Scalable training of hierarchical topic models

Chen, Jianfei; Zhu, Jun; Lu, Jie; Liu, Shixia

doi:10.14778/3192965.3192972

Cited by 17 publications

(21 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The most commonly used topic model is LDA [3], [13], [14], which is represented as a probabilistic graphical model in Figure 1, and it has shown great success in various NLP tasks for discovering latent topics in the biomedical literature [15], [16]. However, topics are naturally organized in a hierarchy [4], and the above models are flat topic models that cannot capture the hierarchical information of the topics.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

A Top-Down Binary Hierarchical Topic Model for Biomedical Literature

Lin

Liu

2020

IEEE Access

View full text Add to dashboard Cite

Over the past two decades, a number of advances in topic modeling have produced sophisticated models that are capable of generating topic hierarchies. In particular, hierarchical Latent Dirichlet Allocation (hLDA) builds a topic tree based on the nested Chinese Restaurant Process (nCRP) or other sampling processes to generate a topic hierarchy that allows arbitrarily large branch structures and adaptive dataset growth. In addition, hierarchical topic models based on the latent tree model, such as Hierarchical Latent Tree Analysis (HLTA), have been developed over the last five years. However, these models do not work well in cases with millions of documents and hundreds of thousands of terms. In addition, the topic trees generated by these models are always poorly interpretable, and the relationships among topics in different levels are relatively simple. The biomedical literature, including Medline abstracts, has large-scale documents in two major categories: biological laboratory research and medical clinical research. We propose a top-down binary hierarchical topic model (biHTM) for biomedical literature by iteratively applying a flat topic model and adaptively processing subtrees of the hierarchy. The biHTM topic hierarchy of complete Medline abstracts with more than 14 topic node levels shows good bimodality and interpretability. Compared to hLDA and HLTA, biHTM shows promising results in experiments assessed in terms of runtime and quality.

show abstract

Section: Related Workmentioning

confidence: 99%

“…These models build a topic hierarchy that allows for arbitrarily large branch structures and adaptive dataset growth. In addition, they have been successfully applied for document modeling, online advertising and microblog location prediction and outperformed flat topic models [4].…”

Section: Introductionmentioning

confidence: 99%

A Top-Down Binary Hierarchical Topic Model for Biomedical Literature

Lin

Liu

2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Pachinko-allocation and hLDA are both focusing on the logic of topic hierarchy in real case. For example, the nested Chinese Restaurant Process used in hLDA refers to a topic model structure which is very close to the real case [13,42,43]. By taking the root node as the door of a Chinese restaurant, and its child nodes as tables, we may take each document as the customer, and figure out the path it follows.…”

Section: Relevant Research Of Topic Modelingmentioning

confidence: 99%

“…43.675232, -70.321317):3The final structure of the hierarchical range tree of the toy dataset is shown inFig. Anexample of 2-branching hierarchical tree with sub range trees as leaf nodes for the toy dataset…”

mentioning

confidence: 99%

Topic Ranger : a tool for topic exploration and analysis of spatio-temporal documents

Lü¹

View full text Add to dashboard Cite

“…Recentelly, researchers have proposed new approaches to effectively extract the hierarchical structure from text [94,17,112,22]. For example, Kim [47] and Titov [90] studied the problem by proposing a model that discovers a hierarchy from review data.…”

Section: Introductionmentioning

confidence: 99%

Structured sentiment analysis in social media

Almars¹

View full text Add to dashboard Cite

In recent years, social media platforms, such as Twitter and Webio, have become popular sources of information on the web. These platforms contain a wealth of valuable information about user opinions, user interests, events and more. People typically use these platforms to discuss different topics, share their opinions about them and engage in question-andanswer sessions. For example, regarding smartphones, users might discuss the main aspects of a smartphone, such as the overall design, battery capacity, screen size and camera. The natural hierarchical structure of those concepts is often hidden in social media. Discovering the hidden structure can helps users understand people' preference to a certain topic at different levels of granularity, and show the reasons why they prefer this topic. Over the past decade, research on hierarchical topic models has shown considerable progress. However, these studies may not always be directly applicable to social media due to the shortness and the shallow meaning of social media messages.There are three major challenges when dealing with social media texts. Firstly, compared with traditionally long texts, social media texts suffer from sparsity, and this issue may result in an incomprehensible and incorrect concept hierarchy. Secondly, social media contains useful information such as social opinions and information about users. Most existing methods perform a flat sentiment analysis on each extracted aspects independently, and ignore the concept hierarchy. In fact, we need to make the sentiment analysis finegrained in order to simultaneously extract the aspects and summarise people' opinions on those discovered aspects. Thirdly, the current models only discover the concept hierarchy ignoring the community structure of users. Maintaining the consistency of user's interest on several communities according to various topics and sentiment information is a challenging problem.In this thesis, the limitations of the existing work are addressed and effective solutions are proposed. First, in order to discover the hierarchical structure of social media content, a novel approach called the context coherence model (CCM) is proposed. It recursively top ii down: (1) organizes the concepts discussed by users in social media texts; and (2) identifies the hierarchical relations among concepts. In the CCM, a new measurement called context coherence is introduced that analyses words in social media texts and determines the similarities among them. Then, the hierarchical relationship between words is determined by recursively partitioning the whole corpus into smaller parts according to the similarity results. Finally, a merging operation is performed to find similar words, group them under the same topic and remove duplicated topics. The approach is evaluated on two real-world data sets. The experiments show that the proposed approach can effectively reveal the hidden structure in social media.Opinions are now reflected in social media on a wide range of topics: trends in pop music, fashion...

show abstract

Scalable training of hierarchical topic models

Cited by 17 publications

References 29 publications

A Top-Down Binary Hierarchical Topic Model for Biomedical Literature

A Top-Down Binary Hierarchical Topic Model for Biomedical Literature

Topic Ranger : a tool for topic exploration and analysis of spatio-temporal documents

Structured sentiment analysis in social media

Contact Info

Product

Resources

About