A graph convolutional topic model for short and noisy text streams

Learning hidden topics from data streams has become absolutely necessary but posed challenging problems such as concept drift well short and noisy data. Using prior knowledge to enrich a topic model is one of potential solutions cope with these challenges. Prior that derived human (e.g. Wordnet) or pre-trained Word2vec) very valuable useful help models work better. However, in streaming environment where arrives continually infinitely, existing studies are limited exploiting resources effectively. Especially, graph, contains meaningful word relations, ignored. In this paper, aim at graph effectively, we propose novel convolutional (GCTM) which integrates networks (GCN) into learning method learns the simultaneously for streams. each minibatch, our not only can exploit an external also balance old perform on new We conduct extensive experiments evaluate both (Wordnet) built embeddings (Word2vec). The experimental results show achieves significantly better performances than state-of-the-art baselines terms probabilistic predictive measure coherence. particular, when dealing texts drift.