How machine learning is making search more human

Data might be the new oil but too many companies are drowning. While most understand the need to collect and make use of data, too few know how to do it.

“The problem,” says Bob Hayward Chief Customer Officer of Search365, “is that you can’t really act on your data unless you can find it.”

This has been the driving force behind Search365 since its inception in 2014. The company was founded on the idea that too much important information was either internally inaccessible or not structured for searches to deliver useful results.

The solution? Getting machines to do the work. Search365’s intuitive enterprise search function has seen it work with Australian government clients and large enterprises such as NAB, Qantas and Seek.

Now it’s raising the stakes with artificial intelligence (AI).

“We believe there’s a way of automating the insights that stem from the process of search – transcribing speech, translating names, scanning through a million documents trying to find particular context and putting them in some order,” says Hayward.

“AI is doing things people can’t attempt because it would take too long and be too expensive – analysing 22 million emails for example. It’s providing insights that will make them more productive.”

An extension of search

Search has always been in the DNA of Search365. Most of the company’s founding staff came from an enterprise search technology vendor called Fast, which was acquired by Microsoft in 2008. Today it’s a Microsoft partner and its products are built on Microsoft technologies.

Search365 started out building search capability that allowed clients to find information inside content. Increasingly, these organisations needed help analysing vast pools of data.

“It’s an evolution. As we helped clients open up search we were inevitably asked to improve relevance and accuracy. That very quickly leads you into the world of natural language processing, which is a division of AI,” Hayward says.

“Character matching isn’t good enough when it comes to searching for names. You’re going to miss too many names out there and that’s a problem. Machine learning needs to be enriched and we’re being driven in this direction by client requirements.”

Diamonds in the rough

Think about a large company holding tens of millions of unstructured emails on a number of servers. Imagine if it was possible to classify that information and make it useful by producing relevant analysis within hours.

Search365 is able to accomplish this with three of its tools – Achromat, Refraktor and Lentikular.

“We find data by going to the source instead of building data lakes and data warehouses. Then we act upon it by analysing, interpreting, transforming and enriching that data,” Hayward says.

“Our specialty is in text analytics – using machine learning techniques from Microsoft to convert rich content into text. Once it’s in text format we can do our magic which, from an AI perspective, includes natural language processing, auto-classification and deep learning.”

The three products are all are named after lenses because they provide a view of data through different filters. Achromat is an enterprise search product but Refraktor, a data analytics platform, and Lentikular, an auto-classification tool, both take this idea of search and refine it through the power of artificial intelligence.

The products use SharePoint and Azure and can also search and analyse content in Dynamics, Outlook and other Microsoft systems. Security is a priority for Search365 with clients including the Department of Foreign Affairs and Trade, the Department of Communications and the Arts and the Fair Work Ombudsman.

Natural language processing is a core strength, particularly when it comes to fuzzy entity matching. This allows the system to fill in gaps, grouping data even when there might be a spelling mistake or information written in a different language. It’s able to pull information on the relationships between entities and people and show it graphically, making it easier to see patterns in data.

Automation is humanising work

The impact of the technology is huge. For government bureaus it simplifies the process of investigation by drawing links between disparate data sets. Now clients can easily understand the relationship between people and entities.

It also has use in large enterprises where they need to see the relationship between files across different ecosystems. For example, insurers can look at a large number of unstructured text files in claims management to see if there are patterns that can identify fraud. This would be impossible without smart technology.

Future development plans are exciting. Search365 is collaborating with Microsoft’s Azure Cognitive Services Toolkit to build game-changing voice analytics. The aim is to make a lot of current processes voice-activated in a way that also satisfies security requirements. It’s the next step in machine learning and promises to have a huge impact on Search365 clients.

“Technology should make life easier,” Hayward says. “We believe in human-aware technology, where the machine adapts to our way of seeing and doing things. Thanks to artificial intelligence, we’re one step closer.”