GraphAware is proud to announce the release of Hume 2.6. This new release brings some major updates and exciting new features to our customers. In particular:
- [New] Hume.Perspectives: a mechanism for specifying multiple subschemas of the main Knowledge Graph. It improves security and readability, allowing users to specify who can read what.
- [Improved] Hume.Labs 2.0: the latest version of Hume.Labs improves security, aligns it to the Hume RBAC model, simplifies management of multiple projects and enormously reduces (almost to 0) the data scientists’ effort for building language models and related skills from an annotated text or dictionary.
- [Improved] Hume.Viz: the graph visualisation has been dramatically improved. The list of new features spans from multiple selection options, to the ability to undo changes. Hume.Actions has been empowered with a new return type - Preview, the “customer-favorite”.
Hume also grew in terms of stability, usability, and scalability. Several components have been added to Hume.Orchestra, multiplying the number of supported data source types - to cite the most exciting, Azure Blob Storage and Samba. Also supported are the processing components - such as the MessageTransformer and a secure python interpreter.
This new version introduces the concept of perspectives. Hume allows our customers to manage Knowledge Graphs of any size. Big companies leverage Hume to merge siloed data sources and organize them in this single connected source of truth. This approach overcomes the isolation of the data silos and unlocks new types of analysis, but it comes at some costs:
- Data security : even in the merged data fabric, data security and privacy must be guaranteed.
- Data Volume: it could happen that due to the size of the Knowledge Graph, navigation is hard. The analyst could be distracted by parts of the graph that are not relevant for them, but appear during the navigation.
Perspectives address those issues, making even a huge Knowledge Graph manageable and secure. Formally, they are subschemas of the main one where a subset of nodes, relationships and even properties are selected. If the schema is generic and an abstraction of a graph database instance, perspectives are “operational schemas”- they map the schema with a concrete graph. Graph visualisations access the final database via perspectives. Once created, a perspective is a new object that can be managed using the RBAC mechanisms native in Hume; in this way it is possible to assign specific perspectives to specific roles and via these, to well identified people or groups of people.
The video shows how to create two different perspectives from the main one and how this influences the way in which graph visualisations can access the nodes, relationships and the properties stored in the Knowledge Graph.
Hume.Labs is at the core of the Hume capability to customize the NLP behavior. It provides an easy-to-use interface where data scientists can work with the domain expert to instruct Hume to recognize domain-specific entities and relationships. Hume.Labs, in combination with the Orchestra and other features available in the Hume ecosystem, becomes a tool that covers a wider spectrum of needs where manual supervision/training is required for creating (or updating) Machine Learning or Deep Learning models. This new version fills the gap in 1.0 that requires a lot of manual work of data scientists for creating the language models after domain experts have annotated a bunch of documents.
The video shows an end-to-end process that allows a project manager, responsible for creating a new custom model for NER, to:
- Create a new NER project and the Type System with the term classes to identify
- Create a first dictionary of terms and the related skill to help to pre-annotate the documents with known entity terms
- Annotate a bunch of documents
- Review the annotations
- Build the language model
- Create the skill out of the model just created
Empowered Hume Visualisation
Hume Visualisation and Analytics represent the way in which Hume delivers the insights to the end users. It allows users to navigate the Knowledge Graph and explore it, offering a rich toolset. This specific component of the Hume ecosystem is not a generic tool for accessing graph databases, instead, it has been thought, designed and realized as an analytics tool for extracting insights out of a Knowledge Graph. With this clear vision in mind, Hume.Visualisation is constantly evolving considering the needs of analysts and data scientists that have to crunch the data as fast as possible in order to find what they are looking for. In some circumstances and for some domains, “time” is a critical variable. That’s why the focus of this new release was: speed and simplicity. In this context, some new features emerge as milestones in this journey:
- Search improvement: In most cases, search is the entry point of the analysis. You can search for a person, location or whatever is a key entity in your domain. This search has been improved, offering amongst other things, more flexibility via an extended syntax that allows the user to specify multi-words, exact search or partial multi-words search. In addition, it is now possible to search directly in the canvas if the nodes already exist.
- Multiple selection options: Often, performing analysis means select multiple nodes or relationships, or all the nodes that satisfy some conditions. Hume now provides several types of options that make selection even easier and the keyboard shortcuts are the cherry on the cake.
- Make Room Expand: While the analysis is going in depth, keeping the focus is a clear need. Having the graph layout reshuffling all the time clearly represents a big distraction. MakeRoom takes care of moving the camera just enough to see the new results without losing the current context.
- Undo: Everybody makes mistakes and in some cases you regret, for instance, an expansion that added too many nodes to your canvas. Undo comes to rescue you.
- Preview Action: Hume 2.6 introduces this nice feature as a new option in the return type for the action. It allows you to “peep” inside the results of an action before adding to the graph. Moreover, in the case you have multiple possible results, previews allow you to choose what to add to your analysis.
Further sources of knowledge
For those who are interested in these topics and in the way in which Hume delivers high value solutions to customers and users, the next resources provide an overview of the spectrum of possibilities offered.
[Blog] Knowledge Graphs with Entity Relations: Is Jane Austen employed by Google? Our Lead Data Scientist, Vlasta, illustrates in detail, Hume-provided mechanisms for extracting relationships among named entities using a rule-based engine.
[Video] Improving Information Retrieval with Knowledge Graphs and Natural Language Processing Christophe Willemsen, CTO, GraphAware, explains how to apply NLP to extract entities and key phrases to build and search Knowledge Graphs.
From the archives: [Blog] Bring Order to Chaos: A Graph-Based Journey from Textual Data to Wisdom Alessandro Negro, Chief Scientist, and Vlasta Kus, Lead Data Scientist, present a journey for converting textual data first in a knowledge graph and from it, extract some insights.