Network-Based Pooling for Topic Modeling on Microblog Content

Abstract

Topic modeling with tweets is difficult due to the short and informal nature of the texts. Tweet-pooling (aggregation of tweets into longer documents prior to training) has been shown to improve model outputs, but performance varies depending on the pooling scheme and data set used. Here we investigate a new tweet-pooling method based on network structures associated with Twitter content. Using a standard formulation of the well-known Latent Dirichlet Allocation (LDA) topic model, we trained various models using different tweet-pooling schemes on three diverse Twitter datasets. Tweet-pooling schemes were created based on mention/reply relationships between tweets and Twitter users, with several (non-networked) established methods also tested as a comparison. Results show that pooling tweets using network information gives better topic coherence and clustering performance than other pooling schemes, on the majority of datasets tested. Our findings contribute to an improved methodology for topic modeling with Twitter content.

Community pooling: LDA topic modeling in Twitter

“…In contrast to Conversation pooling, only direct replies to an original tweet were considered since a conversation could shift its topic in time. This pooling scheme showed better results than the previous methods (Ollagnier & Williams, 2019).…”

“…In this section we showed and discussed the results of the evaluation tasks described in the previous section. For each pooling scheme, we replicated the training workflow used in the literature and used an LDA model with 10 topics (Mehrotra et al, 2013;Ollagnier & Williams, 2019).…”

“…Each evaluation measured the performance of the pooling method differently. In order to present a complete and exhaustive analysis of previous schemes and the proposed pooling method, in this work we evaluated the proposed methodology using topic clustering metrics (Purity and Normalized Mutual Information) (Alvarez-Melis & Saveski, 2016;Hajjem & Latiri, 2017;Mehrotra et al, 2013;Quezada & Poblete, 2019;Ollagnier & Williams, 2019;Akhtar & Beg, 2019), a supervised machine learning classification task (Giorgi et al, 2018;Hong & Davison, 2010) and a document retrieval task (Alvarez-Melis & Saveski, 2016; Al-Sultany & Aleqabie, 2019).…”

“…Our experiments used data from Twitter Streaming API 2 . Similar to previous works, we constructed a dataset collecting tweets containing generic terms and each tweet was labeled by the query that retrieved it (Hong & Davison, 2010;Al-Sultany & Aleqabie, 2019;Mehrotra et al, 2013;Ollagnier & Williams, 2019). We removed all tweets that were retrieved by more than one query, so as to preserve uniqueness of the tweet labels, which was important for our analysis.…”

“…Given the fact that Twitter has become a platform where a tremendous amount of content is generated, shared and consumed, this problem become of interest for the scientific community Hong and Davison presented an intuitive solution: tweet pooling (making a longer document by aggregating multiple tweets following different schemes) (Hong & Davison, 2010). Tweet-pooling has been shown to improve topic decomposition, but the performance varies depending on the pooling method (Hong & Davison, 2010;Mehrotra et al, 2013;Ma et al, 2019;Alvarez-Melis & Saveski, 2016;Ollagnier & Williams, 2019). For example, Mehrotra et al (Mehrotra et al, 2013) extended this idea by pooling all tweets that mention a given hashtag.…”