As research in machine learning and AI continues to progress, front-facing systems have become steadily more complicated, expensive, and limited to specific applications like Business Intelligence (BI). Democratizing technology is a complex issue that involves multiple stakeholders with different sets of abilities. A particular user whose needs are very real, but which are seldom addressed except in BI or military-specific situational awareness situations, is a domain expert with extremely limited technical abilities. In particular, such users do not know how to program, let alone cope with complex machine learning or deep learning algorithms, and cannot satisfy their information needs through a simple Google search for several subsequently-described reasons.

A real-world example of such a domain expert that we encountered in the securities fraud domain is an employee from the Securities and Exchange Commission (SEC) who is attempting to identify actionable cases of penny stock fraud. Penny stock offerings in Over-The-Counter (OTC) markets are frequently suspected of being fraudulent, but without specific evidence, usually in the form of a false factual claim (that is admissible as evidence), and their trading cannot be halted. With thousands of penny stock offerings, investigators do not have the resources or time to investigate all of them. One (technical) way to address this problem is to first crawl a corpus of relevant pages from the web describing the domain, a process alternately known in the Information Retrieval (IR) literature as relevance modeling or domain discovery. The latter term is more encompassing, as it involves not just relevance modeling, but the actual crawling of the data.

Once such a corpus is obtained, an expert in information extraction and machine learning would elicit opinions from the users on what fields (e.g., location, company, stock ticker symbol) are important to the user for answering domain-specific questions, along with example extractions per field. This sequence of Knowledge Graph Construction (KGC) steps results in a graph-theoretic representation of the data (a Knowledge Graph or KG) where nodes are entities and field values and the (directed, labeled) edges are relationships between entities or assignments of field values to entities. Since the KG is structured, it is amenable to aggregations and to both keyword and structured querying. With a good interface, for example, the domain expert can identify all persons and organizations (usually shell companies) associated with a stock ticker symbol, aggregate prices, or zero in on suspicious activity by searching for hyped-up phrases that indicate fraud.

A Google search is inadequate for the knowledge discovery described above as it does not allow one to specify fields nor to limit the search to a specific corpus. Google also does not offer the specialized aggregation facilities that such domain experts would need. While Google works exceedingly well for document-level information indexing and retrieval, it is not designed to support complex analytical tasks, either with or without training.

In our group, we have developed a system called myDIG (my Domain-specific Insight Graphs) that ingests the output of domain discovery, i.e., a raw corpus of web pages, and presents an intuitive multi-step KGC and search interface to a user that requires no programming, can be refined iteratively, and requires an hour or less of example-based training. An implemented prototype of myDIG has already undergone evaluations by domain experts in five different investigative domains, each of which involves significant technological potential. myDIG's dataflow is modular, since it allows users to construct and search knowledge graphs, and thereby discover new and relevant knowledge by conducting data-driven analysis. We presented qualitative and quantitative data collected from the five case studies to illustrate the real-world potential of myDIG in making an advanced set of knowledge discovery technologies accessible to non-technical users. Using myDIG, practicing, non-technical domain experts from the five case study domains were able to build a personalized domain-specific search engine over corpora containing more than a million raw web pages in 4--6 working hours. The myDIG project is open-source, with a dedicated web page (http://usc-isi-i2.github.io/dig/) and a GitHub (https://github.com/usc-isi-i2/dig-etl-engine) repository associated with it. More details on the project can be found in the open article 'myDIG: Personalized Illicit Domain-Specific Knowledge Discovery with No Programming' published in Future Internet MDPI, 2019.