Mukkai S. Krishnamoorthy

Syntactic segmentation and labeling of digitized pages from technical journals

A method for extracting alternating horizontal and vertical projection profiles are from nested sub-blocks of scanned page images of technical documents is discussed. The thresholded profile strings are parsed using the compiler utilities Lex and Yacc. The significant document components are demarcated and identified by the recursive application of block grammars. Backtracking for error recovery and branch and bound for maximum-area labeling are implemented with Unix Shell programs. Results of the segmentation and labeling process are stored in a labeled x-y tree. It is shown that families of technical documents that share the same layout conventions can be readily analyzed. Results from experiments in which more than 20 types of document entities were identified in sample pages from two journals are presented. >

A random walk method for alleviating the sparsity problem in collaborative filtering

Collaborative Filtering is one of the most widely used approaches in recommendation systems which predicts user preferences by learning past user-item relationships. In recent years, item-oriented collaborative filtering methods came into prominence as they are more scalable compared to user-oriented methods. Item-oriented methods discover item-item relationships from the training data and use these relations to compute predictions. In this paper, we propose a novel item-oriented algorithm, Random Walk Recommender, that first infers transition probabilities between items based on their similarities and models finite length random walks on the item space to compute predictions. This method is especially useful when training data is less than plentiful, namely when typical similarity measures fail to capture actual relationships between items. Aside from the proposed prediction algorithm, the final transition probability matrix computed in one of the intermediate steps can be used as an item similarity matrix in typical item-oriented approaches. Thus, this paper suggests a method to enhance similarity matrices under sparse data as well. Experiments on MovieLens data show that Random Walk Recommender algorithm outperforms two other item-oriented methods in different sparsity levels while having the best performance difference in sparse datasets.

A Mechanizable Induction Principle for Equational Specifications

Automating proofs of properties of functions defined on inductively constructed data structures is important in many computer science and artificial intelligence applications, in particular in program verification and specification systems. A new induction principle based on a constructor model of a data structure is developed. This principle along with a given function definition as a set of equations is used to construct automatically an induction scheme suitable for proving inductive properties of the function. The proposed induction principle thus gives different induction schema for different function definitions, just as Boyer and Moores prover does. A novel feature of this approach is that it can also be used for proving properties by induction for data structures such as integers, finite sets, whose values cannot be freely constructed, i.e., constructors for such data structures are related to each other. This method has been implemented in RRL, a rewrite-rule based theorem prover. More than a hundred theorems in number theory including the unique prime factorization theorem, have been proved using the method.

Graph Theoretic and Spectral Analysis of Enron Email Data

Analysis of social networks to identify communities and model their evolution has been an active area of recent research. This paper analyzes the Enron email data set to discover structures within the organization. The analysis is based on constructing an email graph and studying its properties with both graph theoretical and spectral analysis techniques. The graph theoretical analysis includes the computation of several graph metrics such as degree distribution, average distance ratio, clustering coefficient and compactness over the email graph. The spectral analysis shows that the email adjacency matrix has a rank-2 approximation. It is shown that preprocessing of data has significant impact on the results, thus a standard form is needed for establishing a benchmark data.

Exact and Approximate Solutions for the Gate Matrix Layout Problem

We consider the gate matrix layout problem for VLSI circuits, which is known to be NP-complete. We present an efficient algorithm for determining whether two tracks suffice. For the general problem of minimizing the number of tracks (and, hence, the area) needed, we design an attractive dynamic programming formulation to guarantee optimality. We also investigate the performance of fast heuristic algorithms published in the literature and demonstrate that there exist families of problem instances for which the ratio of the number of tracks required by these heuristics to the optimal value is unbounded. Moreover, we show that this result holds for any on-line layout algorithm. We additionally prove that, unless P = NP, no polynomial-time layout algorithm can ensure that the number of tracks it requires never exceeds k plus the optimum, for any constant k.

Fast parallel computation of hermite and smith forms of polynomial matrices

Boolean circuits of polynomial size and polylogarithmic depth are given for computing the Hermite and Smith normal forms of polynomial matrices over finite fields and the field of rational numbers. The circuits for the Smith normal form computation are probabilistic ones and also determine very efficient sequential algorithms. Furthermore, we give a polynomial-time deterministic sequential algorithm for the Smith normal form over the rationals. The Smith normal form algorithms are applied to the rational canonical form of matrices over finite fields and the field of rational numbers.

Node-Deletion NP-Complete Problems

The entire class of node-deletion problems can be stated as follows: Given a graph G, find the minimum number of nodes to be deleted so that the remaining subgraph g satisfies a specified property

Two complementary techniques for digitized document analysis

\pi

Web Usage Mining — Languages and Algorithms

. For each

LOGML: Log Markup Language for Web Usage Mining

\pi