Let’s not forget that the little emotions are the great captains of our lives and we obey them without realizing it.
Premise
Creating a shopping assistant chatbot that will recognize the emotions of Millenials and will recommend books based on the 5 Basic Emotions.
Synopsis
While shopping online I noticed how difficult it is to make decisions without personalized advice or recommendation. During Covid-19 and lockdown, it became even more difficult to get personalized advice due to closed shops. People started to switch more and more to online shopping where a chatbot is waiting for them to answer questions and provide help. However, the experience feels very plain and passive, which makes you feel awkward. On average 70% of people shopping online express frustration when their experience is impersonal (Segment, 2017). So then, how can we personalize a conversation to be more life-like, intimate, and representative of human interaction? Through emotions.
Preface
Lack of digital personalization is considered the main cause of customers’ anxiety from a variety of options available and decreased probability of purchasing. Consumers appreciate the personalized experience and request personalization online, expecting retailers to find out smart technological solutions supporting individually targeted in-store approaches digitally (Adhi et al., 2019).
Creating a shopping assistant chatbot, which will provide personalized suggestions, is the way to support customers online. Chatbot use will enable effective automated interactions which can deliver a high-quality user experience as well as boost customer satisfaction and retention (Juniper Research Ltd., 2019).
One of the ways virtual service agents influence customer satisfaction is through social responses and feelings of personalization. AI-based chatbots have a feature to show human-like characteristics – friendliness and expertise which are considered crucial for providing personalized service. A study confirmed that chatbots with perceived human-likeness are more preferred by people when reaching out to customer support (Verhagen et al., 2014)
In order to understand whether emotions can increase personalization and help in making a purchasing decision, it was important to research the types of emotions and their role in people’s choices. The first step was to emotions in order to understand what are the types of emotions and what are the common ones. The next step was creating IBM Watson Tone Analyzer code in Python to analyze user input and detect emotion. Then the conversational part with empathy was created in IBM Watson Assistant.
Prototype and experiments
Research & 2nd iteration
How do Millenials buy books online?
Based on the feedback received, I decided to study the characteristics and preferences of my focus group and conduct research on the online book purchasing behavior of Millenials. My focus was on Millenials, those ages 18 to 35, as this is generation has been shaped and very much influenced by technology and connectedness, they spend 50% more time shopping online than their older counterparts (Harris, 2021). In addition to this fact, Millennials also read more than any other age group. Reading became a social thing for Millenials and most of them will not purchase a book without consulting social media/ website before. According to statistics, 8 out of 10 Millennials purchase a book after checking reviews, comments, videos. While choosing a book, cover and price have more influence on them than the author, and the feel of authenticity and strong emotional connection are very much valued (Brown, 2019).
Creating 2nd iteration
Based on the preferences of Millennials when purchasing books online, the chatbot was updated by focusing more on the problem-solving and trust aspects of it, creating a strong emotional connection, acknowledging the problem by being more empathetic and understandable. I also decided to add pictures of the books’ covers to the chatbot as this has more influence on Millenials when choosing books. The feedback of my friend was also taken into account when updating a chatbot such as lack of welcoming message.
[aesop_video src=”youtube” id=”BpsRDZx6IYc” align=”center” disable_for_mobile=”on” loop=”on” controls=”on” mute=”off” autoplay=”off” viewstart=”off” viewend=”off” show_subtitles=”off” revealfx=”off” overlay_revealfx=”off”]
Testing 2nd iteration
Based on the feedback of my friends, the second version of the chatbot felt more personal and emotional. Also, the covers of the book helped them to decide which book they prefer more. However, the updated version still needs a design that is critical to grab attention as well as the addition of open-ended questions to make the conversation more human-like. Those things can be improved with the next iteration using the conversation base already created.
Research & 3rd iteration
In recent years, the research interest in the usage of emotions and personalization in recommendation systems is increasing. Some research works were done on how user emotions help in content personalization (Bielozorov et al, 2019; Mazaheri et al, 2012). Moreover, emotional parts of the brain have a powerful influence on the choices made which is especially useful to process the volume of information available online (Morse, 2006).
In the industry of books, Amazon makes recommendations using ratings and purchasing history to recommend similar books (Rejoiner Inc., 2021). Goodreads applies recommendations based on books that similar users read and discussed by looking at users’ bookshelves (Chung, 2011). Those recommendations do not tend to consider the emotions the user wants to experience when reading which is an important factor in determining whether the user likes something or not (Ellis, 2019).
For the third iteration, the focus was put on the recognition of the emotions of users and empathetically reacting to the input. To do this IBM Tone Analyzer was used to recognize the emotion of the user and then empathetically react to it and recommend a book based on that emotion. IBM Tone Analyzer can detect 5 types of emotions from text such as joy, fear, sadness, disgust, and anger.
Creating 3rd iteration
As long as the chatbot itself can not recognize emotions, for the third iteration, the IBM Tone Analyzer was used to understand the emotions of the user input. The service uses an advanced machine learning model to do multi-label classification. The IBM Tone Analyzer requires little coding which can be easily done by following the documentation of the program. I added a snippet of code to Python to analyze the tone of the user input. It detected the user’s emotional tone and then empathetically reacted to it. By using Python function def it recommended the top 5 book titles the user can read in this emotional state. Below is an example of how Tone Analyzer detects and emotion from user input and then recommends books.
