[Purpose/Significance] In the era of big data, scientific and technological texts exhibit characteristics such as multi-source, multi-type, and multi-structured. Designing topic detection methods to accurately organize scientific and technological information from massive texts has become a supportive task for formulating scientific and technological development strategies, optimizing resource allocation, and promoting scientific and technological innovation. [Method/Process] This paper proposes a neural network model, NNMMFF, based on multi-view feature fusion. The model utilizes a self-supervised training framework to extract feature vectors of scientific and technological texts from three perspectives: network, statistical, and contextual. It uses an embedded “multi-view feature fusion module” to integrate the three feature vectors and obtain fused text features. Finally, the effectiveness of this approach is validated through the topic detection task. [Result/Conclusion] Experiments are conducted on a Chinese dataset from the “National Security” domain (2015–2023). The results show that in NNMMFF, a deep feature-level fusion algorithm using gated multimodal units can effectively fuse feature vectors from the network, statistical, and contextual perspectives. In contrast, shallow feature-level fusion algorithms, such as feature concatenation and feature weighting, perform relatively poorly. The findings suggest a clear performance hierarchy for topic detection in multi-source scientific and technological texts: three-view fusion outperforms two-view fusion, which in turn outperforms a single view.
[1] BLONDEL V D, GUILLAUME J L, LAMBIOTTE R, et al. Fast unfolding of communities in large networks[EB/OL]. [2025-07-03]. https://arxiv.org/abs/0803.0476.
[2] HUMMON N P, DEREIAN P. Connectivity in a citation network: the development of DNA theory[J]. Social networks, 1989, 11(1): 39-63.
[3] TU Y N, HSU S L. Constructing conceptual trajectory maps to trace the development of research fields[J]. Journal of the Association for Information Science and Technology, 2016, 67(8): 2016-2031.
[4] DEVLIN J, CHANG M W, LEE K, et al. BERT: pretraining of deep bidirectional transformers for language understanding[EB/OL]. [2025-07-03]. https://arxiv.org/abs/1810.04805.
[5] MORRIS S A, VAN D V, MARTENS B. Mapping research specialties[J]. Annual review of information science and technology, 2008, 42(1): 213-295.
[6] 张子振, 储煜桂, 吴小兰. 基于LDA的多源文献主题及其差异研究——以“机器学习”为例[J]. 情报科学, 2019, 37(6): 108-112, 150. (ZHANG Z Z, CHU Y G, WU X L. Multi-source literature topics based on LDA and their differences: taking “machine learning” as an example[J]. Information science, 2019, 37(6): 108-112, 150.)
[7] LIU Y, XU S H. A local context-aware LDA model for topic modeling in a document network[J]. Journal of the Association for Information Science and Technology, 2017, 68(6): 1429-1448.
[8] HU L M, LI C, SHI C, et al. Graph neural news recommendation with long-term and short-term interest modeling[J]. Information processing & management, 2020, 57(2): 102142.
[9] 刘非凡, 张爽, 罗双玲, 等. 基于深度图神经网络方法的领域知识结构探测[J]. 情报学报, 2021, 40(11): 1209-1220. (LIU F F, ZHUANG S, LUO S L. Detection of scientific knowledge structure based on graph representation learning[J]. Journal of the China Society for Scientific and Technical Information, 2021, 40(11): 1209-1220.)
[10] 裘惠麟, 邵波. 多源数据环境下科研热点识别方法研究[J]. 图书情报工作, 2020, 64(5): 78-88. (QIU H L, SHAO B. Research on identification methods of scientific research hotspots under multi-source data[J]. Library and information service, 2020, 64(5): 78-88.)
[11] 颜端武, 梅喜瑞, 杨雄飞, 等. 基于主题模型和词向量融合的微博文本主题聚类研究[J]. 现代情报, 2021, 41(10): 67-74. (YAN D W, MEI X R, YANG X F, et al. Research on microblog text topic clustering based on the fusion of topic model and word embedding[J]. Journal of modern information, 2021, 41(10): 67-74.)
[12] 陈一帆, 张志强, 丁敬达, 等. 图书情报领域多源数据特征级融合方法研究综述[J]. 图书情报工作, 2024, 68(18): 134-146. (CHEN Y F, ZHANG Z Q, DING J D, et al. A review of the multi-source data feature-level fusion method in the field of library and information science[J]. Library and information service, 2024, 68(18): 134-146.)
[13] ABDI A, SHAMSUDDIN S M, HASAN S, et al. Deep learning-based sentiment classification of evaluative text based on multi-feature fusion[J]. Information processing & management, 2019, 56(4): 1245-1259.
