Showing papers presented at "Knowledge Discovery and Data Mining in 2008"

Factorization meets the neighborhood: a multifaceted collaborative filtering model

Abstract: Recommender systems provide users with personalized suggestions for products or services. These systems often rely on Collaborating Filtering (CF), where past transactions are analyzed in order to establish connections between users and products. The two more successful approaches to CF are latent factor models, which directly profile both users and products, and neighborhood models, which analyze similarities between products or users. In this work we introduce some innovations to both approaches. The factor and neighborhood models can now be smoothly merged, thereby building a more accurate combined model. Further accuracy improvements are achieved by extending the models to exploit both explicit and implicit feedback by the users. The methods are tested on the Netflix data. Results are better than those previously published on that dataset. In addition, we suggest a new evaluation metric, which highlights the differences among methods, based on their performance at a top-K recommendation task.

ArnetMiner: extraction and mining of academic social networks

Abstract: This paper addresses several key issues in the ArnetMiner system, which aims at extracting and mining academic social networks. Specifically, the system focuses on: 1) Extracting researcher profiles automatically from the Web; 2) Integrating the publication data into the network from existing digital libraries; 3) Modeling the entire academic network; and 4) Providing search services for the academic network. So far, 448,470 researcher profiles have been extracted using a unified tagging approach. We integrate publications from online Web databases and propose a probabilistic framework to deal with the name ambiguity problem. Furthermore, we propose a unified modeling approach to simultaneously model topical aspects of papers, authors, and publication venues. Search services such as expertise search and people association search have been provided based on the modeling results. In this paper, we describe the architecture and main features of the system. We also present the empirical evaluation of the proposed methods.

Get another label? improving data quality and data mining using multiple, noisy labelers

Abstract: This paper addresses the repeated acquisition of labels for data items when the labeling is imperfect. We examine the improvement (or lack thereof) in data quality via repeated labeling, and focus especially on the improvement of training labels for supervised induction. With the outsourcing of small tasks becoming easier, for example via Rent-A-Coder or Amazon's Mechanical Turk, it often is possible to obtain less-than-expert labeling at low cost. With low-cost labeling, preparing the unlabeled part of the data can become considerably more expensive than labeling. We present repeated-labeling strategies of increasing complexity, and show several main results. (i) Repeated-labeling can improve label quality and model quality, but not always. (ii) When labels are noisy, repeated labeling can be preferable to single labeling even in the traditional setting where labels are not particularly cheap. (iii) As soon as the cost of processing the unlabeled data is not free, even the simple strategy of labeling everything multiple times can give considerable advantage. (iv) Repeatedly labeling a carefully chosen set of points is generally preferable, and we present a robust technique that combines different notions of uncertainty to select data points for which quality should be improved. The bottom line: the results show clearly that when labeling is not perfect, selective acquisition of multiple labels is a strategy that data miners should have in their repertoire; for certain label-quality/cost regimes, the benefit is substantial.

Relational learning via collective matrix factorization

Abstract: Relational learning is concerned with predicting unknown values of a relation, given a database of entities and observed relations among entities. An example of relational learning is movie rating prediction, where entities could include users, movies, genres, and actors. Relations encode users' ratings of movies, movies' genres, and actors' roles in movies. A common prediction technique given one pairwise relation, for example a #users x #movies ratings matrix, is low-rank matrix factorization. In domains with multiple relations, represented as multiple matrices, we may improve predictive accuracy by exploiting information from one relation while predicting another. To this end, we propose a collective matrix factorization model: we simultaneously factor several matrices, sharing parameters among factors when an entity participates in multiple relations. Each relation can have a different value type and error distribution; so, we allow nonlinear relationships between the parameters and outputs, using Bregman divergences to measure error. We extend standard alternating projection algorithms to our model, and derive an efficient Newton update for the projection. Furthermore, we propose stochastic optimization methods to deal with large, sparse matrices. Our model generalizes several existing matrix factorization methods, and therefore yields new large-scale optimization algorithms for these problems. Our model can handle any pairwise relational schema and a wide variety of error models. We demonstrate its efficiency, as well as the benefit of sharing parameters among relations.

