Semantic Annotation for Microblog Topics Using Wikipedia Temporal Information

Abstract

Trending topics in microblogs such as Twitter are valuable resources to understand social aspects of real-world events. To enable deep analyses of such trends, semantic annotation is an effective approach; yet the problem of annotating microblog trending topics is largely unexplored by the research community. In this work, we tackle the problem of mapping trending Twitter topics to entities from Wikipedia. We propose a novel model that complements traditional text-based approaches by rewarding entities that exhibit a high temporal correlation with topics during their burst time period. By exploiting temporal information from the Wikipedia edit history and page view logs, we have improved the annotation performance by 17-28%, as compared to the competitive baselines.

Semantic analysis of Twitter content

“…The task of linking textual mentions to their proper entities from structured knowledge bases has attracted a lot of attention over the past several years [51,68,109,115,130,139]. This task is primarily composed of two major steps: i) The first step is concerned with the identification of the terms or phrases that have the potential to be linked to some entity in the knowledge base.…”

“…There are typically two types of features that have been used in the literature, namely local and global features. Local features include things such as the distance obtained from a cosine similarity measure [68], edit distance similarity [68], the probability of the mention serving as the anchor text for the candidate entity [68] and the temporal relevance of a candidate entity for the given mention [115]. The detail related work will be presented in Section 2.2.4.…”

“…Therefore, these models will not be able to use dynamic information about the temporal co-evolution of mentions and entities. Tran et al [115] is one of the only few that considers the notion of temporality. The authors incorporate temporal information from the Wikipedia edit history and page view logs to link hashtags to entities.…”

“…In this dataset, the average number of annotations for each tweet is 2.17 and there are 127 tweets that do not have any annotations. There are two other datasets released by Tran et al [115] and Shen et al [109]. However, we were not able to use them because they are limited only to annotations for trending hashtags and annotations customized to specific users and are not publicly available.…”

Tweeki: Linking Named Entities on Twitter to a Knowledge Graph

“…For the latter step, roughly two types of features are used. Local features identify one mention at the time and disambiguate it separately such as using prior probability in Liu et al (2013) or temporal relevance mention in Tran et al (2015). Global features take a more comprehensive view and consider the relations between the entity candidates for the different mentions of the tweet (Huang et al, 2014;Feng et al, 2018).…”