Community Topic: Topic Model Inference by Consecutive Word Community Discovery

Proceedings of the 29th International Conference on Computational Linguistics, pages 971–983 October 12–17, 2022. 971 Community Topic: Topic model inference by consecutive word community discovery Eric Austin and Osmar R. Zaïane University of Alberta Alberta Machine Intelligence Institute Edmonton, Alberta eaustin@ualberta.ca zaiane@ualberta.ca Christine Largeron Université Jean Monnet Hubert Curien Laboratory Saint-Etienne, France largeron@univ-st-etienne.fr Abstract We present our novel, hyperparameter-free topic modelling algorithm, Community Topic. Our algorithm is based on mining communi- ties from term co-occurrence networks. We empirically evaluate and compare Community Topic with Latent Dirichlet Allocation and the recently developed top2vec algorithm. We find that Community Topic runs faster than the com- petitors and produces topics that achieve higher coherence scores. Community Topic can dis- cover coherent topics at various scales. The net- work representation used by Community Topic results in a natural relationship between topics and a topic hierarchy. This allows sub- and super-topics to be found on demand. These features make Community Topic the ideal tool for downstream applications such as applied research and conversational agents. 1 Introduction Topic modelling discovers the themes and concepts of large collections of unstructured text documents. These topics can fulfill multiple roles. They can act as features for document classification and indices for information retrieval. However, one of the most important functions of these topics is to assist in the exploration and understanding of large corpora. Researchers in all fields and domains seek to better understand the main ideas and themes of document collections too large for a human to read and sum- marize. This requires topics that are interpretable and coherent to the human users. In more recent years, another new area has emerged where topics can provide a great deal of utility: conversational agents or “chat bots”. A conversational agent is a computer program that is able to carry on a conversation with a human. The conversation is an end in itself; the purpose of speaking with a conversational agent is to converse, to be entertained, to express emotion and be sup- ported. This goes well beyond asking Siri to set a timer. One key component of having an actual conversation with a human is the awareness and use of the topic of conversation. Work has been done on enriching the agent’s response using the detected topic (Dziri et al., 2019). However, more can be done with topics to improve a conversational agent given the right topic model. It can be used to detect and control topic drift in the conversation so that the agent’s responses make sense in context. If the user is engaged with the current topic, then the agent can stay on topic or detect sub-topics to focus the conversation. The agent can detect super-topics to broaden the range of conversation. The agent should be able to move to related topics or, if the user becomes bored or displeased, jump to dissimi- lar topics. This type of control over the flow of the conversation is crucial to human communication and is needed for human-computer interaction. The …

Content truncated to 500 words