Learning classifiers from only positive and unlabeled data

Abstract: The input to an algorithm that learns a binary classifier normally consists of two sets of examples, where one set consists of positive examples of the concept to be learned, and the other set consists of negative examples. However, it is often the case that the available training data are an incomplete set of positive examples, and a set of unlabeled examples, some of which are positive and some of which are negative. The problem solved in this paper is how to learn a standard binary classifier given a nontraditional training set of this nature.Under the assumption that the labeled examples are selected randomly from the positive examples, we show that a classifier trained on positive and unlabeled examples predicts probabilities that differ by only a constant factor from the true conditional probabilities of being positive. We show how to use this result in two different ways to learn a classifier from a nontraditional training set. We then apply these two new methods to solve a real-world problem: identifying protein records that should be included in an incomplete specialized molecular biology database. Our experiments in this domain show that models trained using the new methods perform better than the current state-of-the-art biased SVM method for learning from positive and unlabeled examples.

Microscopic evolution of social networks

Abstract: We present a detailed study of network evolution by analyzing four large online social networks with full temporal information about node and edge arrivals. For the first time at such a large scale, we study individual node arrival and edge creation processes that collectively lead to macroscopic properties of networks. Using a methodology based on the maximum-likelihood principle, we investigate a wide variety of network formation strategies, and show that edge locality plays a critical role in evolution of networks. Our findings supplement earlier network models based on the inherently non-local preferential attachment.Based on our observations, we develop a complete model of network evolution, where nodes arrive at a prespecified rate and select their lifetimes. Each node then independently initiates edges according to a "gap" process, selecting a destination for each edge according to a simple triangle-closing model free of any parameters. We show analytically that the combination of the gap distribution with the node lifetime leads to a power law out-degree distribution that accurately reflects the true network in all four cases. Finally, we give model parameter settings that allow automatic evolution and generation of realistic synthetic networks of arbitrary scale.

Influence and correlation in social networks

Abstract: In many online social systems, social ties between users play an important role in dictating their behavior. One of the ways this can happen is through social influence, the phenomenon that the actions of a user can induce his/her friends to behave in a similar way. In systems where social influence exists, ideas, modes of behavior, or new technologies can diffuse through the network like an epidemic. Therefore, identifying and understanding social influence is of tremendous interest from both analysis and design points of view.This is a difficult task in general, since there are factors such as homophily or unobserved confounding variables that can induce statistical correlation between the actions of friends in a social network. Distinguishing influence from these is essentially the problem of distinguishing correlation from causality, a notoriously hard statistical problem.In this paper we study this problem systematically. We define fairly general models that replicate the aforementioned sources of social correlation. We then propose two simple tests that can identify influence as a source of social correlation when the time series of user actions is available.We give a theoretical justification of one of the tests by proving that with high probability it succeeds in ruling out influence in a rather general model of social correlation. We also simulate our tests on a number of examples designed by randomly generating actions of nodes on a real social network (from Flickr) according to one of several models. Simulation results confirm that our test performs well on these data. Finally, we apply them to real tagging data on Flickr, exhibiting that while there is significant social correlation in tagging behavior on this system, this correlation cannot be attributed to social influence.

Angle-based outlier detection in high-dimensional data

Abstract: Detecting outliers in a large set of data objects is a major data mining task aiming at finding different mechanisms responsible for different groups of objects in a data set All existing approaches, however, are based on an assessment of distances (sometimes indirectly by assuming certain distributions) in the full-dimensional Euclidean data space In high-dimensional data, these approaches are bound to deteriorate due to the notorious "curse of dimensionality" In this paper, we propose a novel approach named ABOD (Angle-Based Outlier Detection) and some variants assessing the variance in the angles between the difference vectors of a point to the other points This way, the effects of the "curse of dimensionality" are alleviated compared to purely distance-based approaches A main advantage of our new approach is that our method does not rely on any parameter selection influencing the quality of the achieved ranking In a thorough experimental evaluation, we compare ABOD to the well-established distance-based method LOF for various artificial and a real world data set and show ABOD to perform especially well on high-dimensional data

