Transforming Data Hurdles into Semantic Knowledge Graphs for Machine Learning

At the top of an organization’s worry list, poor data management is one of the bigger headaches. An organization’s data and content can be quite overwhelming, especially when more than half of it is unstructured — in other words, cluttered and hard to derive knowledge from. 

Semantic tools are the logical solution for many of these data management issues; where a semantic knowledge graph is the driver behind great knowledge maintenance and discovery. With all its capabilities, users of a knowledge graph can overcome the common roadblocks associated with unstructured data.

Addressing the many challenges of data management

Though each organization may have specific struggles according to their unique business practices, many organizations face similar obstacles while managing their data. Across the board, the following challenges are often shared:

• Textual synonyms: In layman’s terms, this refers to different terms meaning the same thing. A food vendor has the same meaning as a food seller but they show up as different concepts in unstructured data.
• Data ambiguity: For people familiar with semantic terminology, data ambiguity can be classified as “homography,” which is when a word has two different meanings. The word, “apple,” for example, can mean the gala apple that you eat, but it can also mean the technology company that sells you an iPhone. The takeaway here is that only one meaning of the word “apple” is relevant to a food vendor. 
• Language discrepancies: Particularly for international companies, unstructured data can be expressed in different languages. Much like a casual conversation between two foreigners, it is quite difficult to interpret the meaning of words and terminologies when a language is not shared.
• Lack of background knowledge: In this regard, much of an organization’s data is missing linkages to knowledge or understanding. Without context, many words are floating around a database seemingly unconnected because there is no clear information there to link the words together.

These examples may not seem so challenging at first. After all, an employee working for a grocery store chain will understand that their data is pointing towards apple like the fruit and not Apple like the company. However, a computer has a more difficult time processing this logic if not helped with semantic knowledge graphs.

While Excel sheets and relational databases are often used to support data organization, they still require a lot of maintenance to make the connections that are necessary to understand data. In our apple scenario, every citation of apple in the text document would have to be combed through and verified by an employee which is a very arduous job.

The good news is that semantic knowledge graphs have the answers to these recurring headaches that organizations experience.

Users of a knowledge graph can tackle each of these problems by taking steps that are fundamental to the process of building knowledge graphs. 

Consider the taxonomy, for example, which serves as a backbone to all things semantics. In a taxonomy, terms, documents, etc., can be classified into a hierarchical system that allows for precise organization and management of data. A thesaurus manager helps facilitate the usage of entity extraction, which is text mining at the highest level, and automated tagging. With these tools, you can pull key terminologies from large expanses of text so that they can be mapped as semantic concepts. The concepts allow users to take advantage of synonyms, multilingual, and alternative labels to avoid many of the problems previously stated about language discrepancies. 

Think of it in terms of an actor’s bio on Wikipedia. In this example, Meryl Streep’s page is filled with text, most of which are common “filler” words used to string sentences together, but there are words that also provide a lot of context and meaning, highlighted below: 

“Often described as the “best actress of her generation”, Streep is particularly known for her versatility and accents. She has received a number of accolades, including being nominated for a record 21 Academy Awards, of which she has won three, and a record 32 Golden Globe nominations, winning nine.” 

An entity extractor will hone in on a few words out of many to derive the fundamental meaning out of the unstructured text — in this case, we derive that Meryl Streep is an actor who has been nominated for 53 awards, and won 12 of them. The entity extractor will help automate this process so you don’t have to comb through it yourself.

Knowledge graphs as a vehicle for better search experiences

On top of all this knowledge discovery, knowledge graphs provide a plethora of tools that advance an organization’s existing applications – both from an internal and customer-facing perspective. On front-end websites often found in the ecommerce sector, a semantic search can fetch results that match the meaning of a user query instead of focusing exclusively on the exact words and phrases. This way, the search engine can interpret user intent and behavior to make search experiences more accurate.

Semantic recommender systems can be used to “matchmake” employees within a company for HR purposes. Colleagues can be connected to each other based on their semantic “footprints,” or profiles, that contain various descriptive skills and background information to work on relevant projects together and be recommended to interesting career opportunities. 

Question and answering machines can be used to generate structured search results from organic text, which can be helpful to an organization when trying to figure out which employees can perform certain tasks. In this case, an employer could input “employees that speak Spanish” in the search field, and the search will pull specific employee data from CV archives to make the connection between the natural language search, and the keyword “Spanish.”

These are just a few of the powerful semantic tools that organizations can use to transform their existing processes and elevate their data to the next level. The knowledge graph sits at the center of these tools to make them possible. 

Adding context and reason to knowledge graphs to enable machine learning

Where there are gaps in the data, a semantic knowledge graph can provide missing background information by linking contextual pieces together. The knowledge graph gives the data the ability to “reason” in ways a computer cannot because it maps the relationships of data objects to each other. 

Additionally, a well thought out knowledge graph can help to minimize risks in data management and actually increase the quality of data. Since knowledge graphs, and therefore the way you model your knowledge, are based on rules defined in an expressive ontology, knowledge graphs can help organizations make educated conclusions about their data, and train it for machine learning algorithms.

As stated by Semantic Web Company CEO Andreas Blumauer in KMWorld’s recently published article, “If you take a look at how your knowledge model was used to contextualize your training data, and that produces a certain probability of this or that prediction, then of course you can steer the enhancement, development, and evolution of your model in a good direction.”

The article, written by editorial consultant Jelani Harper for KMWorld, summarizes knowledge graphs as a central point for AI and machine learning. Knowledge graphs facilitate interoperability between the departments of large enterprises, so that data is scalable and prepared for machine learning. The context and labels embedded within knowledge graphs help produce “supervised learning deployments” and training datasets. 

Furthermore, it is with natural language processing and entity extraction that organizations can quickly create training sets without having to deal with the typical headache of scraping the web. Because it has the ability to process natural language found in text documents, the semantic knowledge graph provides a useful format for feeding machine readable data into the training algorithms.

“There’s a big turnaround in the market for eventually fusing symbolic AI, which is represented by knowledge models, and statistical AI, which is represented by machine learning,” said Blumauer.

Knowledge graphs help the machine learn how to understand an organization’s data better, so that they can continuously automate improvements and knowledge discovery.