Chinese Short-Text Classification Based on Topic Model with High-Frequency Feature Expansion

Research publications related to the novel coronavirus disease COVID-19 are rapidly growing in number. However, current online literature hubs, even with artificial intelligence, are inadequate for identifying the relative strength of research topics. Hence, we aimed to develop a comprehensive Latent Dirichlet Allocation (LDA) topic model using natural language processing (NLP) techniques, provide visualisations for temporal trends, and apply our methodology to improve existing online literature hubs.Using the search term “COVID”, abstracts were extracted from PubMed®, from January to July 2020 (N=16346). An LDA topic model was trained on 81% of abstracts. Weekly temporal trends were visualised as a heatmap on all abstracts. Then, we tested our methodology on over 23,000 abstracts gathered from January 2020 to September 2020 from LitCovid, a literature hub from the National Center for Biotechnology Information. We use our topic model to subdivide LitCovid’s eight categories into corresponding LDA topics.The optimised LDA topic model, created using PubMed® data, produced 25 comprehensive topics with no significant overlap. There were temporal changes for topics: prominence of “Mental Health” and “Socioeconomic Impact” increased, “Genome Sequence” decreased, and “Epidemiology” remained relatively constant. We identified inadequate representation of “Airborne Transmission Protection”. Importantly, research on masks and PPE is skewed towards clinical applications with a lack of population-based epidemiological research. Our methodology, when applied to LitCovid, identified important topics within each LitCovid category. For example, “Case Report” was split into topics such as “Pulmonary” and “Oncology” as well as the under-represented topics “Haematology” and “Gastroenterology”. Our work allows for comprehensive topic identification and intuitive visualisation of temporal trends in COVID-19 research. Implementation of the methodology complements existing online literature hubs and identifies underrepresented topics such as population-based studies on masks that may be of significant public interest.Funding Statement: None to declare.Declaration of Interests: There are no conflicts of interest.

To derive latent topics from free-text responses on the negative impact of the pandemic on research activities and determine similarities and differences in the resulting themes between academic-based and clinical-based researchers. We performed a secondary analysis of free-text responses from a cross-sectional online survey conducted by the Japan Academy of Nursing Science of its members in early 2020. The participants were categorized into two groups by workplace (academic-based and clinical-based researchers). Latent Dirichlet allocation (LDA) topic modeling was used to extract latent topics statistically and list important keywords/text associated with the topics. After organizing similar topics by principal component analysis (PCA), we finally derived topic-associated themes by reading the keywords/texts and determining the similarity and differences of the themes between the two groups. A total of 201 respondents (163 academic-based and 38 clinical-based researchers) provided free-text responses. LDA identified eight and three latent topics for the academic-based and clinical-based researchers, respectively. While PCA re-grouped the eight topics derived from the former group into four themes, no merging of the topics from the latter group was performed resulting in three themes. The only theme common to the two groups was "barriers to conducting research," with the remaining themes differing between the groups. Using LDA topic modeling with PCA, we identified similarities and differences in the themes described in free-text responses about the negative impact of the pandemic between academic-based and clinical-based researchers. Measures to mitigate the negative impact of pandemics on nursing research may need to be tailored separately.

Despite the unique characteristics of urban forests, the motivating factors of urban forest visitors have not been clearly differentiated from other types of the forest resource. This study aims to identify the motivating factors of urban forest visitors, using latent Dirichlet allocation (LDA) topic modeling based on social big data. A total of 57,449 cases of social text data from social blogs containing the keyword “urban forest” were collected from Naver and Daum, the major search engines in South Korea. Then, 17,229 cases were excluded using morpheme analysis and stop word elimination; 40,110 cases were analyzed to identify the motivating factors of urban forest visitors through LDA topic modeling. Seven motivating factors—“Cafe-related Walk”, “Healing Trip”, “Daily Leisure”, “Family Trip”, “Wonderful View”, “Clean Space”, and “Exhibition and Photography”—were extracted; each contained five keywords. This study elucidates the role of forests as a place for healing, leisure, and daily exercise. The results suggest that efforts should be made toward developing various programs regarding the basic functionality of urban forests as a natural resource and a unique place to support a diversity of leisure and cultural activities.