Feedback effects between similarity and social influence in online communities

Abstract: A fundamental open question in the analysis of social networks is to understand the interplay between similarity and social ties. People are similar to their neighbors in a social network for two distinct reasons: first, they grow to resemble their current friends due to social influence; and second, they tend to form new links to others who are already like them, a process often termed selection by sociologists. While both factors are present in everyday social processes, they are in tension: social influence can push systems toward uniformity of behavior, while selection can lead to fragmentation. As such, it is important to understand the relative effects of these forces, and this has been a challenge due to the difficulty of isolating and quantifying them in real settings.We develop techniques for identifying and modeling the interactions between social influence and selection, using data from online communities where both social interaction and changes in behavior over time can be measured. We find clear feedback effects between the two factors, with rising similarity between two individuals serving, in aggregate, as an indicator of future interaction -- but with similarity then continuing to increase steadily, although at a slower rate, for long periods after initial interactions. We also consider the relative value of similarity and social influence in modeling future behavior. For instance, to predict the activities that an individual is likely to do next, is it more useful to know the current activities of their friends, or of the people most similar to them?

Discrimination-aware data mining

Abstract: In the context of civil rights law, discrimination refers to unfair or unequal treatment of people based on membership to a category or a minority, without regard to individual merit. Rules extracted from databases by data mining techniques, such as classification or association rules, when used for decision tasks such as benefit or credit approval, can be discriminatory in the above sense. In this paper, the notion of discriminatory classification rules is introduced and studied. Providing a guarantee of non-discrimination is shown to be a non trivial task. A naive approach, like taking away all discriminatory attributes, is shown to be not enough when other background knowledge is available. Our approach leads to a precise formulation of the redlining problem along with a formal result relating discriminatory rules with apparently safe ones by means of background knowledge. An empirical assessment of the results on the German credit dataset is also provided.

Fast collapsed gibbs sampling for latent dirichlet allocation

Abstract: In this paper we introduce a novel collapsed Gibbs sampling method for the widely used latent Dirichlet allocation (LDA) model. Our new method results in significant speedups on real world text corpora. Conventional Gibbs sampling schemes for LDA require O(K) operations per sample where K is the number of topics in the model. Our proposed method draws equivalent samples but requires on average significantly less then K operations per sample. On real-word corpora FastLDA can be as much as 8 times faster than the standard collapsed Gibbs sampler for LDA. No approximations are necessary, and we show that our fast sampling scheme produces exactly the same results as the standard (but slower) sampling scheme. Experiments on four real world data sets demonstrate speedups for a wide range of collection sizes. For the PubMed collection of over 8 million documents with a required computation time of 6 CPU months for LDA, our speedup of 5.7 can save 5 CPU months of computation.

Context-aware query suggestion by mining click-through and session data

Abstract: Query suggestion plays an important role in improving the usability of search engines. Although some recently proposed methods can make meaningful query suggestions by mining query patterns from search logs, none of them are context-aware - they do not take into account the immediately preceding queries as context in query suggestion. In this paper, we propose a novel context-aware query suggestion approach which is in two steps. In the offine model-learning step, to address data sparseness, queries are summarized into concepts by clustering a click-through bipartite. Then, from session data a concept sequence suffix tree is constructed as the query suggestion model. In the online query suggestion step, a user's search context is captured by mapping the query sequence submitted by the user to a sequence of concepts. By looking up the context in the concept sequence sufix tree, our approach suggests queries to the user in a context-aware manner. We test our approach on a large-scale search log of a commercial search engine containing 1:8 billion search queries, 2:6 billion clicks, and 840 million query sessions. The experimental results clearly show that our approach outperforms two baseline methods in both coverage and quality of suggestions.

Joint latent topic models for text and citations

Abstract: In this work, we address the problem of joint modeling of text and citations in the topic modeling framework. We present two different models called the Pairwise-Link-LDA and the Link-PLSA-LDA models.The Pairwise-Link-LDA model combines the ideas of LDA [4] and Mixed Membership Block Stochastic Models [1] and allows modeling arbitrary link structure. However, the model is computationally expensive, since it involves modeling the presence or absence of a citation (link) between every pair of documents. The second model solves this problem by assuming that the link structure is a bipartite graph. As the name indicates, Link-PLSA-LDA model combines the LDA and PLSA models into a single graphical model.Our experiments on a subset of Citeseer data show that both these models are able to predict unseen data better than the baseline model of Erosheva and Lafferty [8], by capturing the notion of topical similarity between the contents of the cited and citing documents. Our experiments on two different data sets on the link prediction task show that the Link-PLSA-LDA model performs the best on the citation prediction task, while also remaining highly scalable. In addition, we also present some interesting visualizations generated by each of the models.

