Bring you to the past: Automatic Generation of Topically Relevant Event Chronicles

Abstract

An event chronicle provides people with an easy and fast access to learn the past. In this paper, we propose the first novel approach to automatically generate a topically relevant event chronicle during a certain period given a reference chronicle during another period. Our approach consists of two core components-a timeaware hierarchical Bayesian model for event detection, and a learning-to-rank model to select the salient events to construct the final chronicle. Experimental results demonstrate our approach is promising to tackle this new problem.

News Stream Summarization using Burst Information Networks

“…In this paper, we adopt the same evaluation setting and use the same reference summaries and the annotations with our previous work (Ge et al, 2015b) to evaluate our summaries' quality. For the event entries that are not in Ge et al (2015b)'s annotations, we have 3 human judges annotate them according to the previous annotation guideline and consider an entry correct if it is annotated as correct by at least 2 judges.…”

“…• B-HAC: similar to NB except that BurstVSM representation (Zhao et al, 2012) is used for event detection using Hierarchical Agglomerative Clustering algorithm. • TAHBM: similar to NB except that the stateof-the-art event detection model (TaHBM) proposed by Ge et al (2015b) is used for event detection. • Ge et al (2015b): the state-of-the-art stream summarization approach which used TaHBM to detect events and L2R model to rank events.…”

“…• TAHBM: similar to NB except that the stateof-the-art event detection model (TaHBM) proposed by Ge et al (2015b) is used for event detection. • Ge et al (2015b): the state-of-the-art stream summarization approach which used TaHBM to detect events and L2R model to rank events. Note that we did not compare with previous multidocument summarization models because the goal and setting of stream summarization are different from multi-document summarization, as Section 1 https://en.wikipedia.org/wiki/2009 sports politics disaster military comprehensive P@50 P@100 P@50 P@100 P@50 P@100 P@50 P@100 P@50 P@100 discussed.…”

“…Run time Can be run in parallel Table 6: Run time of BINet-based approaches and Ge et al (2015b)'s approach trivial) event entries that are not important enough to be included in the stream summary account for the majority of errors for both models. This is because it is difficult to distinguish these trivial events since the corpus we used as a text stream is not as ideal as the assumption that the more important events, the more times they are reported.…”

Generating sets of related sentences from input seed features

“…NLG has been widely applied to several fields, for instance to the generation of recommendations (Lim-Cheng et al, 2014). However, generation systems are currently designed for very specific domains (Ramos-Soto et al, 2015) and predefined purposes (Ge et al, 2015). The use of SW's technologies can facilitate the development of more flexible and domain independent systems, that could be adapted to the target audience or purposes, which would considerably advance the state of the art in NLG.…”