This study employs the Latent Dirichlet Allocation (LDA) topic model to extract and analyze the evolution of factors influencing enterprise resilience. By applying LDA to 318 resilience-related literature, optimal topics were identified based on topic confusion scores and pyLDAvis visualization results. High-frequency keywords were used to identify key themes and summarize their evolutionary patterns. Four core themes were distilled regarding enterprise resilience mechanisms: “Digital Transformation Drives Corporate Resilience (Topic 1),” “Financial Policies Empower Corporate Resilience (Topic 2),” “Green Innovation Promotes Corporate Resilience (Topic 3),”, “Artificial Intelligence Factors Boost Industrial Chain Resilience (Topic 4),”. Among these, “Digital Transformation Drives Corporate Resilience (Topic 1),” represents a current hot research direction. This study reveals the patterns by which different key factors enhance corporate resilience, providing theoretical foundations for corporate strategy formulation.

With the emerging of massive short texts, e.g., social media posts and question titles from Q&A systems, discovering valuable information from them is increasingly significant for many real-world applications of content analysis. The family of topic modeling can effectively explore the hidden structures of documents through the assumptions of latent topics. However, due to the sparseness of short texts, the existing topic models, e.g., latent Dirichlet allocation, lose effectiveness on them. To this end, an effective solution, namely Dirichlet multinomial mixture (DMM), supposing that each short text is only associated with a single topic, indirectly enriches document-level word co-occurrences. However, DMM is sensitive to noisy words, where it often learns inaccurate topic representations at the document level. To address this problem, we extend DMM to a novel Laplacian Dirichlet Multinomial Mixture (LapDMM) topic model for short texts. The basic idea of LapDMM is to preserve local neighborhood structures of short texts, enabling to spread topical signals among neighboring documents, so as to modify the inaccurate topic representations. This is achieved by incorporating the variational manifold regularization into the variational objective of DMM, constraining the close short texts with similar variational topic representations. To find nearest neighbors of short texts, before model inference, we construct an offline document graph, where the distances of short texts can be computed by the word mover’s distance. We further develop an online version of LapDMM, namely Online LapDMM, to achieve inference speedup on massive short texts. Carrying this implications, we exploit the spirit of stochastic optimization with mini-batches and an up-to-date document graph that can efficiently find approximate nearest neighbors instead. To evaluate our models, we compare against the state-of-the-art short text topic models on several traditional tasks, i.e., topic quality, document clustering and classification. The empirical results demonstrate that our models achieve very significant performance gains over the baseline models.

Topic modeling extracts useful potential topics that reflect market information from massive financial news and is widely used in data mining and economic research. Traditional topic modeling approaches such as Latent Dirichlet Allocation (LDA) and Non-Negative Matrix Factorization (NMF) lack semantic information, and short texts have feature sparse problems. We develop a topic clustering model based on BERT-LDA joint embedding that takes both contextual semantics and thematic narrative into account. We cluster document embeddings with the HDBSCAN algorithm and utilize a class-based TF-IDF (c-TF-IDF) method to create topic representations. Empirical results show that the BERT-LDA model is competitive compared with traditional and single topic models. It generates coherent topic words that are dissimilar to each other.

Latent Dirichlet allocation (LDA) is a statistical model that is often used to discover topics or themes in a large collection of documents. In the LDA model, topics are modeled as discrete distributions over a finite vocabulary of words. The LDA is also a popular choice to model other datasets spanning a discrete domain, such as population genetics and social networks. However, in order to model data spanning a continuous domain with the LDA, discrete approximations of the data need to be made. These discrete approximations to continuous data can lead to loss of information and may not represent the true structure of the underlying data. We present an augmented version of the LDA topic model, where topics are represented using Gaussian mixture models (GMMs), which are multi-modal distributions spanning a continuous domain. This augmentation of the LDA topic model with Gaussian mixture topics is denoted by the GMM-LDA model. We use Gibbs sampling to infer model parameters. We demonstrate the utility of the GMM-LDA model by applying it to the problem of clustering sleep states in electroencephalography (EEG) data. Results are presented demonstrating superior clustering performance with our GMM-LDA algorithm compared to the standard LDA and other clustering algorithms.

