othiru / emotionally-aware-chatbot-edaic

Undergraduate thesis using machine learning and Python

Emotionally Aware Chatbot EDAIC

Table of contents

• General info
• Author
• Supervisor
• Technologies
• Dataset
• Models
• Result
• Conclusion

General info

A chatbot is a software application that uses Artificial Intelligence and generates human-like conversations. In this research, EDAIC (Emotion Detector & Artificial Intellegent Chatbot) is a chatbot capable of textual interactions and detecting emotions from the text. It can work as an emotion detector. EDAIC will answer the question of a user after analyzing that question’s keywords. Many apps use this kind of chatbot in the FAQ and Help sections so that users can converse with assigned chatbots and find answers to their queries. Instead of interacting with real humans, users converse with a human-like chatbot that acts like an actual human. In this case, chatbots are very helpful that can assist in different applications and websites.

Artificial Intelligence is the latest technology. Creating a chatbot using this technology can make anyone’s professional profile stand out in this competitive era. As Python language is used to create this chatbot, learning this language can also enhance anyone’s professional profile.

EDAIC, a chatbot that uses Artificial Intelligence to detect emotions from text, is introduced in this research. Emotions refer to various types of consciousness or states of mind that are shown as feelings. They can be expressed through facial expressions, gestures, texts, and speeches. Emotion detection through chatbots also helps to understand humans more precisely while conversing with them. Emotions like Joy, Love, Surprise, Neutral, Sadness, Fear and Anger are considered while creating this chatbot. This chatbot will communicate with people using the help of natural language processing (NLP) and classification algorithms.

In this research, EDAIC is more likely to work as an emotion detector. By detecting emotions, it can understand the customers’ likes or dislikes for the products and services of a company more accurately. It can assist users with swift responses that will reduce time and carry out the service faster. Therefore, this chatbot helps to reduce the workload. It aids the business teams in communicating with customers to resolve their queries in a faster and optimized way.

Author

• Nowshin Rumali
• Amin Ahmed Toshib
• Rejone E Rasul Hridoy
• Mehedi Hasan Sami

Supervisor

• Mr. Md. Khairul Hasan

Technologies

Project is created with:

• Google Colab
• Pycharm

Language used in this project is:

• Python

Dataset

• Chatbot Dataset: There are 2064 data in this dataset and 3 columns: User text, Chatbot reply and intent. This dataset is merged from 2 datasets and contains 27 unique intents.
• Tweet Emotion Dataset: Tweet emotions from SemEval-2018 Affect in Tweets Distant Supervision Corpus (AIT-2018 Dataset) is used.This dataset has two columns content and sentiment. It has 25000 unique text classified with 7 emotions such as anger, love, surprise, fear, joy, sadness and Neutral.

Both Dataset are given in this repository

Models

Here we have used 7 machine learning models for both chatbot and emotion detector and the models are:

• Support Vector Machine (SVM)
• Logistic Regression
• Random Forest Classifier
• XGBoost Classifier
• Multinomial Naive Bayes
• Decision Tree Classifier
• Multilayer Perceptron (MLP)

Result

1. Chatbot: Accuracy of the seven models - Support Vector Machine (SVM), Logistic Regression, Random Forest Classifier, XGBoost Classifier, Multinomial Naive Bayes, Decision Tree Classifier and Multilayer Perceptron are 71.71%, 56.01%, 48.45%, 58.72%, 44.19%, 53.68% and 70.93% respectively.
2. Emotion Detector: Accuracy of the seven models - Support Vector Machine (SVM), Logistic Regression, Random Forest Classifier, XGBoost Classifier, Multinomial Naive Bayes, Decision Tree Classifier and Multilayer Perceptron are 88.47%, 86.12%, 66.87%, 88.84%, 71.59%, 75.75% and 85.31% respectively.

Conclusion

In our model, we have developed an application that has created an atmosphere where our chatbot (EDAIC) and a human can make a conversation. We have applied seven machine learning models to our chatbot. They are SVM, Logistic Regression, Random Forest, XGBoost, Naive Bayes, Decision Tree, and Multilayer Perceptron. We have generated the reply based on the best intent and cosine distance. At first, We have used ensemble learning to choose the best intent. After finding the best intent, we have used cosine distance to receive relatively suitable responses. Among all the models, SVM has given the most accuracy for chats that is 71.7%.

Detecting emotions from text is relatively complex. Because, in some cases, it is not easy to recognize appropriate emotions from the text. Again, we have used seven machine learning models for our emotion detector, and XGBoost gave the most accuracy that is 88.84%.

We have implemented an application combining two models - chatbot and emotion detector. Here, users can chat with EDAIC, and they can also see the emotions of their texts. There is a feature for rating emotions. So, one can easily give a rating to the emotion that has been predicted by the chatbot. Also, a user can choose the correct emotion if the predicted emotion seems incorrect to the user and gives a bad rating. If the predicted emotion is rated by a user, this feedback will be stored in a file. So, after collecting a certain amount of feedback we can re-train our model for better performance.