Research on the automatic generation method of telecom re-complaints report based on improved Transformer model
In the telecommunications industry, repeated customer complaints about unresolved or unsatisfactory issues are a common challenge. Manually generating re-investment reports is not only time-consuming and prone to subjectivity but also fails to meet enterprise demands for efficiency and consistency....
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Beijing Xintong Media Co., Ltd
2025-06-01
|
| Series: | Dianxin kexue |
| Subjects: | |
| Online Access: | http://www.telecomsci.com/zh/article/doi/10.11959/j.issn.1000-0801.2025110/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In the telecommunications industry, repeated customer complaints about unresolved or unsatisfactory issues are a common challenge. Manually generating re-investment reports is not only time-consuming and prone to subjectivity but also fails to meet enterprise demands for efficiency and consistency. To address this issue, an automatic report generation method based on an improved Transformer model was proposed. This method introduced emotion embedding, enabling the model to effectively capture dynamic emotional changes in customer interactions and better understand customer attitudes and demands during the dialogue. Additionally, the incorporation of customized position encoding enhanced the model’s ability to perceive complaint time series information, significantly improving the time logic and detailed completeness of the generated content. Experimental results demonstrate that the proposed model achieves BLEU (bilingual evaluation understudy) and ROUGE (recall-oriented understudy for gisting evaluation) scores of 0.352 and 0.482, respectively, outperforming the original Transformer and other baseline models. Moreover, compared to manual efforts, the proposed model improves work efficiency by 89%. The generated reports not only align more accurately with real-world requirements but also exhibit superior performance in semantic detail and time sequence consistency. |
|---|---|
| ISSN: | 1000-0801 |