[et_pb_section transparent_background=”off” allow_player_pause=”off” inner_shadow=”off” parallax=”off” parallax_method=”off” padding_mobile=”off” make_fullwidth=”off” use_custom_width=”off” width_unit=”on” make_equal=”off” use_custom_gutter=”off” background_color=”#ffffff”][et_pb_row make_fullwidth=”on” use_custom_width=”off” width_unit=”on” use_custom_gutter=”off” padding_mobile=”off” background_color=”rgba(168,168,168,0.79)” allow_player_pause=”off” parallax=”off” parallax_method=”off” make_equal=”off” background_color_1=”#ffffff”][et_pb_column type=”4_4″][et_pb_text admin_label=”Text” background_layout=”light” text_orientation=”left” use_border_color=”off” border_color=”#ffffff” border_style=”solid”]
Every day, humans type out more than 200 billion emails, hundreds of millions of tweets, and innumerable texts, chats, and private messages. No one person could pick through even a tiny sliver of this information and stitch together themes and trends—but computers are starting to be able to. For more than a decade, researchers have been developing computer programs that can ingest enormous amounts of writing to try and understand the emotions stirred up by an idea or a product.
The field—known as sentiment analysis—got its start in market research. As online reviews started to gather steam in the mid-2000s, companies who wanted to understand how their products—or their competitors’ offerings—were being received began to use algorithms to aggregate reviews, says Bing Liu, a professor of computer science at the University of Illinois, Chicago, who has written extensively about the history of sentiment analysis. The algorithmic approach could reveal broader insights than a focus groups or surveys, the thinking went.
Sentiment analysis has bloomed into a large and lucrative industry. Dozens of startups now focus exclusively on providing these services to other companies, Liu says, and many bigger tech corporations have developed their own software.
More recently, the corporate world has turned these same tools inward. Large companies like Accenture, Intel, IBM, and Twitter have started using the software to understand how their own employees feel about their jobs, and identify problems that might escape a harried supervisor during annual-review time.
Twitter, for example, hired a company called Kanjoya to analyze employees’ responses to regular surveys about their workplace experiences. The surveys used to be administered twice yearly, and included just one or two open-ended questions, The Wall Street Journal reported last year. But with Kanjoya, Twitter started sending the survey to one-sixth of its workforce every month—and increased the number of open-ended questions it asked. Kanjoya’s analysis tools ran through the narrative answers, extracting patterns that were then shared with executives.
Other companies have focused on analyzing employee chatter outside of formal reviews or surveys. To catch grievances that might not surface in structured responses, or identify policies that are working, IBM has for years been scooping up employees’ posts and comments on the company’s internal social networking platform.
That platform, called Connections, is available to all of IBM’s 380,000 employees in 170 countries. It functions like Facebook, Dropbox, and Wikipedia bundled into one package, allowing employees to publish posts, comment on others’, or collaborate in smaller groups. (IBM also sells a version of Connections to other companies.) An internally developed sentiment-analysis tool called Social Pulse monitors posts and comments for trends and red flags.
Last year, IBM used the program to engage its employees in a revamp of its performance-review system. Its HR department set up a forum to solicit feedback on proposals for a new system, and received tens of thousands of responses. Instead of assigning a team of analysts to comb through the reams of feedback, IBM set Social Pulse loose on the data.
The software helped surface a widespread complaint: Employees were unhappy that their performances were graded on a curve. The company promptly did away with the method, says Sadat Shami, who manages IBM’s Center for Engagement & Social Analytics.
“Without our social listening capabilities, we wouldn’t have been able to surface that in time to make that decision,” says Shami. “What traditionally happens in a month or two, we did in real-time.”
Kanjoya, the company that helps Twitter interpret its employee-survey results, also offers a service that piggybacks on workplace social networks. A page on its website advertises that the company’s sentiment-analysis tools work with Yammer, a social network that was acquired for more than a billion dollars by Microsoft in 2012. (Microsoft announced this week that it’s combining Yammer with its Office 365 Groups service.) Kanjoya’s product offers “employee engagement tracking,” which promises to trace positive or negative emotions over time, and a search function that responds to queries with an analysis of the sentiment surrounding it.
Broadening the scope of data-gathering from surveys and reviews to social-media posts risks invading employees’ privacy. That’s why IBM restricts its data-mining to posts and comments that are shared with the entire company, and doesn’t touch emails, chats, or interactions in private groups. Kanjoya was not available for comment on its privacy policies.
(Federal insider threat-detection programs are experimenting with sentiment analysis, too, to suss out moles inside their own agencies. The National Geospatial-Intelligence Agency will award a contract to implement sentiment-analysis technology, NextGov reported earlier this month, the details of which are still vague.)
Sentiment analysis is far from a polished technology. “The computer’s understanding of natural language is still bad,” says Liu. “Accuracy is not very easy.” A research project that tested basic analysis tools on a trove of emails sent between developers of an open-source server software suite only had a maximum accuracy rate of 30 percent. (Interestingly, when two people tried to determine the emotions expressed in 50 emails, they could only agree on three-quarters of them, says Tourani Parastou, the main author of the research paper at Polytechnique Montréal.)
The human element still remains an important check on emotion-sensing algorithms. Even IBM’s 3-year-old Social Pulse software is bolstered by human eyes: A small team of analysts routinely examine the trends it identifies to make sure it got them right before sending them up the chain to management.
Advancements in machine learning continue to improve algorithms’ ability to understand human writing—but researchers are already looking for the next emotion-detection technology. A pair of computer scientists at Sathyabama University in India published a paper last year that proposed a new way of determining employees’ attitudes and well-being: facial scans. The system they created captures images of employees’ faces every time they enter the building to determine whether they’re happy, sad, depressed, or angry, with the intention of using that data to optimize productivity and employee performance.
[/et_pb_text][et_pb_image admin_label=”Image” src=”https://sharesource.com.au/wp-content/uploads/2016/02/sharesource-logo.jpg” show_in_lightbox=”off” url_new_window=”off” animation=”left” sticky=”off” align=”left” force_fullwidth=”off” always_center_on_mobile=”on” use_border_color=”off” border_color=”#ffffff” border_style=”solid”] [/et_pb_image][/et_pb_column][/et_pb_row][/et_pb_section]