Text-based comments have long been the best way for employers to capture the employee voice, and gain visibility into their sentiments.
Whether its comments on an engagement survey, or feedback provided on an employer review site, employee text comments offer valuable insight into where an organization needs to focus and take effective action to improve the overall employee experience.
Traditionally, companies have collected employee text data but haven’t looked at it analytically. Text data can be difficult to analyze and quantify and is often a massive drain on resources. This means that companies are missing out on the true goal of this feedback – identifying the stories and sentiments that deeply influences employee satisfaction, engagement, and retention.
In recent years, the value of text feedback has been firmly established and most employers have turned to different text analytics tools and methods to analyze their employee text comments.
Below, we outline the most popular comment analysis methods available, and some of the game-changing options that can better break through barriers to text analytics and achieve real sentiment analysis.
1. Keyword Analysis
When confronted by information, we automatically search for meaning and outlines. That’s why, when faced with a substantial amount of text data, one approach is to hunt for the most straight-forward nuggets of information – keywords.
Typically, keyword analysis focuses on highly qualitative terms, such as “good”, “bad” as well as rating how frequently they appear. This approach is supported by defining terms that serve as relative synonyms in relation to these (e.g. “great” for “good”, “awful” for “bad”). By cross referencing the qualitative (e.g. “good”) with quantity (e.g. “good” used 40 times in 50 assessments, as opposed to “bad” being used 5 times in the same number of assessments), certain conclusions can be drawn.
Keyword analysis can be used to tackle enormous tracts of text and extract basic findings. It can be particularly useful in identifying the prevalence of a particular term or brand name in social media over a defined period of time.
However, this method has inherent weakness for those seeking to accurately chart sentiment. It will generally return no results if a section of text does not use those exact terms. Additionally, the subtlety and flexibility of language means any contextual mention of these terms may not be correctly interpreted. For example, a sentence such as “I was badly underperforming until the company provided this particular training” being tagged purely for the use of the negative term “badly”. Likewise, linguistic devices such as irony or sarcasm can go completely undetected.
2. Text Analytics Using Fixed Dictionaries
The reason why this type of definition-based keyword analysis is so limited is primarily to do with stretching the underlying definitions beyond their scope.
When we want to understand something, we look it up in a dictionary. However, that does not mean that we can recognize shades of that concept, or precisely that thing when we encounter it in the wild, unless it exactly matches what the dictionary has outlined. This is the same shortcoming that dictionary-based text analysis comes up against when paired with freeform content.
What’s more, for a dictionary-based tool employed in comment analysis, every time will be the first time. The capacity to tag, learn, and build on recognizing qualitative information is simply not there.
Nevertheless, the power of a definition, concise and authoritative, combined with automation, has provided a foundation for text analysis for many years. It can concretely prove that something was said, but not exactly what was expressed. Just as a person may need access to context, knowledge of patterns, and rich experience to reliably identify something they have been presented with, so text analysis needs to take that step of being supported in each instance where sentiment is trying to be reliably identified.
A more effective approach is when this enriched capacity is employed towards sufficient amounts of content that reflect employee concerns and experiences.
3. Pattern Discovery / Machine Learning
New applications of Machine Learning (ML) is one way in which to overcome the definition-based barrier.
ML-enabled systems have been trained, through processing massive amounts of data, on how to recognize the definitions that dictionaries traditionally provided, but also to recognize the connective and contextual tissue that surrounds these concepts. It is a training process that builds towards an applied model that is deeply effective in terms of sorting textual data.
Once this learning process is completed, free-form text content can be processed to the point where it is rapidly categorized. This can be fine-tuned and built upon in a continual fashion, essentially providing the applied ‘memory’ that is required to increasingly recognize sentiment accurately.
This improved sorting power can be invaluable for an organization looking to rapidly parse feedback from a huge number of respondents. By cross-referencing enough data, a truly accurate picture of concerns and sentiment can be realized.
Supplied with enough data over time, predictive insights are even possible, giving the greatest possible intelligence to an organization looking to strike a path through uncertain times.
An Employee Experience Management (EXM) based machine learning platform like BlueML represents a major shift in capabilities for text analytics. BlueML uses millions of real-world employee comments, which allows the system to continuously recognize, reformulate, and reorganize the patterns it detects in comments.
BlueML•Employee engagement•Employee Experience Management•Feedback matters•