By team Kilpatrick Digital
Gabriele Volpi | Head of Digital Technology
Supatchara Schuller | Big Data and Analytics Consultant
It is well known that the health emergency has accelerated the arrival of a re-shaped model of organization with most part of the employees working from home and then partially even after the medical emergency restrictions are loosen up. Plus, the high levels of uncertainty and fear, the sudden changes in routines and lifestyles that this unprecedented crisis has brought, is causing employees to feel helpless, desperate, unmotivated and unproductive. In this context, it’s increasingly difficult to monitor and improve employees’ engagement and well-being.
Fortunately, new technological tools like Sentiment Analysis together with the most advanced Artificial Intelligence techniques of Natural Languages Processing (NLP) come in handy.
Sentiment analysis is the type of emotional artificial intelligence used for processing natural language, computational linguistics, biometrics, and text analysis to quantify, identify, and extract systematically the sentiment beyond what is written. Sentiment Analysis regards indeed the interpretation and classification of emotions (positive, negative and neutral) within text data using text analysis techniques.
Therefore, sentiment analysis allows businesses to identify customer sentiment toward products, brands or services in online conversations and feedback. It also helps to understand how employees feel about the company and their working conditions (even when they are working remotely) enabling companies to develop strategies to improve retention and employees’ satisfaction, thus greatly improving productivity and business results.
The Training and Prediction Processes
In the first part of every Natural Language Processing project, the text to be analysed is tokenized or in single words, parsed, removing non useful parts (points, special characters, articles…) and then embedded into numerical vectors, each representing its tokenized words, which represents their features and the similarities between words (in the sense of topic domain, such as love-affect…).
As a supervised Classification Problem of Machine Learning, each phases of the sequence of action which compose the Sentiment Analysis is divided into a training and a prediction process.
In the training process, our model learns to associate a particular text to the corresponding tag based on the test samples used for training. Pairs of feature vectors and tags (e.g. positive, negative, or neutral) are fed into the machine learning algorithm to generate a model.
In the prediction process, the feature extractor is used to transform unseen text inputs into feature vectors. These feature vectors are then fed into the trained model, which uses a statistical model to predict the most appropriate tag (positive, negative, or neutral).
Before the development of Deep Learning Neural Network, several classical Machine Learning approaches with statistical model Naïve Bayes Classifier, Logistic Regression and Support Vector Machines were used.
In recent years, Neural Networks outperformed classical models in every step, allowing to take into account a large number of hidden features without applying a lot of Feature Engineering work. In particular, Convolutional Neural Networks (CNN) allow to take into account quite all local features words and sentences have, while more recently the application of Recurrent Neural Network (RNN), in particular of Long-Short Term Memory network (LSTM), allowed to obtain even better results taking into account long-distance dependency features, that reflect syntactic and semantic information.
Data Mining: Active and Passive Sentiment Analysis
Sentiment Analysis can be applied to every type of written text, so there could be a lot of possible sources of information to take data from.
Normally, for an Employee Sentiment Analysis, two different sets of data acquisition types are considered: an active one, which basically consists of an open question survey, and a passive one, which through the use of API collects and analyses internal text messages of employees such as emails or Intranet tickets. Both modalities have their advantages and disadvantages, in terms of privacy concerns, reality and continuity of data.
Benefits of Sentiment Analysis
Sentiment Analysis is well known to be used for Customer Retention strategies, analyzing feedbacks and providing a real Data Driven Strategy, but in recent years its application for internal use is becoming more and more interesting.
In fact, the use of Deep Learning technology for Sentiment Analysis can be very useful for managers and HR sector, which can understand and extract deep insights from employees’ feelings and concerns without processing manually a lot of surveys’ answers. Following these insights, target-specific actions can be driven optimally to the best effective actions. Moreover, knowing employees’ feeling is also helpful to better communicate new decisions, and if done together with an ONA (Organizational Network Analysis) it could help manage organizational ineffectiveness.
Success Cases: implementation of Sentiment Analysis
Orange is one of the world’s leading telecommunications operators. The group serves 264 million customers, including 204 million mobile customers and 20 million fixed broadband customers.
In early 2019 Orange began to look for a new social listening tool in order to continue their transformation into a truly data-driven organization. They wanted to have a
tool that could be used at all levels and empower employees to make data-driven decisions. They wanted a tool that was both accessible and easy to use but was also robust enough to provide comprehensive consumer insights into the global markets where Orange operates.
Orange started to use an external Sentiment Analysis tool to monitor reviews on different App Store platforms. In this way Orange can easily understand what people are saying about their applications, understand their feedback, and get instant insight into the sentiment of customers. They also can have instant alerts whenever there is a negative review so the right manager can deal with issue as soon as possible in order to protect the brand. Orange managed to train and deploy more than 800 employees to follow the customers’ sentiments, in order to drive a real Data Driven market strategy.
Tesco is a British multinational groceries and general merchandise retailer, which with an underlying profit of 3.4 billions pounds is the most profitable online grocery retailer in the world.
Some of the most vital Tesco’s objectives were to make the company a better place to shop and work via a more engaged workforce, and to gain a deeper understanding of what really mattered in all areas of the business (stores, distribution, Tesco.com, Tesco Bank).
For this reason, Tesco started an Employees’ Sentiment Analysis project called Listen and Fix. With the aim to also facilitate two-way communication between company and them, Tesco invited its employees to send in their thoughts by text or email, which meant a low barrier to participation. Feedback was then aggregated and categorized through functions, departments, topics. All these insights were very useful for Tesco’s Data Scientists team to identify key issues and then take actions accordingly.
At Kilpatrick, we have realized that Sentiment Analysis is a unique instrument, useful to have meaningful insights for company’s Data Driven strategies, both for Customer and Employees purpose. As a resilient company, and due to our expertise in People and HR Tech we can assure that investing in digitalization processes with frameworks like this could be essential for companies to survive to present and future challenges.
Ramesh Wadawadagi, Veerappa Pagi, “Sentiment analysis with deep neural networks: comparative study and performance assessment”, Springer Nature B.V. 2020
Li Deng, Yang Liu, “Deep Learning in Natural Language Processing”, Springer
Kilpatrick Digital Know How