Hierarchical Re-estimation of Topic Models for Measuring Topical Diversity

Abstract

A high degree of topical diversity is often considered to be an important characteristic of interesting text documents. A recent proposal for measuring topical diversity identifies three elements for assessing diversity: words, topics, and documents as collections of words. Topic models play a central role in this approach. Using standard topic models for measuring diversity of documents is suboptimal due to generality and impurity. General topics only include common information from a background corpus and are assigned to most of the documents in the collection. Impure topics contain words that are not related to the topic; impurity lowers the interpretability of topic models and impure topics are likely to get assigned to documents erroneously. We propose a hierarchical re-estimation approach for topic models to combat generality and impurity; the proposed approach operates at three levels: words, topics, and documents. Our re-estimation approach for measuring documents' topical diversity outperforms the state of the art on PubMed dataset which is commonly used for diversity experiments.

Do not blame it on the algorithm: an empirical assessment of multiple recommender systems and their impact on content diversity

“…We relied on standard LDA for building our topic models. However, very recently, Azarbonyad et al (2017) have proposed that hierarchical re-estimation of topic models can lead to better estimates of topic diversity. While they are interested in calculating the diversity within a document, we are more interested in differences between documents, future research should evaluate in how far their approach could improve ours.…”

Fusion-based methods for result diversification in web search

“…Instead of formulating user intents for a query as a flat list of subtopics, Hu et al (2015) presented hierarchical diversification models. Hierarchical topic models are estimated for measuring topical diversity of documents in Azarbonyad et al (2017). Wang et al (2016a) investigated methods of evaluating search result diversity using intent hierarchies Result diversification has also been investigated in various specific search applications such as image retrieval (Ionescu et al, 2016), historic entity or event search (Gupta and Berberich, 2016), medical records retrieval (Li et al, 2015), music recommendations (Schedl and Hauger, 2015), search in Twitter (Wang et al, 2016b), among others.…”

Impact of Text Diversity on Review Helpfulness: A Topic Modeling Approach

“…In recommender systems and search engine retrieval, diversified results are considered more informative and useful to help consumer mitigate information overload than highly homogeneous results (Boim et al, 2011;Drosou & Pitoura, 2012;Hurley & Zhang, 2011). In the text mining area, several studies have proposed approaches to measure topical diversity of documents, which indicates how diverse a text document is relative to other documents in a corpus (Azarbonyad et al, 2017;Bache et al, 2013;Derzinski & Rohanimanesh, 2014;Liang et al, 2014). Most of these approaches first extract topics from documents using topic modeling techniques and then estimate the diversity of documents using the extracted topics.…”