[Purpose/Significance] This paper explores the construction of an independent prediction model based on RoBERTa and fusion prediction models combining ensemble tree model and RoBERTa for public libraries to carry out book acquisition based on Chinese book subscription bibliography, in order to break through the current weak research on intelligent book selection models in public libraries. [Method/Process] According to the RoBERTa-based independent prediction model, the article built a bibliographic texts classification model architecture, focused on analyzing and determining the data selection and combination of input text, and formed a plan for model fine-tuning and prediction. Based on fusion models, it analyzed the advantages and strategies of model, and constructed a feature system consisting of fifteen variables with three different types of features. With the goal of improving recall rate, Bayesian optimization method was used for hyperparameter optimization and model construction to better adapt to the actual needs of book acquisition in public libraries. [Result/Conclusion] The experimental results show that both the RoBERTa-based independent prediction model and the fusion prediction models can achieve good performance in book selection and classification prediction, with AUC values and accuracy exceeding 0.94 and 0.88. By examining the F1 scores and F2 scores, the fusion models can further optimize the overall prediction performance and significantly improve the recall rate. The predicted probability values of book purchase formed by model operation, as well as the four acquisition strategies which are recommended purchase, appropriate purchase, cautious purchase and non-recommended purchase, can provide effective decision support and reference for public libraries to carry out intelligent book acquisition of Chinese book subscription bibliography.
[1] 张鹏彬,吴羽冰.广东省立中山图书馆联合新书采选系统的构建与设计[J].图书馆论坛, 2022, 42(10):115-120.(ZHANG P B, WU Y B. Construction and design of joint selection and acquisition system of new books at Sun Yat-sen Library of Guangdong Province[J]. Library tribune, 2022, 42(10):115-120.)
[2] 钟建法,孟子正.基于RoBERTa和LightGBM的中文图书采选模型研究[J].大学图书馆学报, 2025, 43(1):82-92.(ZHONG J F, MENG Z Z. Research on Chinese book acquisition model based on RoBERTa and LightGBM[J]. Journal of academic libraries, 2025, 43(1):82-92.)
[3] 黄小华,苗松,施化吉,等.基于改进遗传神经网络的图书采购模型研究[J].现代情报, 2009, 29(9):162-165.(HUANG X H, MIAO S, SHI H J, et al. Research on book purchase model based on improved genetic neural network[J]. Journal of modern information, 2009, 29(9):162-165.)
[4] 卞丽琴,陈峰.基于人工智能的图书订购策略分析[J].图书馆杂志, 2015, 34(8):39-43.(BIAN L Q, CHEN F. Analysis of book-ordering strategy based on artificial intelligence[J]. Library journal, 2015, 34(8):39-43.)
[5] 周志强.基于混合智能算法的高校图书馆图书采购模型研究[D].呼和浩特:内蒙古大学, 2019.(ZHOU Z Q. Research on book purchasing model of the university library based on hybrid intelligent algorithms[D]. Hohhot:Inner Mongolia University, 2019.)
[6] 傅立云,胡芸.基于随机森林的图书采购决策模型[J].情报探索, 2020(5):34-39.(FU L Y, HU Y. The model of book purchasing decision-making based on random forest[J]. Information research, 2020(5):34-39.)
[7] 鞠静.基于决策树算法的高校图书馆图书采访决策模型研究[D].保定:河北大学, 2021.(JU J. Research on the book acquisition decision model of university library based on decision tree algorithm[D]. Baoding:Hebei University, 2021.)
[8] 王红,雷菊霞.人工智能图书采访模式设计及流程运维[J].图书馆学研究, 2018(5):71-76.(WANG H, LEI J X. Design and process operation of artificial intelligence book interview mode[J]. Research on library science, 2018(5):71-76.)
[9] 原小玲,王红.人工智能图书采访系统的学习训练集构建与维护[J].晋图学刊, 2020(3):22-25.(YUAN X L, WANG H. Construction and maintenance of learning training set for artificial intelligence in book interview[J]. Shanxi library journal, 2020(3):22-25.)
