Semantic Abstraction for generalization of tweet classification: An evaluation of incident-related tweets

Abstract

Social media is a rich source of up-to-date information about events such as incidents. The sheer amount of available information makes machine learning approaches a necessity to process this information further. This learning problem is often concerned with regionally restricted datasets such as data from only one city. Because social media data such as tweets varies considerably across different cities, the training of efficient models requires labeling data from each city of interest, which is costly and time consuming. To avoid such an expensive labeling procedure, a generalizable model can be trained on data from one city and then applied to data from different cities. In this paper, we present Semantic Abstraction to improve the generalization of tweet classification. In particular, we derive features from Linked Open Data and include location and temporal mentions. A comprehensive evaluation on twenty datasets from ten different cities shows that Semantic Abstraction is indeed a valuable means for improving generalization. We show that this not only holds for a two-class problem where incident-related tweets are separated from non-related ones but also for a four-class problem where three different incident types and a neutral class are distinguished. To get a thorough understanding of the generalization problem itself, we closely examined rule-based models from our evaluation. We conclude that on the one hand, the quality of the model strongly depends on the class distribution. On the other hand, the rules learned on cities with an equal class distribution are in most cases much more intuitive than those induced from skewed distributions. We also found that most of the learned rules rely on the novel semantically abstracted features.

Keyphrase Extraction from Disaster-related Tweets

“…For the embedding-based metric, we used both GloVe embeddings pre-trained on general Twitter (General Embd), and GloVe embeddings that we trained on a crisis-related Twitter corpus (Crisis Embd). The crisis-related embeddings were trained on three previously published crisis datasets, specifically, CrisisLexT6 [39], CrisisLexT26 [40] and 2CTweets [50], together with tweets collected using the Twitter Streaming API during Hurricanes Harvey, Irma, and Maria, and Mexico Earthquake. The total corpus contains approximately 5.8 million tweets.…”

A framework for event classification in tweets based on hybrid semantic enrichment

“…As a means to deal with the poor textual content of tweets, related work has suggested the use of external information to add context to tweet contents, in applications such as Event Classification (ABEL et al, 2012a;PACKER et al, 2012;PAULHEIM, 2013;GUCKELSBERGER;JANSSEN, 2015) and Sentiment Analysis (SAIF; HE; ALANI, 2012). As a general approach, the textual features to be enriched are selected according to some criterion, and mapped into the resources available in external knowledge source (e.g.…”

Mining the Web of Linked Data with RapidMiner

“…Schulz et al use the extension for developing an approach for semantic abstraction for generalization of tweets classification [32,33]. Shidik et al [34] make use of the extension for developing a machine learning approach for predicting forest fires using LOD as background knowledge.…”

“…All factors not influenced by factors originating in online data access, such as local performance of data analysis, can be addressed by design in RapidMiner, since there, e.g., are cloud computing services 31 as well as the Radoop extension 32 for data analysis with Hadoop, which allow for scaling the analytics operations.…”