Conversation Trees: A Grammar Model for Topic Structure in Forums

Abstract

Online forum discussions proceed differently from face-to-face conversations and any single thread on an online forum contains posts on different subtopics. This work aims to characterize the content of a forum thread as a conversation tree of topics. We present models that jointly perform two tasks: segment a thread into subparts, and assign a topic to each part. Our core idea is a definition of topic structure using probabilistic grammars. By leveraging the flexibility of two grammar formalisms, Context-Free Grammars and Linear Context-Free Rewriting Systems, our models create desirable structures for forum threads: our topic segmentation is hierarchical, links non-adjacent segments on the same topic, and jointly labels the topic during segmentation. We show that our models outperform a number of tree generation baselines.

Microblog Conversation Recommendation via Joint Modeling of Topics and Discourse

“…Our work is also in line with conversation modeling for social media discussions (Ritter et al, 2010;Budak and Agrawal, 2013;Louis and Cohen, 2015;Cheng et al, 2017). Topic modeling has been employed to identify conversation content on Twitter (Ritter et al, 2010).…”

A Large-Scale Corpus for Conversation Disentanglement

“…By removing these labels and mixing conversations they create a disentanglement problem. While convenient, this risks introducing a bias, as people write differently when explicit structure is defined, and only a few papers have released data (Abbott et al, 2016;Zhang et al, 2017;Louis and Cohen, 2015). Models: Elsner and Charniak (2008) explored various message-pair feature sets and linear classifiers, combined with local and global inference methods.…”

Latent-Variable PCFGs: Background and Applications

“…As long as the backbone structure of the skeletal grammar is of the form a → α where α includes nonterminals in one form or the other, the nonterminals can be decorated with additional latent state information. Work about using other grammar formalisms with latent states includes the work of Fowler and Penn (2010) who introduced latent states into a combinatory categorial grammar (CCG) for syntactic parsing, the work of Saluja et al (2014), who generalized L-PCFGs to synchronous LPCFGs and proposed to estimate them using both a spectral algorithm and EM for machine translation and the work of Louis and Cohen (2015) who modeled online forum topic structure by using LCFRS with latent states (the latent states corresponded to topics that need to be inferred from data). Models similar to L-PCFGs have been used for parsing with discontinuous elements (Nederhof and Yli-Jyrä, 2017).…”