This study utilizes LDA topic modeling to examine research trends related to COVID-19 within the field of social work, analyzing these trends through peer-reviewed articles from the Korea Citation Index database. Latent Dirichlet Allocation (LDA) topic modeling, a statistical method for discovering abstract topics within a collection of documents, is applied to categorize and summarize the thematic concentration of the literature. Five themes have emerged: healthcare service and digitalized methods, exploring mental health status, qualitative approaches to social service program responses to COVID-19, evaluation of care services, and public service and program responses to COVID-19. Key findings reveal that the Korean social work academia focused on digital-based non-face-to-face services, evaluating the adequacy of public social services, and analyzing mental health and caregiving services during the pandemic. These findings indicate a reassessment of social work practices in response to the pandemic, underscoring the need to explore the challenges and opportunities presented by varied national responses. Additionally, the application of Python and machine learning in this research has shown significant benefits for social work studies, enabling a deeper analysis of complex, big data and facilitating more informed decision-making.

While the widespread adoption of intelligent customer service systems has significantly boosted service efficiency, the persistent surge in complaints has revealed notable service failure challenges. This study addresses the shortcomings of current analytical approaches, which suffer from inefficiency and high subjectivity. To surmount these obstacles, we utilized the Latent Dirichlet Allocation (LDA) topic model to perform a thorough analysis of 1,247 genuine complaint texts. By means of dual assessment using the perplexity metric and coherence score, we determined that the optimal number of topics is three. We pinpointed three primary issues: functional failure, procedural failure, and emotional failure, which account for 42%, 35%, and 23% of the total, respectively. Our research uncovers that the root causes of intelligent customer service failures lie in the discrepancy between technological capabilities and user expectations, the clash between cost-cutting and efficiency-enhancement measures versus service quality, as well as a bias in system design that favors efficiency over a user-centric approach. Drawing on these insights, this paper puts forward tailored improvement strategies, including enriching the knowledge base, streamlining the process for escalating to human support, and integrating sentiment analysis capabilities. These strategies aim to furnish enterprises with a theoretical foundation and practical guidance for refining their intelligent customer service systems.

A wide range of web corpora are in the form of short text, such as QA queries, search queries and news titles. Entity linking for these short texts is quite important. Most of supervised approaches are not effective for short text entity linking. The training data for supervised approaches are not suitable for short text and insufficient for low-resourced languages. Previous unsupervised methods are incapable of handling the sparsity and noisy problem of short text. We try to solve the problem by mapping the sparse short text to a topic space. We notice that the concepts of entities have rich topic information and characterize entities in a very fine-grained granularity. Hence, we use the concepts of entities as topics to explicitly represent the context, which helps improve the performance of entity linking for short text. We leverage our linking approach to segment the short text semantically, and build a system for short entity text recognition and linking. Our entity linking approach exhibits the state-of-the-art performance on several datasets for the realistic short text entity linking problem.

