://searchengineland.com/google-search-chatbot-features-this-year-391977

[6] https://cloud.google.com/enterprise-knowledge-graph/docs/search-api

[7] https://ai.googleblog.com/2021/03/progress-and-challenges-in-long-form.html

Jeannie Hill:
Related Post
  1. Strategic B2B Lead Generation
  2. SEO AI Strategy
  3. Knowledge Graph vs. Knowledge Panel vs. Google Business Listing

This website uses cookies.