[aesop_image img=”https://cmdstudio.nl/mddd/wp-content/uploads/sites/58/2021/03/Screenshot-2021-06-11-at-20.38.12.png” panorama=”off” align=”center” lightbox=”on” captionsrc=”custom” captionposition=”left” revealfx=”off” overlay_revealfx=”off”]
In addition to the analysis of the user emotion in IBM Tone Analyzer, the chatbot in IBM Watson was created to show how it can react empathetically adapting to the emotional state of the user. The architecture of the Watson Assistant consists of three concepts: intents, entities, and dialog. It works by firstly manually inputting intents which are goals or topics the potential user will use; in this case, it was asking for the book recommendation.
The conversation in IBM Watson started with the introduction message what is Bookbot and how it can help in recommending books. Then the user could react whether he/she wanted to continue by typing “Hello, yes!”. After that chatbot tried to understand what is the current state of the user by asking general questions like “Please share how was your day?”, “How are you doing?”,” How do you feel today?”. The user input to one of these questions was then put into Python code at the back end for analyzing the real emotions of the user with the help of IBM Tone Analyzer. Below is the video of how the chatbot works.
[aesop_video src=”youtube” id=”DywErWYtgS4″ align=”center” disable_for_mobile=”on” loop=”on” controls=”on” mute=”off” autoplay=”off” viewstart=”off” viewend=”off” show_subtitles=”off” revealfx=”off” overlay_revealfx=”off”]
Testing 3rd iteration
Based on the feedback of my friends, the third version of the chatbot felt empathetic and emotional. In general, everyone liked the idea of implementing emotions for the book recommendation. They saw it as a good perspective for implementation on the book-selling websites. Also, it helped them to faster choose the book. However, as they mentioned it would be nice to have a longer conversation with a chatbot, where he asks more questions before defining emotions.
Conclusion
Before experimenting with a chatbot, I conducted extended research on the role of emotions in personalization and decision-making. Understanding the issues and needs helped me to structure the process, however, it took me a lot of time which decreased the time for the actual creation of prototypes. The main thing that I have learned is that it is possible to create an advanced chatbot without any coding in a short period of time. However, the chatbot created cannot be integrated with Tone Analyzer, which makes it impossible to create a fully working prototype. In order, the chatbot to actually recommend books based on emotions, the recommender system needs to be created in Python which will combine both emotion detection with the Machine Learning algorithm as well as the recommender algorithm.
References
Adhi, P., Burns, T., Davis, A., Lal, S., Mutell, B. (2020). A transformation in store. Retrieved from https://www.mckinsey.com/business-functions/operations/our-insights/a-transformation-in-store
Brown B. (2019). INFOGRAPHIC: The Surprising Reading Habits of Millennials. Retrieved from https://experteditor.com.au/blog/infographic-surprising-reading-habits-millennials/
Harris W. (2021). 15 Actionable Tips for Selling to Millenials: As Ecommerce Guide. Retrieved from https://www.sellbrite.com/blog/15-actionable-tips-for-selling-to-millennials-an-ecommerce-guide/
Juniper Research Ltd., (2019). Chatbot interactions in retail to reach 22 billion by 2023, as AI offers compelling new engagements. Retrieved from https://www.juniperresearch.com/press/press-releases/chatbot-interactions-retail-reach-22-billion-2023
Lim A. G. Y. (2020). The Big Five Personality Traits. Retrieved from https://www.simplypsychology.org/big-five-personality.html
Marr B. (2019). Why AI and Chatbots Need Personality. Retrieved from https://www.forbes.com/sites/bernardmarr/2019/08/02/why-ai-and-chatbots-need-personality/?sh=1fde2f9f14f8
McKenzie J. P. (2019). The Ultimate Book Genres List to Help Pick Your Next Page-Turner. Retrieved from https://www.oprahdaily.com/entertainment/books/a29576863/types-of-book-genres/
Mehra L. (2021). Chatbot personality preferences in Global South urban English speakers. Social Sciences & Humanities Open, 3(1). https://doi.org/10.1016/j.ssaho.2021.100131
Modi A., Sharma P. & Bhatt D. (2016). Various Approaches to Recognise Human Emotions. Computer Science & Electronics Journals, 7(1), pp. 173-176. https://doi.org/10.09592/ijcsc.2016.025
Opening the Book. (2021). Discover your next great read. Retrieved from https://www.whichbook.net
Segment. (2017). The 2017 State of the Personalization Report. Retrieved from http://grow.segment.com/Segment-2017-Personalization-Report.pdf
Verhagen, T., Nes, V. J., Feldberg, F., Dolen, V. W. (2014). Virtual Customer Service Agents: Using Social Presence and Personalization to Shape Online Service Encounters. Journal of Computer-Mediated Communication, 19(3). https://doi.org/10.1111/jcc4.12066