Research on Keywords Extraction of Film Reviews Based on the KeyBERT Model
Qikang Huang · Transactions on Computer Science and Intelligent Systems Research · 2024
Current film and television platforms still struggle to effectively gather information for users and companies through public film reviews. There is a lack of applications or research on keyword extraction from film reviews. This study aims to evaluate the effectiveness and feasibility of the KeyBERT model for extracting keywords from film reviews. The precision and recall rate are utilized to assess the impact of model extraction. The test results indicate that the average precision and recall rate of film review extraction are 0.600 and 0.387, respectively, which are slightly lower than those of other types of text. Specifically, the precision rate and recall rate of plot description film reviews are 0.80 and 0.50, respectively, which are higher than the rates for multidimensional subjective analysis film reviews (0.40 and 0.33). Furthermore, they surpass the precision rate of 0.20 and the recall rate of 0.20 for subjective emotional expression type reviews. It is worth noting that the precision rates decrease from 0.80 to 0.20 as the number of words reviewed increases from 100 to 500 in subjective emotional expression type reviews. While the KeyBERT model is suitable for extracting keywords from movie reviews, it is essential to consider the classification of such reviews, the structural breakdown of lengthy text, and minimizing personal bias as much as possible.