training an intent-entity prediction model using a set of historical intent-entity sequences automatically generated from a corpus of historical chat transcripts, wherein each historical intent-entity sequence in the set of historical intent-entity sequences is organized in a historical utterance order corresponding to a first ordered vector of historical user utterances in the corpus of historical chat transcripts;

extracting a second ordered vector of current user utterances from a current chat conversation between a user and a chatbot;

generating a set of current intent-entity sequences from the extracted second ordered vector of current user utterances, wherein each current intent-entity sequence in the generated set of current intent-entity sequences is organized in a current utterance order corresponding to the extracted second ordered vector of current user utterances;

determining, using the generated set of current intent-entity sequences on the trained intent-entity prediction model, a set of next predicted intents based on the determined set of next predicted intents including a corresponding set of highest confidence intents;

generating a set of prediction inputs by appending respective next predicted intents from the determined set of next predicted intents to the generated set of current intent-entity sequences;

predicting, using the generated set of prediction inputs on the trained intent-entity prediction model, a next intent-entity sequence based on the predicted next intent-entity sequence including an intent as a first token; and

converting the predicted next intent-entity sequence into at least one suggested question for the user to ask the chatbot, wherein the at least one suggested question is displayed to the user in a current chat window, and wherein the at least one suggested question displayed in the current chat window is selectable by the user for input in the current chat conversation as a next question from the user.