Language model construction method and device

The embodiment of the invention discloses a language model construction method and device, and the method comprises the steps: calling corpora from a database, the corpora comprising a plurality of business corpora with labels and a plurality of non-label corpora; respectively training the k first models by using the business corpus to obtain k first language models; respectively predicting each corpus by using k first language models to obtain k first prediction probability matrixes of each corpus; performing mean value calculation on the k first prediction probability matrixes of each corpus to obtain a second prediction probability matrix of each corpus; performing multiple rounds of training on the second model by using the corpus, determining a loss function value of the second model after each round of training according to the second prediction probability matrix, stopping training until a preset condition is met, and obtaining a second language model, wherein the second model after each round of train