To Draw Is Human: Toward No-Code Subgraph Search

Due to the worldwide shortage of developers, growing talent gap, and budgetary challenges faced by small- and medium-sized businesses in hiring software teams, low-code or no-code frameworks are the latest disruption in the business world.¹ For example, SAP recently launched SAP AppGyver, which is a “no-code application development platform that enables developers of all skill levels to create enterprise-ready applications with drag-and-drop simplicity.”⁵

The demand for such low-code or no-code frameworks is not limited to software applications development but also for easy access and search of data residing in databases. Specifically, lay users should be able to access them without needing to write a single line of code. However, query languages (QL)—the primary means to access data residing in databases—enforce end users to be proficient in these languages before they can take advantage of databases for their tasks.

Visual query interfaces are popular low-code/no-code frameworks that facilitate access and search of databases by drawing queries with “drag-and-drop simplicity” instead of writing them.

Visual query interfaces (VQIs) are popular low-code/no-code frameworks that facilitate access and search of databases by drawing queries with “drag-and-drop simplicity” instead of writing them using a QL. Given the ubiquity of graphs to model data in a wide variety of domains (for example, biology, chemistry, ecology, social science, and journalism), we summarize our 15-year odyssey that began in 2008, long before low-code/no-code frameworks became an industry disruption, to address unique research challenges brought by such frameworks in subgraph search environment. Specifically, subgraph search query is one of the most popular query paradigms for accessing graph data. Since graphs are intuitive to draw, graph data management tools from academia and industry (for example, PubChem^a) are increasingly exposing VQIs to enable an end user to draw a subgraph search query interactively without writing a single line of code in a proprietary QL.

We focus on two novel streams of research that exploit the central role played by no-code visual environments in subgraph search. First, classical VQI frameworks for low-code/no-code search suffers from several limitations, such as high creation and maintenance cost, lack of superior support for no-code subgraph query formulation, and poor portability across application domains and data sources.² We introduce the paradigm of plug-and-play (PnP) VQI that addresses these limitations by automatically constructing a no-code VQI framework for a graph repository.² It departs from the traditional mantra of “manual” VQI construction using software developers by automatically generating and maintaining a VQI for a given graph data source in a data-driven manner without resorting to any coding. A user can simply “plug” a PnP interface on top of their graph data source (that is, socket) and “play” by drawing subgraph queries using drag-and-drop. In particular, the design of PnP interfaces is grounded on well-founded principles of human-computer interaction (HCI) and cognitive psychology to enhance the usability and reach of no-code subgraph querying frameworks.

Second, given a classical or PnP VQI that supports no-code visual querying, we explore the unique opportunity of blending visual query formulation with query processing.³ Specifically, we interleave (that is, blend) query construction and query processing to prune false results and prefetch partial query results by exploiting the latency offered by the drag-and-drop activities during query formulation. The key benefits of this paradigm are at least twofold. First, it significantly improves the system response time (SRT), which is the time a user waits to view the query results. This is because the query processor does not remain idle during visual query formulation by processing the potential subgraph query early based on “hints” received from the user. In traditional query processing paradigm, SRT is identical to the time taken to evaluate the entire query. In contrast, in this paradigm SRT is the time taken to process a part of the query yet to be evaluated (if any). Second, as a visual query is iteratively processed during query formulation, it paves the way for realizing efficient techniques that can enhance usability of graph databases, such as query suggestion, empty result feedback, and exploratory search.

The design of PnP interfaces is grounded on well-founded principles of human-computer interaction and cognitive psychology to enhance the usability and reach of no-code subgraph querying frameworks.

Both of these research directions depart from the long-standing classical paradigm of query formulation and processing in a no-code/low-code environment. Since the inception of VQIs, they have been constructed manually by programmers. We depart from this tradition and demonstrate how it can be constructed in a data-driven manner paving the way for no-code generation of no-code subgraph search frameworks. Second, since the inception of database technology several decades ago, the classical paradigm of querying has always been served “neat.” A query is formulated by an end user or application and the query processor is responsible for evaluating it once it is completely formulated. That is, query formulation precedes query processing. They were not blended. Our research opened the possibility of blending them together in the context of graph data by exploiting the unique characteristics of no-code querying frameworks, bringing in significant benefits to usability and query performance. Subsequently, some visual data systems have leveraged this paradigm of blending for data processing and analytics. For instance, Northstar⁴ exploits “think time” available in an interactive VQI for approximate query processing.

In conclusion, we note the two key enablers of this research—HCI and data management—have evolved into two disparate and vibrant scientific fields, rarely making any systematic effort to leverage techniques and principles from each other toward superior realization of these efforts. Through our research, we continuously strive to bridge this chasm to realize superlative and user-friendly no-code querying frameworks for diverse end users.