AI Translation Analysis Based on Text Features and Similarity Calculation

In the context of accelerated globalization, the communication barriers between different languages have gradually become one of the key factors restricting international cooperation and information sharing. In order to cope with the challenge of large-scale data processing, improve the effectiveness and quality of translation, innovative research combines the expansion convolution neural network, long attention mechanism and user embedding, and through the dependent graph convolution model similarity calculation, build a new based on text features and similarity calculation of artificial intelligence translation model. The outcomes revealed that the full model reaches 0.85,0.80,0.82 and 35.6 in accuracy, recall, F1 scores and BLEU scores, respectively. Ablation experiments showed that each component had a significant effect on the performance. For example, after removing the DCNN, the BLEU score drops to 33.2. In addition, when processing 200,000 words, the system runs for about 30 ms, superior to other systems. Therefore, the translation model not only improves the translation quality, but also shows good time efficiency when processing large-scale data sets.