2021 2nd International Conference on Artificial Intelligence and Information Systems | 2021
Irony recognition combined with LDA and improved one-dimensional intra-attention model
Abstract
For short texts with ironic emotions, because the data is sparse and irony features are difficult to predict and extract, which causes the problem of substitution of irony text recognition accuracy, a more accurate irony recognition model is proposed. On the basis of the internal attention model, focus on the unique contradictory emotional vocabulary of ironic sentences; then, simultaneously input the sentences into the LDA model to obtain the maximum probability topic of the short text and use the Bi-LSTM model to obtain the two-way semantic dependence of the text; Finally, before the prediction layer, the above three are spliced for softmax classification. Compared with traditional irony recognition models such as LSTM, it has achieved better results on the Weibo comment data set and Ptacek data set.