The cost of privacy: destruction of data-mining utility in anonymized data publishing

Abstract: Re-identification is a major privacy threat to public datasets containing individual records. Many privacy protection algorithms rely on generalization and suppression of "quasi-identifier" attributes such as ZIP code and birthdate. Their objective is usually syntactic sanitization: for example, k-anonymity requires that each "quasi-identifier" tuple appear in at least k records, while l-diversity requires that the distribution of sensitive attributes for each quasi-identifier have high entropy. The utility of sanitized data is also measured syntactically, by the number of generalization steps applied or the number of records with the same quasi-identifier. In this paper, we ask whether generalization and suppression of quasi-identifiers offer any benefits over trivial sanitization which simply separates quasi-identifiers from sensitive attributes. Previous work showed that k-anonymous databases can be useful for data mining, but k-anonymization does not guarantee any privacy. By contrast, we measure the tradeoff between privacy (how much can the adversary learn from the sanitized records?) and utility, measured as accuracy of data-mining algorithms executed on the same sanitized records.For our experimental evaluation, we use the same datasets from the UCI machine learning repository as were used in previous research on generalization and suppression. Our results demonstrate that even modest privacy gains require almost complete destruction of the data-mining utility. In most cases, trivial sanitization provides equivalent utility and better privacy than k-anonymity, l-diversity, and similar methods based on generalization and suppression.

Composition attacks and auxiliary information in data privacy

Abstract: Privacy is an increasingly important aspect of data publishing. Reasoning about privacy, however, is fraught with pitfalls. One of the most significant is the auxiliary information (also called external knowledge, background knowledge, or side information) that an adversary gleans from other channels such as the web, public records, or domain knowledge. This paper explores how one can reason about privacy in the face of rich, realistic sources of auxiliary information. Specifically, we investigate the effectiveness of current anonymization schemes in preserving privacy when multiple organizations independently release anonymized data about overlapping populations.1. We investigate composition attacks, in which an adversary uses independent anonymized releases to breach privacy. We explain why recently proposed models of limited auxiliary information fail to capture composition attacks. Our experiments demonstrate that even a simple instance of a composition attack can breach privacy in practice, for a large class of currently proposed techniques. The class includes k-anonymity and several recent variants.2. On a more positive note, certain randomization-based notions of privacy (such as differential privacy) provably resist composition attacks and, in fact, the use of arbitrary side information.This resistance enables "stand-alone" design of anonymization schemes, without the need for explicitly keeping track of other releases.We provide a precise formulation of this property, and prove that an important class of relaxations of differential privacy also satisfy the property. This significantly enlarges the class of protocols known to enable modular design.

iSAX: indexing and mining terabyte sized time series

Abstract: Current research in indexing and mining time series data has produced many interesting algorithms and representations. However, the algorithms and the size of data considered have generally not been representative of the increasingly massive datasets encountered in science, engineering, and business domains. In this work, we show how a novel multi-resolution symbolic representation can be used to index datasets which are several orders of magnitude larger than anything else considered in the literature. Our approach allows both fast exact search and ultra fast approximate search. We show how to exploit the combination of both types of search as sub-routines in data mining algorithms, allowing for the exact mining of truly massive real world datasets, containing millions of time series.

