A Label Distribution Topic Model for Multi-label Classification
Lin Liu, Lin Tang · 2019
At present, multi-label supervised topic model is a kind of effective multi-label classification model applied to various domain. However, due to the limitation of traditional label-topic correspondence in existing multi-label supervised topic model, there are still some aspects that need to be improved. This paper proposed a label distributed LDA model(LD-LDA) for providing more complete label description, which overcomes the disadvantage that labels can only be associated with a fixed hidden topic set or a set of non-overlapping hidden topics, so as to describe labels by the form of probability distribution of all hidden topics. The experimental results show that LD-LDA model has better prediction effect than comparative models in protein function prediction. Although the description of observable variables, parameters and hidden variables in LD-LDA model is based on the problem of protein function prediction, LD-LDA model is essentially a multi-label topic model, which is also applicable to various multi-label application scenarios.