Aiming at the problem that most existing models of art sentiment analysis only consider text encoding from the word level, this paper proposes a novel long and short-term memory network-based krill herd algorithm for explainable art sentiment analysis in interior decoration environment. Firstly, multi-scale convolution is used to capture local correlation of different granularity, so as to obtain more semantic information of different levels and form richer text representation. Then, a gating mechanism is introduced to control the path of sentiment information flowing to the aggregation layer. An improved krill swarm algorithm based on cosine control factor and Cauchy factor is proposed to solve the model. Finally, the full connection layer and argmax function are used to achieve sentiment classification. The experimental results show that compared with other advanced models, the proposed model can improve the accuracy of emotion classification by 2.3% and 0.8% respectively on two public data sets of IMDB and Yelp2014, and obtain the minimum root mean square error (RMSE).