What is sentiment analysis?
Sentiment analysis(also known as opinion mining or emotion artificial intelligence) is a natural language processing technique used to determine whether data is positive, negative, or neutral. It studies the mood and attitudes expressed in a written text (Heires, 2015). A sentiment analysis model analyzes a text string and categorizes it using one of the labels you specify; for example, you could analyze a tweet to see if it is positive or negative, or an email to see if it is pleased, annoyed, or sad.
Sentiment analysis is mostly used in marketing. Companies search social media for brand references and determine if customer opinions are favorable or bad.
Leading employers, on the other hand, approach their employees as internal customers in today's employee-centric environment. As a result, using sentiment analytics to engage your workers and improve your people results makes sense.
Employee sentiment analysis
Employee sentiment analysis sentiment analysis is often performed on text data to help businesses monitor customer feedback and to provide qualitative feedback that offers actionable insights. Despite being important organizational assets, employees are almost always overlooked by organizations when it comes to employee sentiment analysis. An Employee engagement survey does a great job of identifying whether employees are satisfied with their work or not, but it is only limited to the responses from questions given and can be hard to deal with if the surveys are in huge amounts.
Employee sentiment analysis aims to understand how employees are feeling and why they feel that way. Many forward-thinking businesses are using employee surveys on a regular basis to collect and analyze data using sentiment analysis which becomes a great source of constructive feedback. It can uncover the strengths and weaknesses of an organization and also detect positive and negative feelings towards a new change in the organization. This helps HR managers to make decisions that address employees' concerns. This creates a culture of open communication. However, some organizations use sentiment analysis technology to monitor employees.
Heires (2015) states that companies are increasingly improving focus on monitoring internal communications between employees to better understand employees' moods. Organizations are monitoring employee communication via company chat rooms, emails, customer call logs, etc. The idea behind this is to use the results to manage risks posed by employees and to reduce employee turnover rate. Unhappy employees could pose risks such as leaking company information, fraud, and money laundering, and so on.
By monitoring positive and negative sentiment, companies can gain a better understanding of your employees and move one step closer to creating and sustaining an engaged workforce.
The sentiment analysis process
The process of sentiment analysis can be summarized in the following steps:
- Data collection
- Cleaning the data
- Determining sentiment
- Reporting results
Most sentiment analysis applications have been based on data from Twitter, where tweets are extracted and analyzed. This works well for customers and media reviews for a company, however, for employees, this might be a challenge. Firstly, on social media, people do not always use their correct names. Secondly, employees are highly unlikely to tweet negative comments about an organization that they work for. Therefore, even if the correct usernames are found, the data may be biased towards high positive sentiment.
Heires (2015) mentions that organizations can make use of internal communications platforms that employees use to identify employee sentiment. Data can be collected from these platforms and used for sentiment analysis. Employee sentiment analysis can be done from the organizational level, departmental level and the individual level. The data collected for this purpose is usually in the form of text(e.g. Employee feedback from emails). It is possible to carry out a sentiment analysis on audio and image data but this is a growing field of research. For images that have captions on them, the captions could be extracted from them and then further processed as text data. However, this is difficult as the meaning of a caption could depend on the image from which it is taken. Such context cannot be extracted when only extracting captions. There has been interest in sentiment analysis of images with captions by the American company, Facebook, and other researchers.
For this article, we will focus on text data.
The data collected is text. Computers do not understand text data so it has to be simplified and changed into numbers. A field of Artificial Intelligence called Natural Language Processing (NLP) deals with the processing of human natural language. It is common practice in NLP to remove punctuation marks such as full stops, commas, apostrophes, etc. as they do not have a huge impact on the meaning of words. After this, unnecessary words, called stop words are also removed from the text. Stop words are words such as to, the, a, etc. as they do not have a huge impact on the meaning of the text as well. The result is a list of words without punctuation.
