A Consolidated Open Knowledge Representation for Multiple Texts

Abstract

We propose to move from Open Information Extraction (OIE) ahead to Open Knowledge Representation (OKR), aiming to represent information conveyed jointly in a set of texts in an open textbased manner. We do so by consolidating OIE extractions using entity and predicate coreference, while modeling information containment between coreferring elements via lexical entailment. We suggest that generating OKR structures can be a useful step in the NLP pipeline, to give semantic applications an easy handle on consolidated information across multiple texts.

Interactive Abstractive Summarization for Event News Tweets

“…None of these works considers interactive summaries, and in particular none incorporates sufficient data for our modes of user interaction. We next briefly review the Open Knowledge Representation recently introduced by Wities et al (2017), which is used by our system.…”

“…We illustrate the components of the OKR formalism that are central to our summarization method via the example OKR structure in Figure 1 (see Wities et al (2017) for full details). On the top, there are four original tweets.…”

“…To assess and improve the value of our system, we have conducted two usability studies employing standard usability tests. The tests were performed on a dataset of human annotated OKR structures (of the form of Figure 1) released by Wities et al (2017). We took their 6 largest clusters of event tweets, of about 100 tweets each.…”

“…• Static: the full ordered list of sentences generated by our system (Section 3), with no interactive features nor metadata (such as concept expansion, knowledge scores, etc.). As mentioned earlier, we have used the goldstandard OKR structures for 6 of the events released by Wities et al (2017). Six users were each presented with two events in each interface (IAS, Tweet, Static), where the assignment of event to interface and order of interfaces were different for each user.…”

“…To facilitate the modular construction of interactive summaries, we utilize as input a consolidated representation of texts, in particular the recent Open Knowledge Representation (OKR) of Wities et al (2017). Briefly, this representation captures the propositions of the texts, where corefferring concepts or propositions are collapsed together while keeping links to the original mentions (see Section 2).…”