Mark James Carman

SIR-Hawkes: Linking Epidemic Models and Hawkes Processes to Model Diffusions in Finite Populations

Among the statistical tools for online information diffusion modeling, both epidemic models and Hawkes point processes are popular choices. The former originate from epidemiology, and consider information as a viral contagion which spreads into a population of online users. The latter have roots in geophysics and finance, view individual actions as discrete events in continuous time, and modulate the rate of events according to the self-exciting nature of event sequences. Here, we establish a novel connection between these two frameworks. Namely, the rate of events in an extended Hawkes model is identical to the rate of new infections in the Susceptible-Infected-Recovered (SIR) model after marginalizing out recovery events -- which are unobserved in a Hawkes process. This result paves the way to apply tools developed for SIR to Hawkes, and vice versa. It also leads to HawkesN, a generalization of the Hawkes model which accounts for a finite population size. Finally, we derive the distribution of cascade sizes for HawkesN, inspired by methods in stochastic SIR. Such distributions provide nuanced explanations to the general unpredictability of popularity: the distribution for diffusion cascade sizes tends to have two modes, one corresponding to large cascade sizes and another one around zero.Two of the main frameworks used for modeling information diffusions in the online are epidemic models and Hawkes point processes. The former consider information as a viral contagion which spreads into a population of online users, and employ tools initially developed in the field of epidemiology. The latter view individual broadcasts of information as events in a point process and they modulate the event rate according to observed (or assumed) social principles; they have been broadly used in fields such as finance and geophysics. Here, we study for the first time the connection between these two mature frameworks, and we find them to be equivalent. More precisely, the rate of events in the Hawkes model is identical to the rate of new infections in the Susceptible-InfectedRecovered (SIR) model when taking the expectation over recovery events – which are unobserved in a Hawkes process. This paves the way to apply tools developed for one framework across the gap, to the other framework. We make three further contributions in this work. First, we propose HawkesN, an extension of the basic Hawkes model, in which we introduce the notion of finite maximum number of events that can occur. Second, we show HawkesN to explain real retweet cascades better than the current state-of-the-art Hawkes modeling. The size of the population can be learned while observing the cascade, at the expense of requiring larger amounts of training data. Third, we employ an SIR method based on Markov chains for computing the final size distribution for a partially observed cascade fitted with HawkesN. We propose an explanation to the generally perceived randomness of online popularity: the final size distribution for real diffusion cascades tends to have two maxima, one corresponding to large cascade sizes and another one around zero. ACM Reference format: Marian-Andrei Rizoiu, Swapnil Mishra, Quyu Kong, Mark Carman, and Lexing Xie. 1997. SIR-Hawkes: on the Relationship Between Epidemic Models and Hawkes Point Processes. In Proceedings of arXiv submission, Canberra, Australia, 2017, Nov. 2017 (arXiv’18), 16 pages. https://doi.org/10.475/123_4

Leveraging Label Category Relationships in Multi-class Crowdsourcing

Current quality control methods for crowdsourcing largely account for variations in worker responses to items by interactions between item difficulty and worker expertise. Few have taken into account the semantic relationships that can exist between the response label categories. When the number of the label categories is large, these relationships are naturally indicative of how crowd-workers respond to items, with expert workers tending to respond with more semantically related categories to the categories of true labels, and with difficult items tending to see greater spread in the responded labels. Based on these observations, we propose a new statistical model which contains a latent real-valued matrix for capturing the relatedness of response categories alongside variables for worker expertise, item difficulty and item true labels. The model can be easily extended to incorporate prior knowledge about the semantic relationships between response labels in the form of a hierarchy over them. Experiments show that compared with numerous state-of-the-art baselines, our model (both with and without the prior knowledge) yields superior true label prediction performance on four new crowdsourcing datasets featuring large sets of label categories.

Sarcasm Detection Using Incongruity Within Target Text

Prior work in sarcasm detection uses indicators such as (a) unigrams and pragmatic features (such as emoticons, etc.) by Gonzalez-Ibanez et al. (Proceedings of the 49th annual meeting of the association for computational linguistics: human language technologies: short papers-volume 2, pp 581–586, 2011), Carvalho et al. (Proceedings of the 1st international CIKM workshop on topic-sentiment analysis for mass opinion. ACM, pp 53–56, 2009), Barbieri et al. (Modelling sarcasm in twitter: a novel approach, ACL 2014, p 50, 2014b), or (b) patterns extracted from techniques such as hashtag-based sentiment by Maynard and Greenwood (Proceedings of LREC, 2014), Liebrecht et al. (The perfect solution for detecting sarcasm in tweets# not, 2013), a positive verb being followed by a negative situation by Riloff et al. (Proceedings of the conference on empirical methods in natural language processing 2013, pp 704–714, 2013), or discriminative n-grams by Tsur et al. (ICWSM, 2010), Davidov et al. (Proceedings of the fourteenth conference on computational natural language learning. Association for Computational Linguistics, pp 107–116, 2010b).

Sarcasm Detection Using Contextual Incongruity

In the previous chapter, we presented approaches that capture incongruity within target text. However, as observed in errors reported by these approaches, some sarcastic text may require additional contextual information so that the sarcasm to be understood. This is true in case of sentences like ‘Nicki Minaj, don’t I hate her!’ or ‘Your parents must be really proud of you!’ These forms of sarcasm can be detected using contextual incongruity. Here, ‘contextual’ refers to information beyond the target text. In this chapter, we present approaches that capture contextual incongruity in order to detect sarcasm. We consider two settings. The first setting is a monologue (in Sect. 4.1) where a single author is being analyzed. In this case, we consider the historical context of the author, i.e., the text created by the author of the target text and create a sentiment map of entities. The second setting is a dialogue (in Sect. 4.2) where multiple participants take part in a conversation. In this case, we use sequence labeling as a novel formulation of sarcasm detection to capture contextual incongruity in the dialogue.

Understanding the Phenomenon of Sarcasm

In the monograph so far, we introduced computational sarcasm and presented past work related to sarcasm in linguistics and computational linguistics. In this chapter, we aim to understand the phenomenon of sarcasm through three studies. Before we take on the problems of sarcasm detection and generation, these studies help us understand the challenges of computational sarcasm. Each of these studies could also lead to detailed areas of research themselves.