Department of Computer Science
The Intelligent Systems Research Group aims to concentrate on the mathematical and computational tools necessary to develop systems capable of adaptive behaviour in complex real-world environments. The group is especially interested in applications of these tools to explore problems in areas like Biology, Economics, Language, Sociology and Text Analysis.
Inductive Logic Programming:
Research in Inductive Logic Programming (ILP) is at the intersection of machine learning, logic programming, statistics, and optimization. The existence and rapid growth of scientific and industrial databases have brought into focus the need for automated methods that assist the discovery of trends and predictive patterns in data, and communicating them in a manner designed to provoke insight. This has turned attention to machine learning techniques capable of representing complex real-world concepts, and at the same time, constructing hypotheses in some human-comprehensible form. ILP is at the cutting-edge of this. Given a set of observations and background knowledge encoded in the form of statements in first-order logic, an ILP system attempts to construct models for the observations. The models are in the same language as the observations and background knowledge, namely first-order logic. Read more...
Social Computing and IR
Research in Social Computing and Information Retrieval is an interdisciplinary attempt to use computational formulations to solve problems in social sciences, text processing and retrieving relevant information from the world wide web (including social media). Social computing broadly has two varied perspectives. One is to devise computer simulations for constructing generative models to analyze complex social phenomena. The other perspective focuses on mining relevant inferences from the social media.
Constructing Generative Models
The focus in this perspective is to design agent based models and simulations for the social process or structure under study. This kind of generative modeling is different from the traditional approaches of equation based models and macro & micro-simulations. It is often visualized as a transition in social modeling from factors to actors. These models use a bottom-up approach and take the actors (role players in the concerned social setting) as the basic point of model. The actors are modeled as agents along with their attributes and their typical behaviours. Any rules governing agent interactions are also coded into the model and the model is then let to operate on its own. The system-level behaviour (and properties) are then observed and the outcome analyzed. What is interesting to see is how different system-level behaviours emerge out of local interactions between agents. This kind of modeling often does not require any explicit theory about the social phenomenon under study and is very useful for modeling phenomenon having emergent properties. This modeling approach has been used in the recent past to model variety of complex social processes and structures ranging from settlement segregation patterns to organizational behaviours.
Social Media Analytics
Social media analytics uses computational ideas from Information Retrieval (such as crawling, clustering, classification, topic discovery and sentiment analysis) to discover useful patterns in the posts on social web and also to mine relevant inferences out of it. With the transformation of the World Wide Web into a more participative form, often referred to as Web 2.0, we now have have billions of users who are not only consuming the data on the Web but are co-creators. We now have huge amount of information contained in social media in various forms ranging from blogposts to youtube videos. Social media analytics focuses on the data of textual kind and aims to process the underlying texts to extract and identify various useful inferences. This inference mining may be aimed at extracting inferences for commercial exploitation (such as finding the sentiment of bloggers on a product launch) or for a cross-cultural sociological analysis. It is the second aim which is more challenging. We can use IR based computational formulations for identifying the variation in opinions of people from different demographic (and cultural) regions on an issue of relevance to them.Both these perspectives sometime compliment each other in the sense that one helps in theorizing and the other verifies the theory. Computation of Social Network Analysis measures sometimes also produce valuable indicators for structural properties of the underlying social groups.
