Artificial Intelligence (AI) Based Sentiment Analysis for News Items

Sheafer Tamir , HUJI, Faculty of Social Sciences, Political Science
Shenhav Shaul, HUJI, Faculty of Social Sciences, Political Science
Fogel - Dror Yair, HUJI


  • Advanced tools for analyzing, monitoring, and understanding public and political discourse around the world
  • The tools bring together theoretical understanding of political and public discourse with state-of- the-art computational text analysis methods
  • Unique textual features of public discourse, such as multiple players, various perspectives, conflict-orientation, and dynamics, are addressed
  • Sophisticated algorithms that mimic human reading of texts, taking into account elements such as context, dynamic terminology, and framing, have been developed


Our Innovation

Both inductive and deductive learning methods are used to differentiate between different topics, events and actions in the public and political domains. The technology differentiates between perspectives by which topics, events and actions are described, using contextual, sequential and spatial information

The technology enables the fast decomposition of a new domain at high resolution with minimal human intervention. It uses a proprietary variation of topic modeling for inductive learning of discourse, nlp parsing tools, and deep neural networks to produce an accurate analysis of the public domain that closely resembles the way human beings perceive it


Key Features

  • Rapid decomposition of new domains through high-resolution analysis, using inductive and deductive learning technologies
  • Combines theoretical understanding of political and public discourse with state-of- the-art computational text analysis methods
  • High-resolution analysis of topics, themes, events, and perspectives
  • Expert human supervision and continuous training of the algorithm
  • Able to operate in different languages

Topics described by a specific media outlet and the relations between them. This network can reveal the way an outlet frames different topics using the context in which they are presented. In the example, the difference between the way FOX News and the BBC frames the BDS is seen. While both discuss the BDS in the context of the Israeli-Palestinian conflict, FOX also creates ties between BDS and anti-Semitism (in accordance with the Israeli government's strategy), while the BBC does not.

Development Milestones

  • Research and develop additional methods:
  • Framing and story analysis
  • Speeding up the process of learning a new domain in order to fully automate the process
  • Predicting the behavior of "waves" or bursts of media coverage of different topic domains
  • Develop end-to-end prototypes of media monitoring systems focusing on hate crimes and hate discourse and international interventions in conflicts

The Opportunity

The technology can provide strategic solutions for key questions such as how to associate or disassociate actors with certain topics or values; how messages are accepted in different outlets; and what are the typical dynamic of topics to gain or lose media and public attention. The technology has relevance for anyone needing a continuously-updated sophisticated picture of political and public domains, with an ability to forecast public discourse, such as: 

  • International corporations
  • International investors and the stock market
  • Government and state agencies
  • Media organizations
  • International and inter-governmental organizations and bodies, such as EU, UN, G8, NATO
  • Political leaders, parties and candidates in dense media markets where tracking media coverage is a challenge
  • NGOs such as civil rights movements and immigrant rights organizations

Patent Status

Granted US 9,772,996

Contact for more information:

Anna Pellivert