Graph Mining / Complex Networks

Speaker Profiling in Conversation

Given a dialgue as input, extract the persona of the speakers involved in the conversation

Speaker persona can benefit applications like chat bots to curate personalised responses for a particular user which may further lead to better user engagement. However, personality information does not often accompany the speaker information. But, we observe that during a conversation a speaker often elaborates on her or someone else's qualities or personality information. In this work, we try to extract this personality information provided in a dialogue in a direct or indirect way to construct a persona for all the speakers involved in the chat.

Emotion-Flip Reasoning

Given a dialogue sequence, identify the emotions of the speakers and identify the reason/s behind an emotion flip of a speaker

While having a conversation, it becomes important to understand the emotion of the opposite party in order to converse more effectively. However, in order to avoid unwanted situations it is also important to identify the cause of those emotions as well. Detecting the reason behind emotions and their shift, sometimes, comes as a challenge even for human beings let alone for machines. In this work, we try to train a model which can not only identify the current emotion of a speaker but can also reason out the shift in the emotional state of a speaker.

Dialog System for Social Counselings

Building dialogue counseling for social couseling to give human like experience to users dealing with emotions, intents and various dialogues

Aiming to build a conversation bot that could perform counseling as a therapist virtually. The modular approach involves many modules in a pipeline fashion responsible for various tasks sucha as dialog act classification, intent classification, dialog management. The motto is to improvise the NLU and NLG while counseling.

Goal-Oriented Chat-Bot in Code-Mixed Language

Build a code mix laguage model based chatbot for bank conversations

We have several chatbots that can answer user queries on a specific platform, and methods have been proposed for modified versions too, but there isn't good support for codemix language(say, Hinglish). We aim to build a language model for a conversation system. The final product would be a chatbot for bank conversations to fulfill the user queries and perform actions as per user requests in every way possible. The requests (or utterances) can be in codemix language, i.e., Hinglish. Later these requests are classified into intents, and after the process of slot filling, the database query is made. This complete task is accomplished in the form of a friendly conversation between the user and the chatbot.