Employee engagement is a critical component in any organization’s success. So far, companies have gauged the sentiment of their employees with costly surveys that can be deployed only a limited number of times throughout the year.
However, nowadays, initiatives that employees take are posted on internal social networks such as Microsoft Yammer, and views that employees have about their companies are posted on publicly available rating websites such as GlassDoor.
That is why, under the project Buzz@Work, Nokia Bell Labs developed new Natural Processing Language tools that can extract information from online posts with near-human-level understanding. They can extract psychological aspects of conversations that algorithms have hitherto found it difficult to capture, including: (1) the types of conversations employees are having, (2) whether employees are behaving as a cohesive group, (3) whether they are expressing empathy, or (4) whether they tend to suffer from work-related depressive symptoms.
More specifically, by leveraging recent advances in machine learning and natural language processing, Nokia Bell Labs researchers developed models that measure:
(1) The types of conversations employees are having. The researchers found that 10 distinct types of interaction extensively discussed in psychology are sufficient to account for the way most people conceptualize their relationships. These types are: Trust, Respect, Knowledge, Power, Support, Identity, Similarity, Fun, Romance, and Conflict. The researchers then built deep-learning tools that can extract expressions of these 10 dimensions from any conversational text. With these tools at hand, work-related social networks can automatically not only limit abusive language but also identify and promote opportunities for positive and meaningful social interactions.
(2) Whether employees are behaving as a cohesive group. These models are based on the psychological concept of Integrative Complexity. This refers to a person’s ability to bridge opposing points of view. Recent work in Social Psychology has found Integrative Complexity to be related to the person’s use of language (for example, to the use of figurative expressions like “on the other hand”). The general idea is that, by preferentially promoting content contributing to healthy dialogues, work-related social networks can automatically identify the employees who are able to synthesize points of view and, in so doing, can encourage the circulation of valuable information across the entire company.
(3) Whether employees are expressing empathy. By mining linguistic and semantic characteristics of the language markers expressed by a person’s textual responses, these empathy models can accurately predict each user’s level of situational empathy, which an immediate empathic response of a person to a triggering situation. Check Open Inc out. It makes it possible to explore the peaks and dips that employees of S&P 500 Companies experienced over the years.
(4) Whether employees tend to suffer from work-related depressive symptoms. The researchers developed Med-DL, a novel AI Deep Learning algorithm that mines large-scale social media data to reliably extract medical symptoms and diseases. The algorithm is based on state-of-the-art Recurrent Neural Networks, and can discover health mentions (e.g., workplace stress, fatigue) from the large-scale, noisy, and unstructured data present in work-related social networks.
By tracking the types of conversation employees are having on workplace social channels, these models can monitor the culture of a company in real time, measure the impact of corporate interventions on employee morale, and, ultimately, encourage an open dialogue between leaders and front-line employees.