Knowledge transfer via multiple model local structure mapping

Abstract: The effectiveness of knowledge transfer using classification algorithms depends on the difference between the distribution that generates the training examples and the one from which test examples are to be drawn. The task can be especially difficult when the training examples are from one or several domains different from the test domain. In this paper, we propose a locally weighted ensemble framework to combine multiple models for transfer learning, where the weights are dynamically assigned according to a model's predictive power on each test example. It can integrate the advantages of various learning algorithms and the labeled information from multiple training domains into one unified classification model, which can then be applied on a different domain. Importantly, different from many previously proposed methods, none of the base learning method is required to be specifically designed for transfer learning. We show the optimality of a locally weighted ensemble framework as a general approach to combine multiple models for domain transfer. We then propose an implementation of the local weight assignments by mapping the structures of a model onto the structures of the test domain, and then weighting each model locally according to its consistency with the neighborhood structure around the test example. Experimental results on text classification, spam filtering and intrusion detection data sets demonstrate significant improvements in classification accuracy gained by the framework. On a transfer learning task of newsgroup message categorization, the proposed locally weighted ensemble framework achieves 97% accuracy when the best single model predicts correctly only on 73% of the test examples. In summary, the improvement in accuracy is over 10% and up to 30% across different problems.

Efficient semi-streaming algorithms for local triangle counting in massive graphs

Abstract: In this paper we study the problem of local triangle counting in large graphs. Namely, given a large graph G = (V;E) we want to estimate as accurately as possible the number of triangles incident to every node υ ∈ V in the graph. The problem of computing the global number of triangles in a graph has been considered before, but to our knowledge this is the first paper that addresses the problem of local triangle counting with a focus on the efficiency issues arising in massive graphs. The distribution of the local number of triangles and the related local clustering coefficient can be used in many interesting applications. For example, we show that the measures we compute can help to detect the presence of spamming activity in large-scale Web graphs, as well as to provide useful features to assess content quality in social networks.For computing the local number of triangles we propose two approximation algorithms, which are based on the idea of min-wise independent permutations (Broder et al. 1998). Our algorithms operate in a semi-streaming fashion, using O(jV j) space in main memory and performing O(log jV j) sequential scans over the edges of the graph. The first algorithm we describe in this paper also uses O(jEj) space in external memory during computation, while the second algorithm uses only main memory. We present the theoretical analysis as well as experimental results in massive graphs demonstrating the practical efficiency of our approach.

The structure of information pathways in a social communication network

Abstract: Social networks are of interest to researchers in part because they are thought to mediate the flow of information in communities and organizations. Here we study the temporal dynamics of communication using on-line data, including e-mail communication among the faculty and staff of a large university over a two-year period. We formulate a temporal notion of "distance" in the underlying social network by measuring the minimum time required for information to spread from one node to another - a concept that draws on the notion of vector-clocks from the study of distributed computing systems. We find that such temporal measures provide structural insights that are not apparent from analyses of the pure social network topology. In particular, we define the network backbone to be the subgraph consisting of edges on which information has the potential to flow the quickest. We find that the backbone is a sparse graph with a concentration of both highly embedded edges and long-range bridges - a finding that sheds new light on the relationship between tie strength and connectivity in social networks.

Community evolution in dynamic multi-mode networks

Abstract: A multi-mode network typically consists of multiple heterogeneous social actors among which various types of interactions could occur. Identifying communities in a multi-mode network can help understand the structural properties of the network, address the data shortage and unbalanced problems, and assist tasks like targeted marketing and finding influential actors within or between groups. In general, a network and the membership of groups often evolve gradually. In a dynamic multi-mode network, both actor membership and interactions can evolve, which poses a challenging problem of identifying community evolution. In this work, we try to address this issue by employing the temporal information to analyze a multi-mode network. A spectral framework and its scalability issue are carefully studied. Experiments on both synthetic data and real-world large scale networks demonstrate the efficacy of our algorithm and suggest its generality in solving problems with complex relationships.

Identifying authoritative actors in question-answering forums: the case of Yahoo! answers

Abstract: We consider the problem of identifying authoritative users in Yahoo! Answers. A common approach is to use link analysis techniques in order to provide a ranked list of users based on their degree of authority. A major problem for such an approach is determining how many users should be chosen as authoritative from a ranked list. To address this problem, we propose a method for automatic identification of authoritative actors. In our approach, we propose to model the authority scores of users as a mixture of gamma distributions. The number of components in the mixture is estimated by the Bayesian Information Criterion (BIC) while the parameters of each component are estimated using the Expectation-Maximization (EM) algorithm. This method allows us to automatically discriminate between authoritative and non-authoritative users. The suitability of our proposal is demonstrated in an empirical study using datasets from Yahoo! Answers.

