Accès libre

Application of Artificial Intelligence-based Content Generation Technology in News Publishing

  
21 mars 2025
À propos de cet article

Citez
Télécharger la couverture

Figure 1.

Structure of personalized news title generation model
Structure of personalized news title generation model

Figure 2.

Comparison of different decoding strategies in CSTS dataset
Comparison of different decoding strategies in CSTS dataset

Figure 3.

Training loss value
Training loss value

Figure 4.

Generate the PR curve of different types of titles
Generate the PR curve of different types of titles

Comparison of Different Models on the CSTS Dataset

Methods Rouge-1 Rouge-2 Rouge-L
RNN 19.55 8.7 17.34
RNN-context 30.18 14.06 27.36
HG-News 32.33 15.03 26.22
ABS 34.29 21.34 31.12
LSTM+Point 37.9 24.9 35.13
NAML+HG 36.31 21.85 34.89
This method 39.01 25.41 37.43

Comparison of different models on the LCSTS dataset

Methods Rouge-1 Rouge-2 Rouge-L
RNN 6.1 2.77 5.67
RNN-context 10.8 7.3 10.72
HG-News 22.76 7.72 21.34
ABS 28.15 11.03 25.36
LSTM+Point 29.09 14.74 27.83
NAML+HG 31.18 12.23 27.53
This method 33.87 15.67 28.04

The title of the three indicators is generated

News category Precision Recall F1-score
Political time 0.9666 0.6776 0.8605
international 0.9671 0.7011 0.8626
society 0.9682 0.7505 0.8659
culture 0.9702 0.7611 0.8724
entertainment 0.9711 0.7745 0.8756
health 0.9741 0.7856 0.8789