The Maana Knowledge Graph
The knowledge layer that powers enterprise decision flows
The core of the Maana Knowledge Platform™ is the patented Maana Knowledge Graph™. Unlike traditional “semantic” systems, the Knowledge Graph does not rely exclusively on ontologies and description logics. Instead, the Knowledge Graph separates the structure of data from the content itself. This separation enables a fluidity of modeling, allowing data from any source and in any format to be seamlessly integrated, modeled, searched, analyzed, operationalized and re-purposed.
Each resulting model is a unique combination of three key components – subject matter expertise, relevant data from silos, and the right algorithm – all of which are instrumental in optimizing assets and decision flows.
The first of its kind, the Maana Knowledge Graph is also dynamic, which means that it can represent data and computational models. In addition, it can be used to perform complex transformations and calculations at interactive speeds, making it a game-changing technology for agile development of knowledge applications.
Maana has over 8 Knowledge Assistant modules such as SimilarityAssist™, DocAssist™ and JoinAssist™. These user-guided, machine-assisted tools enable data scientists and technical experts to collaborate and rapidly create new, iterative knowledge models for optimizing assets and processes.
JoinAssist guides users in finding non-obvious relations, even when the user does not fully understand their data relationships. Often, users have a good understanding of the available data, but the effort to manually connect it all to create large data models is difficult and time consuming. Maana’s user-guided, machine-assisted mechanism provides a visual interface to automatically build the relevant joins – and with little manual intervention.
Similarity Assist™ empowers users to identify instances, cases, events or records similar to the one they would like to investigate. For example, when looking for unprofitable insurance contracts, companies aren’t limited to searching for fixed attributes like high claims. With Similarity Assist, they can pinpoint contracts that are most similar to a sample unprofitable contract. The tool’s guiding algorithms search across several dimensions (such as performance, economic, and competitive) to identify similarities given the context of what the user is trying to solve.
DocAssist enables extraction of targeted knowledge – in the relevant context – from unstructured documents. For example, a large amount of data is usually stored in tables or hidden in unstructured documents like PDFs or Word files. Maana’s Knowledge Platform uses machine learning to extract structure from these documents, as well as domain-specific terminology that can be labeled for further training of the machine learning model.
Data are often spread over multiple files, parts or sources, making it both difficult and time consuming to get to more engaging phases of solution development. With SourceAssist, now you can interactively discover and securely acquire relevant data for use within Maana. SourceAssist also helps with the challenge of constantly changing, growing data by letting you configure how Maana deals with repeated or periodic acquisition of the same content.
Building models at speed requires quick and easy access to available components. SearchAssist allows you to browse all data indexed in Maana and easily create or modify models by dragging and dropping results into a workspace. An intuitive search experience ensures that as you type, results continuously update to match what is being typed. Results are displayed in order based on a MaanaRank™, which ensures results with a higher ranking are displayed first.
Visually build knowledge models – the fundamental building blocks for Knowledge Applications – from pertinent “problem” questions such as, “Given a vessel, omitted port, and date, what are the set of alternative port options ranked by score?” With KnowledgeModelAssist, you can search concepts, bring them into the workspace, define relationships between them and apply appropriate functions in order to digitize your subject-matter expertise.
As you build out your Knowledge Application, you can use KnowledgeGraphAssist visualize all knowledge graph elements in your workspace and explore all connected elements, including data sources. This not only helps you get an overview of how the different models are fitting together to form the desired solution, but it also provides the level of transparency required to explain how the models are working to answer specific questions.
As you build knowledge models, you must be able to view the actual data represented by them. PresentationAssist gives you the ability to view information as tables. You can also change visualization types, edit the presented data, and provide inputs to get the answer (i.e., the output) to a specific question (e.g., which units had the most spare parts replacement last quarter).
Maana enables business users and domain experts to utilize Maana’s pre-built Knowledge Applications or easily develop their own so employees can use them to make better, faster decisions. These applications provide recommendations into day-to-day operations to enable optimal decision making and learning for business processes (for example, supply chain, call center, and accounts receivable processes). They can also help optimize asset performance through predictive maintenance and more.
Proactive Financing for Customer Upsell
Identify the most desirable subset of current customers for targeted marketing campaigns – for example, to offer them a line of credit.
Learns and Adapts
Once Maana’s Knowledge Graph recommendations are operationalized via the Knowledge Applications, the models continue to learn and adapt from the actions and feedback of subject-matter experts. This is another key advantage of having flexible models as they can be easily modified when needed. The users doesn’t have to start from scratch, as they can take advantage of prior work.
To solve a complex business problem, the first step is to define the problem using precise problem-questions. Maana enables you to represent these problem-questions within the platform using knowledge models.
Building these knowledge models is fundamental in getting to the right answers that solve business problems. The process requires collaboration between the business leaders, subject matter experts, data scientists and business analysts so that the problem is truly understood within the proper business context and defined using commonly used terms.
Visualize and Configure
Once the precise question has been defined, Maana’s rich, graphical UI can be used to model relevant business concepts and represent their inter-relations as a knowledge model.
Knowledge models are declarative in nature, in that they do not tell the user how to answer the question; rather, they focus attention on what needs to be answered next. This helps user define further questions without getting too deep into the algorithms and data.
Compose and Reuse
Once built, all models are connected to form an intelligent, relationship driven, easily searchable and scalable semantic Knowledge Graph™, which is at the core of the Maana platform. The benefits don’t end once you create your initial Knowledge Graph. The models contained in Knowledge Graph™ can be reused as is or modified and re-applied to another problem across the enterprise.
IT organizations should be able to integrate and extend any new, enterprisewide platform to their current technologies. The Knowledge Platform can support traditional, third-party ETL, data wrangling, visual analytics, dashboards and more through plug-ins, libraries and REST APIs.
Maana supports access to Python and R modules in an interactive environment or via a command line interface. Jobs can be run in a distributed environment for quick results. By using Knowledge Assistants in the platform and by leveraging external libraries, users can solve problems using the best tool available for the job.
The platform also supports backend systems, such as Google’s TensorFlow, gridMathematica, or even a proprietary system through a pluggable interface. Users should have access to the right tool for each job. Whether it is a Python-based optimization package or a complex, deep learning network, fluid integration and composition is necessary and available.
Search and Explore
Maana’s Semantic Search enables users to find the most relevant data and knowledge inputs from across silos in the context of optimizing an asset or process. Once these knowledge inputs are indexed in the Knowledge Graph, this patented technology suggests and completes user queries using domain-specific knowledge from the Knowledge Graph.
By entering a keyword or phrase, the user is presented with a disambiguation dialog that provides a ranked and filtered view of all knowledge in the graph. Once the user guides the Knowledge Graph on the relevancy of the suggestions, data is presented using rich, interactive visualizations for subsequent refinement and action.