Research

Hunting Consequences of Activities from Texts

Motivation

One way to predict future events is to build predictive models that generalize from specific sets of sequences of events to provide likelihoods of future outcomes [2]. Specifically in predicting human activities, one can rely on narratives in movie script, e.g., after hike a hill or climb a mountain, one usually drink water [3].

Given the possible next activities of an activity, the challenge is to identify which of these are the consequences, giving higher probabilities for them to happen. Sometimes consequences of an activity are stated explicitly with causal markers, e.g. “He embraces her because he loves her.” However, more often the resulting activities happen because there exists states in between, e.g., “He climb a mountain. (It makes him tired/thirsty/hungry.) He takes a rest/drinks water/eats lunch.”

Model

We introduce the notion of states, which we can simply assume to be adjectives. An activity, assumed to be an in-/transitive verb, can lead to states (turn off light → dark), and vice versa, a state may result in activities (dark → turn on light).

Use cases

Several ways that states may be useful:

  1. We can better predict (quantitatively and maybe, qualitatively) possible following activities based on the states in between, e.g., climb a mountain may lead to take a rest, drink water or eat lunch.
  2. Sentiment analysis is easier when using adjectives, e.g. SentiWordNet [1]. Hence, the states can help more in determining the emotions evoked by an activity, at least in identifying the emotion’s polarity.

References

[1] Stefano Baccianella, Andrea Esuli, and Fabrizio Sebastiani. Sentiwordnet 3.0: An enhanced lexical resource for sentiment analysis and opinion mining. In Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC’10), Valletta, Malta, 2010.

[2] Kira Radinsky and Eric Horvitz. Mining the web to predict future events. In Proceedings of the Sixth ACM International Conference on Web Search and Data Mining, WSDM ’13, pages 255–264, New York, NY, USA, 2013.

[3] Niket Tandon, Gerard de Melo, Abir De, and Gerhard Weikum. Knowlywood: Mining activity knowledge from hollywood narratives. In Proceedings of the 24th ACM International on Conference on Information and Knowledge Management, CIKM ’15, pages 223–232, New York, NY, USA, 2015.