ADAPT Researcher Has Paper Accepted in Natural Language Computing Journal

ADAPT Researcher Haider Khalid Has Research Paper Accepted in Natural Language Computing Journal

The paper proposes a novel approach to topic detection in conversational dialogue was accepted into the Natural Language Computing Journal.

SFI ADAPT Centre researcher, Haider Khalid, has had his paper accepted to the International Journal of Natural Language Computing (IJNLC). The research paper titled: ‘Comparative Analysis of Existing and a Novel Approach to Topic Detection on Conversational Dialogue Data’, was co-authored by ADAPT’s Director, Professor Vincent Wade.

The paper proposed unsupervised and semi-supervised techniques for topic detection in conversational dialogue compared with existing topic detection techniques. Topic detection focuses on the ability of the device such as a smart speaker, to understand a conversation topic with a human. The novel approach for topic detection, takes pre-processed data as an input and performs similarity analysis with the TF-IDF scores bag of words technique (BOW) to identify higher frequency words like ‘and’, ‘she’, ‘you’, ‘it’, ‘the’, ‘was’, etc. which are refined by integrating the clustering and elbow methods and using the Parallel Latent Dirichlet Allocation (PLDA) model to detect the topics.

The experimental results show that the proposed topic detection approach performs significantly better using a semi-supervised dialogue dataset which, when compared with traditional topic detection approaches, performs better in all evaluation metrics.

To read or find out more about this research paper, click here >


Helpful terminology:
Topic Detection: in a Natural Language Context focuses on an agent’s (Chat-bot, Smart speaker, etc.) ability to understand the topic in a conversation with a human.
Conversational Dialogue: in this context denotes the ability for agents to have a conversation that is not preset and can shift and change topics to emulate the complexity of human conversation.