[14] 曾子明, 张瑜, 李婷婷. 多特征融合的突发公共卫生事件潜在谣言传播者识别[J]. 图书情报工作, 2022, 66(13): 80-90. (ZENG Z M, ZHANG Y, LI T T. Detection of potential rumor spreaders in public health emergencies based on multi-feature fusion[J]. Library and information service, 2022, 66(13): 80-90.)
[15] RANI S, KUMAR M. Social media video summarization using multi-visual features and Kohnen’s self organizing map[J]. Information processing & management, 2020, 57(3): 102190.
[16] CHEN R J, LU M Y, WANG J, et al. Pathomic fusion: an integrated framework for fusing histopathology and genomic features for cancer diagnosis and prognosis[J]. IEEE Transactions on medical imaging, 2020, 41(4): 757-770.
[17] DENG Y, ZHANG W, LAM W, et al. User satisfaction estimation with sequential dialogue act modeling in goal-oriented conversational systems[C]// LAFOREST F, TRONCY R, SIMPERL E, et al. WWW’22: the web conference 2022. Lyon: Association for Computing Machinery, 2022:2998-3008.
[18] XUE J, WANG Y, TIAN Y, et al. Detecting fake news by exploring the consistency of multimodal data[J]. Information processing & management, 2021, 58(5): 102610.
[19] WU X H, MAO J, XIE H, et al. Identifying humanitarian information for emergency response by modeling the correlation and independence between text and images[J]. Information processing & management, 2022, 59(4): 102977.
[20] MOODY C E. Mixing Dirichlet topic models and word embedding to make LDA2vec[EB/OL]. [2025-07-03]. http://arxiv.org /abs/1605.02019．
[21] SUN Y, HAN J, YAN X, et al. PathSim: meta path-based top-k similarity search in heterogeneous information networks[J]. VLDB Endowment, 2011, 4(11): 992-1003.
[22] 飞桨. 基于PEGASUS实现中文文本摘要[EB/OL]. [2025-07-03]. https://aistudio.baidu.com/projectdetail/4903667. (PADDLEPADDLE. Implementing Chinese text summarization based on PEGASUS [EB/OL]. [2025-07-03]. https://aistudio.baidu.com/projectdetail/4903667.)
[23] FU X, ZHANG J, MENG Z, et al. MAGNN: metapath aggregated graph neural network for heterogeneous graph embedding[C]// HUANG Y, KING I, LIU T Y, et al. WWW’20: the web conference 2020. Taipei: Association for Computing Machinery, 2020: 2331-2341.
[24] PAPER WITH CODE. Node clustering on DBLP [EB/OL]. [2025-07-03]. https://paperswithcode.com/sota/node-clustering-on-dblp.
[25] BLEI D M, NG A Y, JORDAN M I. Latent Dirichlet allocation[J]. Journal of machine learning research, 2003, 3(1): 993-1022.
[26] AREVALO J, SOLORIO T, MONTES-Y-GÓMEZ M, et al. Gated multimodal units for information fusion[EB/OL]. [2025-07-03]. https://arxiv.org/abs/1702.01992.
[27] ZHEN L, HU P, WANG X, et al. Deep supervised cross-modal retrieval[C]// Proceedings of 2019 IEEE/CVF conference on computer vision and pattern recognition. Long Beach, 2019: 10386-10395.
[28] NAW N, HLAING E E. Relevant words extraction method for recommendation system[J]. Buletin teknik elektro dan informatika, 2013, 2(3): 293-299.
[29] GROOTENDORST M. BERTopic: neural topic modeling with a class-based TF-IDF procedure[EB/OL]. [2025-07-03]. https://arxiv.org/abs/1310.4546v1.
[30] MCINNES L, HEALY J, SAUL N, et al. UMAP: uniform manifold approximation and projection[J]. Journal of open source software, 2018, 3(29): 861.
[31] 习近平. 习近平：坚持总体国家安全观走中国特色国家安全道路[EB/OL]. [2025-07-03]. http://cpc.people.com.cn/n/2014/0415/c64094-24899781.html. (XI J P. Xi Jinping: adhere to the holistic view of national security and follow a national security path with Chinese characteristics[EB/OL]. [2025-07-03]. http://cpc.people.com.cn/n/2014/0415/c64094-24899781.html.)
[32] KVÅLSETH T O. On normalized mutual information: measure derivations and properties[J]. Entropy, 2017, 19(11): 631.
[33] 吴金红, 黄彩云, 李问秋, 等. 融合深度学习和链路预测的交叉性研究主题预测研究[J]. 情报杂志, 2022, 41(12): 57-63. (WU J H, HUANG C Y, LI W Q, et al. Research on predicting methodology for interdisciplinary research topics-based on deep learning and link prediction[J]. Journal of intelligence, 2022, 41(12): 57-63.)