In this article, first a comprehensive study of the impact of term weighting schemes on the topic modeling performance (i.e., LDA and DMM) on Arabic long and short texts is presented. We investigate six term weighting methods including Word count method (standard topic models), TFIDF, PMI, BDC, CLPB, and CEW. Moreover, we propose a novel combination term weighting scheme, namely, CmTLB. We utilize the mTFIDF that takes into account the missing terms and the number of the documents in which the term appears when calculating the term weight. For further robust term weight, we combine mTFIDF with two weighting methods. We evaluate CmTLB against the studied weighting schemes by the quality of the learned topics (topic visualization and topic coherence), classification, and clustering tasks. We applied weighting schemes to Latent Dirichlet allocation (LDA) and Dirichlet multinomial mixture (DMM) on eight Arabic long and short document datasets, respectively. The experiment results outline that appropriate weighting schemes can effectively improve topic modeling performance on Arabic texts. More importantly, our proposed CmTLB significantly outperforms the other weighting schemes. Secondly, we investigate whether the Arabic stemming process can improve topic modeling performance. We study the three approaches of Arabic stemming including root-based, stem-based, and statistical approaches. We also train topic models with weighting schemes on documents after applying four stemmers related to different stemming approaches. The results outline that applying the stemming process not only reduces the dimensionality of term-document matrix leading to fast estimation process, but also show enhancement of topic modeling performance both on short and long Arabic documents. Moreover, Farasa stemmer achieves the highest performance in most cases, since it prevents the ambiguity that may happen because of the blind removal of the affixes such as in root-based or stem-based stemmers.

With the rapid development of social media and e-commerce on the Internet, many short texts such as instant messages, tweets and product comments have become an important form of Internet information. Compared with regular texts, short texts suffer severe data sparsity and informal format. Because of that, mining latent semantics such as topics in short texts is a critical challenge. There are many limitations directly applying common topic models like Latent Dirichlet Allocation (LDA) to short texts. In result of that, many short text topic models were put forward. The Word Network Topic Model (WNTM) is the typical one among them, which transfers sparse document to word space into dense word to word space and then learn topics from it. However, WNTM has its own limitation. The word co-occurrence matrix of the original document is used to construct the word network. This method is simple and intuitive, which can't express the deep meaning between words and words. To solve this problem, we proposed W2V-WNTM, which uses Word2Vec model instead of word co–occurrence to construct word network. Experiment shows it can describe relationship between words better.

Boys’ love (BL) fiction has flourished in China since the beginning of the twenty-first century. It primarily focuses on the romance between men, while most authors and readers of BL fiction are heterosexual women; thus it is paradoxically patriarchal and feminist. This study aims to explore two main questions: (1) What topics do Chinese BL fiction authors prefer? (2) How do the gender concepts of Chinese culture affect the topics and contents of BL fiction? Adopting machine learning methods––the Latent Dirichlet Allocation (LDA) topic model and word vector technology—this study presents the typical topics and specific contents of 72,548 original BL fiction pieces from synchronic and diachronic dimensions. The findings reveal that Chinese BL fiction extensively uses Chinese myths and history to create heterosexually structured fantasies. Furthermore, under the Chinese patriarchal culture, primarily the clan concept, BL authors attach great importance to marriage and reproduction issues. Such discourses eventually limit Chinese BL women authors’ awakening of a feminist consciousness.

We present a new way called Persona Analysis with Text Topic Modelling (PATTM), which tries to learn the role of personae according to the literal descriptions. It is similar to Latent Dirichlet Allocation (LDA) and Author Topic (AT) model, with the attribute of allowing all text to join in the topic modelling process, even when there is no persona in the text. We experiment on the “Libya Event” data set which contains more than 4,000 texts collected from the Internet. The PATTM gives lower perplexity than LDA and AT model on the data set.

In this paper, we propose a method for short text categorization using topic model and integrated classifier. To enrich the representation of short text, the Latent Dirichlet Allocation (LDA) model is used to extract latent topic information. While for classification, we combine two classifiers for achieving high reliability. Particularly, we train LDA models with variable number of topics using the Wikipedia corpus as external knowledge base, and extend labeled Web snippets by potential topics extracted by LDA. Then, the enriched representation of snippets are used to learn Maximum Entropy (MaxEnt) and support vector machine (SVM) classifiers separately. Finally, viewing that the most possible predicted result will appear in the top two candidates selected by MaxEnt classifier, we develop a novel scheme that if the gap between these candidates is large enough, the predicted result is considered to be reliable; otherwise, the SVM classifier will be integrated with MaxEnt classifier to make a comprehensive prediction. Experimental results show that our framework is effective and can outperform the state-of-the-art techniques.