How well do you think you know your customers? Can you “read” their emotions and attitude towards your brand through written (and spoken) text?
Text analytics and sentiment analysis are typical methods that help derive meaning from customer data. Text analytics studies the topics and value of the words, including the grammar and the relationships among the words, giving you the “meaning”. Sentiment analysis builds on text mining and provides insight into the “emotion” or “sentiment” behind the words.
Since customers express their thoughts and feelings more than ever before, sentiment analysis became an essential tool to track and understand that sentiment. Sentiment analysis (or opinion mining) is a natural language processing technique performed on textual data to help businesses monitor brand and product sentiment in customer feedback and understand customer needs. Analysing customer feedback, such as opinions in survey responses and social media conversations, allows brands to learn what makes customers happy or frustrated to tailor products and services to meet their customers’ needs better.
While most companies use different online and offline communication channels to interact with their customers, some companies gather customer’s comments and feedback. They further analyse this valuable data and use it as a guide to improving their products and services. When extracting this data, businesses mostly apply text mining tools with a pre-coded list of words to identify topics and sentiment. This is also known as a lexicon-based method where the system labels a positive, negative, or neutral message based on some keywords.
Unlike text mining software currently available on the market, the SentEMO project aims to expand sentiment analysis even further. The idea is to develop a deep machine learning tool that will offer organisations insights based on text. With its multilingual feature (EN, NL, FR, DE) and customisation options, SentEMO is envisioned as a robust tool that will show businesses detailed insights retrieved from the aspect-based sentiment analysis (positive, negative, neutral) and emotion analysis (anger, fear, happy,…).
Launched by a group of researchers at Gent University (LT3 research group) and Artevelde University, SentEMO is based on fine-grained sentiment analysis. “When you look at what is in the market, you see that the systems currently being used are simple lexicon-based systems that are old-fashioned. In academic research, we are way ahead of what is currently being used in companies,” says Veronique Hoste, a Professor of Computational Linguistics at the Faculty of Arts and Philosophy at Ghent University and an expert in machine learning and natural language processing.
“Most of the systems used today are coarse-grained systems, meaning the system will do a basic detection “positive – negative – neutral”. Companies often want to go beyond this basic mood analysis and gain insights into what their stakeholders say about specific products, features, etc. This is where aspect-based sentiment analysis comes in: first, it determines what people are talking about, and then the sentiment or emotion towards this specific aspect is determined, leading to a much more comprehensive insight into your customers’ emotions. As different companies have different types of data, we also thought it would be interesting to build a system that could be used and customised by companies.”
The SentEMO project will run for two years, meaning the delivery of the trainable machine learning architecture with an annotation-interface is planned to be launched by the end of 2022.
The prototype is envisioned as a platform with two dashboards. One customisable dashboard will give insights into the “mood” of data and include various sentiment analyses. The goal is to start from a more holistic data view and further focus on different data aspects. A second dashboard will be an annotation dashboard that companies can develop themselves.
“Our goal is to give a methodology, a baseline system. We want to deliver the right tools for companies to tailor the system to their specific needs. As you feed the system constantly with new data, you create a robust and accurate tool designed for your needs,” says Veronique.
“During the project, the different prototypes are accessible as a web service for testing purposes. After the project is finalised, we aim to set up an agreement with interested partners to share the source code (the software is written in Python), so companies can integrate it themselves. Everybody works in a different environment, so this way, we made it easier to adapt the system to the needs and preferences of any organisation,” explains Veronique.
To turn this project into a successful venture, Veronique and her team need access to real, tangible data. However, getting data isn’t always the easiest step because companies are worried about security issues. Regarding data security and GDPR compliance, Veronique says SentEMO has a clear policy: “As researchers, we are not interested in who posts messages; we are only interested in what is being said and how. If you’re able to anonymise the data or pseudonymise the data, then there is no problem. With some companies, we sign the NDA, but other businesses easily split the data from the personal information, and then they can share the data without an NDA,” confirms Veronique.
As they were mapping out the project, the SentEMO team knew they needed external help and tap into the power of a diverse partner network. They decided to collaborate with three types of companies:
SentEMO relies on consulting partners to bring in a business perspective and explain what companies want and need. They close the gap between research and practice, as they are in direct contact with businesses, gathering the necessary data and providing SentEMO with client’s feedback and ideas. This enables SentEMO experts to expand their academic knowledge and create a true business-oriented platform. In return for their participation and data provided, the businesses receive actionable insights and access the SentEMO prototype.
“This project is interesting both for boobook and us because we can see how to integrate text mining and data analysis and gain insights from it. From the academic perspective, our focus is on accuracy, not so much on interpretation and what action points need to be done. This is where boobook’s expertise comes in, as they can take extensive data, analyse it, and translate it into actionable insights,” says Veronique.
Proactive collaboration is crucial when it comes to projects involving many partners and inputs. SentEMO knows they can fully confide in boobook. “I see boobook as a valuable partner throughout the whole project. The back-and-forth discussion is important as we evaluate together the provided data, what could be improved, and how to optimise the system,” says Veronique.
Considering machine translation is constantly evolving, it’s only a matter of time we’ll see more and more platforms like SentEMO. Still, Veronique emphasises the important role of human interpretation and analytical mindset.
“If you want to try text mining for your business, think about a problem where you think the text analysis or text mining would be an interesting approach. Then look if there are solutions on the market that can assist you or reach out to a team of experts, such as boobook, to guide you through the process,” concludes Veronique.
Interested to know more about the SentEMO project? Or you would like to apply and participate in the project with your business? Reach out to firstname.lastname@example.org!