American cloud computing company, Salesforce is looking to develop a novel way to help workers cut down on time spent sorting through emails. Despite the task being a necessity in most working environments, reading over emails is tedious and time-consuming as well as a common killer of productivity. Salesforce is aiming to automate this repetitive process and for this reason, the company has assembled a team of MetaMind researchers who are working on a set of summarization tools for business communications.

In 2016, Salesforce acquired deep learning startup MetaMind with the aim of integrating the company’s technology into its range of products and the summarization algorithms are the latest fruits of this venture.

Over the years, studies have shown that employees in the United States spend up to six hours a day checking their inbox, but with the new tools in place time spent doing so could be significantly shorter. The technology would prove extremely beneficial to Salesforce’s customer-service focused products, as it could, for example, potentially allow sales representatives to quickly familiarize themselves with information and emails, whilst freeing up more time to focus on other important tasks.

In order to achieve this goal Salesforce and MetaMind are harnessing the power of machine learning to develop innovative ways of summarizing larger walls of text. Current methods of producing text summaries are lacking in many areas, including flexibility and coherence. The Salesforce team, on the other hand, is using a slightly different approach which brings machine learning methods into the equation including reinforcement learning, supervised learning and contextually increased learning.

By using reinforcement learning, the researchers are able to basically train neural networks to achieve optimal behavior, which is fairly akin to how animals learn. Machines are being offered positive feedback for working in a certain way, which in turn allows them to determine how to act next. The results are calculated using an evaluation metric known as ROGUE (Recall-Oriented Understudy for Gisting Evaluation). Through applying this technique, researchers have found that ROGUE-optimized reinforcement learning can lead to improved recall, language flow and coherence.

Researchers also found another way of teaching the machine through supervised learning; Models are fed positive examples of good summaries which allows them to determine how to behave in future situations. Researchers also granted machines access to look back at the original texts in order to extract any additional context.

The initial results of these experiments are pretty amazing. Salesforce and MetaMind have shown great success in their research as the abstracts generated by their model are kept short, but manage to contain all the main ideas of the original article. Despite these positive developments, Salesforce’s summarization tools are still far from being anything more than research. Regardless, the work emphasizes the company’s ongoing commitment of pursuing machine learning ventures.

Salesforce is no stranger to the field either when back in 2016, the company announced a new machine-learning artificial intelligence entity named “Salesforce Einstein” that was designed and developed to learn from human behavior in order to predict and personalize customer experience. In 2017, Salesforce continues to bet on machine learning in the hopes of bringing about a new era in Customer Relationship Management.

We hope you liked this article, you can read more articles about AI here.