International School of Information Management

University of Mysore


ISiM Research Colloquium (1)


Date: April 3, 2009


Time: 3:00 p.m. onwards


Presentation (1)


Title of the talk: On the Nature of Co-citations: Interpretation and Analysis


Speaker: Mandar R. Mutalikdesai, Research Scholar, IIITB



Co-citation has long been used as a measure of topical similarity between objects in various data sets. However, there are some characteristic differences in the way co-citations are formed in different data sets. We summarize these differences with respect to scientific literature, WWW pages and Wikipedia pages. Based on this, we propose three interpretations of co-citations: (1) Endorsements of citations, (2) Knowledge aggregation activities, and (3) Measures of conditional topical relevance. We shall briefly discuss all three interpretations. However, this talk will mainly elaborate on the first interpretation: Co-citations as endorsements of citations. Such an interpretation gives a new measure that can be used by focused or topical crawlers/surfers to surf the Web and digital libraries such as CiteSeer in a fashion that is most pertinent to the topic of interest. We also find some interesting patterns of "citation endorsements" on the Web and on CiteSeer, which we will discuss in this talk. Finally, we propose a topical ranking scheme based on citation endorsements, which can be used as a fine tuning measure to existing techniques for relevance ranking in citation graphs.


Presentation (2)


Title: Mining Semantics from Lexical Cooccurrence Data

Speaker: Aditya Ramana Rachakonda, Research Scholar, IIITB



Lexical cooccurrence in documents is not uniformly random. The statistics inferred from the term cooccurrence data enable us to probabilistically model the dependencies between terms. It has been argued in cognitive sciences that these dependencies closely resemble the way semantic memory is organised in human beings. This work attempts to model some commonly used semantic associations as probabilistic measures to enable mining of semantics automatically from text corpora. The semantic associations being modelled are synonyms like car and automobile, semantic siblings like Sachin Tendulkar and Rahul Dravid and topical anchors like Disney for Mickey Mouse, Donald Duck and Pluto. Of these, some results on topical anchors will be presented in more detail.





ISiM Lecture Hall 1

ISiM, University of Mysore, Manasagangotri

Mysore 570 006