Research on Neural Network-based Text Sentiment Recognition and Analysis Model in Artificial Intelligence Translation

Aiming at the problems of sentiment semantic mismatch between the translated text and the source text and insufficient mining of fine-grained sentiment information in artificial intelligence translation, a Neural Network-based Text Sentiment Recognition and Analysis Model (NNSRA) is proposed.With Transformer as the basic architecture, this model integrates the Bidirectional Gated Recurrent Unit and attention mechanism to construct a multi-granularity sentiment feature extraction module, realizing the sentiment semantic representation at the word, sentence and discourse levels.By designing a sentiment feature fusion layer, it organically combines sentiment features of different granularities with contextual semantic features, and introduces a sentiment intensity adaptive adjustment mechanism to optimize the weight allocation of features.Experimental results on public sentiment datasets and translation parallel corpora show that compared with traditional models, the NNSRA model improves the accuracy and F1-score in sentiment classification tasks by 2.1-3.5 percentage points and 1.8-3.2percentage points respectively.In translation tasks, it increases the sentiment semantic matching degree of the translated text by 15.7% and the BLEU value by 1.2-1.58,effectively solving the problem of sentiment semantic distortion in artificial intelligence translation.