The next stage is changing all the words to lower case and extracting the stem of each word. After this, the next stage is changing the words to numbers. Multiple processes are used to achieve this. What is important to note is that each word can be transformed into a vector that represents it. The advantage of this is that similar words can have vectors that are similar or close to each other and the computer can understand that.
Various machine learning algorithms can be applied to determine the sentiment of a given text. A result is a number that is used to determine if the sentiment is a positive, neutral or negative. For example, a +1 for a positive sentiment, a 0 for a neutral sentiment and a -1 for a negative sentiment.
Reporting sentiment analysis results is best done using graphical visualizations. The previous article here discusses data visualization. It can be useful when deciding how to visualize the results. The simplest visualization is a pie chart with the categories - positive, neutral, and negative sentiment, showing the percentages of data in each category. Sentiment analysis tools can produce heatmaps showing positive and negative sentiments giving you a fairly accurate picture of employee opinions within the organization.
The results of the analysis can then be used to make decisions by the organization. Decisions can be:
- How to improve employee sentiment
- How to further analyze neutral data to obtain value from it, and so on.
Drawbacks of sentiment analysis
Context is important when analyzing natural language, and this is a huge challenge for sentiment analysis. Some comments can contain both negativity and positivity and it is challenging for computers to classify the text accordingly. Some comments could be an attempt at sarcasm but will end up being classified in the wrong category. The other biggest problem, especially in Africa is that our natural languages are still difficult to represent in a way computers understand. Most of the packages that transform text to vectors currently can only be applied to languages that are widely spoken such as English. There are very few packages for African languages. Therefore, for text in our language, it is not easy to determine the sentiment.
Best practices for employee sentiment analysis
Collecting employee communication data can be a good source of information but it makes employees feel that they are being watched. Therefore, they will not feel free to express themselves. A transparent process is needed through the process of surveys that employees can respond to. The best practices for finding out employee sentiment are:
- To survey genuine issues
- To make surveys short so that employees can answer without frustration
- To commit to privacy to encourage honest feedback
- To show results and actions that are taken as a result of the survey
Sentiment analysis tools
Many firms use employee sentiment analysis software that combines artificial intelligence and machine learning to automate the process of gathering and analyzing employee sentiment data on a wide scale.
Here are a few to consider for organization looking to implement sentiment analysis:
- Intellica.ai - offers machine learning technology to enable corporate intelligence in a variety of industries including natural language processing. The company provides the basic technology, allowing you to create a custom application based on its core architecture. This means you can create your own custom employee sentiment analysis software using it as the base.
- KeenCorp - One of the leading the leading sentiment analysis tools providers out there. It has features to map periodic employee sentiment trends and to prioritize management decisions based on the findings.
- MeaningCloud - The sentiment analysis tool from MeaningCloud is provided as a free API that can be linked into almost any software application. This gives you a lot of freedom because you can integrate MeaningCloud with your existing HR systems.
- Jive Insights - Jive is a cloud-based HR software company that focuses on workplace communication and collaboration. Positive, negative, and neutral postings are categorized, as well as day-by-day post frequency mapping.
- Lattice Sentiment Analysis - Performance management and employee engagement are two of Lattice's products. Lattice's AI-powered sentiment analysis tool may be integrated into either module and will continuously analyze open-ended comments to determine employee sentiment.
Tatenda Emma Matika is a Business Analytics Trainee at Industrial Psychology Consultants (Pvt) Ltd a management and human resources consulting firm.
Heires K., 2015. Sentiment Analysis: Are You Feeling Risky? Available at http://www.mediakat.com/120115%20RMM%20Sentiment%20Analysis.pdf
Dina, N.Z. and Juniarta, N., 2020. Aspect based Sentiment Analysis of Employee’s Review Experience. Journal of Information Systems Engineering and Business Intelligence, 6(1), pp.79-88.
Moniz, A. and de Jong, F., 2014, April. Sentiment analysis and the impact of employee satisfaction on firm earnings. In European Conference on Information Retrieval (pp. 519-527). Springer, Cham.
View Tatenda Emma Matika's full profile