[10] 涂佳琪,杨新涯,沈敏.需求与决策驱动的图书智能采访系统研究与实践——以重庆大学图书馆为例[J].图书情报工作, 2020, 64(11):28-34.(TU J Q, YANG X Y, SHEN M. Research and practice of decision-driven book intelligent interview system:taking Chongqing University as an example[J]. Library and information service, 2020, 64(11):28-34.)
[11] 刘倩倩,刘圣婴,刘炜.图书情报领域大模型的应用模式和数据治理[J].图书馆杂志, 2023, 42(12):22-35.(LIU Q Q, LIU S Y, LIU W. Data governance and application development of large language models in library and information services[J]. Library journal, 2023, 42(12):22-35.)
[12] 梅侠峰,吴晓鸰,黄泽民,等.融合RoBERTa的多尺度语义协同专利文本分类模型[J].计算机工程与科学, 2023, 45(5):903-910.(MEI X F, WU X L, HUANG Z M, et al. A multi-scale semantic collaborative patent text classification model based on RoBERTa[J]. Computer engineering & science, 2023, 45(5):903-910.)
[13] LIU Y, OTT M, GOYAL N, et al. RoBERTa:a robustly optimized Bert pretraining approach[EB/OL].[2025-02-12]. https://arxiv. org/abs/1907.11692.
[14] 卢小宾,张杨燚,杨冠灿,等.新兴技术识别中的不均衡分类研究——基于代价敏感的随机森林算法[J].情报学报, 2022, 41(10):1059-1070.(LU X B, ZHANG Y Y, YANG G C, et al. Imbalanced classification of emerging technologies identification:based on cost-sensitive random forest[J]. Journal of the China Society for Scientific and Technical Information, 2022, 41(10):1059-1070.)
[15] 王日花.基于XGBoost的图书馆混合推荐模型研究与应用——以中国传媒大学图书馆为例[J].河南图书馆学刊, 2022, 42(1):104-108.(WANG R H. Research and application of library hybrid recommendation model based on XGBoost:a case study of the library of Communication University of China[J]. The library journal of Henan, 2022, 42(1):104-108.)
[16] 刘爱琴,郭少鹏,张卓星.基于LDA模型融合Catboost算法的文本自动分类系统设计与实现[J].国家图书馆学刊, 2023, 32(5):84-92.(LIU A Q, GUO S P, ZHANG Z X. Design and implementation of text auto-classification system based on LDA model fusion Catboost algorithm[J]. Journal of the National Library of China, 2023, 32(5):84-92.)
[17] 钟建法,陈娟,李灿元,等.高校图书馆图书采访决策模型研究[J].大学图书馆学报, 2021, 39(5):38-47.(ZHONG J F, CHEN J, LI C Y, et al. Study on decision model of university library acquisition[J]. Journal of academic libraries, 2021, 39(5):38-47.)
[18] 钟建法,孟子正.数智时代学科图书核心出版社理论思辨与测定应用创新[J].大学图书馆学报, 2024, 42(5):78-87.(ZHONG J F, MENG Z Z. Theoretical reflection, determination and innovative application of subject-specific core publisher in the digital intelligence era[J]. Journal of academic libraries, 2024, 42(5):78-87.)
[19] 顾桐,许国良,李万林,等.基于集成LightGBM和贝叶斯优化策略的房价智能评估模型[J].计算机应用, 2020, 40(9):2762-2767.(GU T, XU G L, LI W L, et al. Intelligent house price evaluation model based on ensemble LightGBM and Bayesian optimization strategy[J]. Journal of computer applications, 2020, 40(9):2762-2767.)
[20] 李保安,张雨祺,李培峦.基于贝叶斯优化的SVM-Xgboost移动支付风险预警模型[J].河南科技大学学报(自然科学版), 2022, 43(5):50-56.(LI B A, ZHANG Y Q, LI P L. SVMXgboost mobile payment risk warning model based on Bayesian optimization[J]. Journal of Henan University of Science and Technology (natural science), 2022, 43(5):50-56.)