Hypergraph spectral learning for multi-label classification

Abstract: A hypergraph is a generalization of the traditional graph in which the edges are arbitrary non-empty subsets of the vertex set. It has been applied successfully to capture high-order relations in various domains. In this paper, we propose a hypergraph spectral learning formulation for multi-label classification, where a hypergraph is constructed to exploit the correlation information among different labels. We show that the proposed formulation leads to an eigenvalue problem, which may be computationally expensive especially for large-scale problems. To reduce the computational cost, we propose an approximate formulation, which is shown to be equivalent to a least squares problem under a mild condition. Based on the approximate formulation, efficient algorithms for solving least squares problems can be applied to scale the formulation to very large data sets. In addition, existing regularization techniques for least squares can be incorporated into the model for improved generalization performance. We have conducted experiments using large-scale benchmark data sets, and experimental results show that the proposed hypergraph spectral learning formulation is effective in capturing the high-order relations in multi-label problems. Results also indicate that the approximate formulation is much more efficient than the original one, while keeping competitive classification performance.

A tree-based approach for frequent pattern mining from uncertain data

Abstract: Many frequent pattern mining algorithms find patterns from traditional transaction databases, in which the content of each transaction--namely, items--is definitely known and precise. However, there are many real-life situations in which the content of transactions is uncertain. To deal with these situations, we propose a tree-based mining algorithm to efficiently find frequent patterns from uncertain data, where each item in the transactions is associated with an existential probability. Experimental results show the efficiency of our proposed algorithm.

Building semantic kernels for text classification using wikipedia

Abstract: Document classification presents difficult challenges due to the sparsity and the high dimensionality of text data, and to the complex semantics of the natural language. The traditional document representation is a word-based vector (Bag of Words, or BOW), where each dimension is associated with a term of the dictionary containing all the words that appear in the corpus. Although simple and commonly used, this representation has several limitations. It is essential to embed semantic information and conceptual patterns in order to enhance the prediction capabilities of classification algorithms. In this paper, we overcome the shortages of the BOW approach by embedding background knowledge derived from Wikipedia into a semantic kernel, which is then used to enrich the representation of documents. Our empirical evaluation with real data sets demonstrates that our approach successfully achieves improved classification accuracy with respect to the BOW technique, and to other recently developed methods.

What is frequent in a single graph

Abstract: The standard, transactional setting of pattern mining assumes that data is subdivided in transactions; the aim is to find patterns that can be mapped onto at least a minimum number of transactions However, this setting can be hard to apply when the aim is to find graph patterns in databases consisting of large graphs For instance, the web, or any social network, is a single large graph that one may not wish to split into small parts The focus in network analysis is on finding structural regularities or anomalies in one network, rather than finding structural regularities common to a set of them This requires us to revise the definition of key concepts in pattern mining, such as support, in the single-graph setting Our contribution is a support measure that we prove to be computationally less expensive and often closer to intuition than other measures proposed Further we prove several properties between these measures and experimentally validate the efficiency of our measure

Mobile call graphs: beyond power-law and lognormal distributions

Abstract: We analyze a massive social network, gathered from the records of a large mobile phone operator, with more than a million users and tens of millions of calls. We examine the distributions of the number of phone calls per customer; the total talk minutes per customer; and the distinct number of calling partners per customer. We find that these distributions are skewed, and that they significantly deviate from what would be expected by power-law and lognormal distributions.To analyze our observed distributions (of number of calls, distinct call partners, and total talk time), we propose PowerTrack , a method which fits a lesser known but more suitable distribution, namely the Double Pareto LogNormal (DPLN) distribution, to our data and track its parameters over time. Using PowerTrack , we find that our graph changes over time in a way consistent with a generative process that naturally results in the DPLN distributions we observe. Furthermore, we show that this generative process lends itself to a natural and appealing social wealth interpretation in the context of social networks such as ours. We discuss the application of those results to our model and to forecasting.

FAST: a roc-based feature selection metric for small samples and imbalanced data classification problems

Abstract: The class imbalance problem is encountered in a large number of practical applications of machine learning and data mining, for example, information retrieval and filtering, and the detection of credit card fraud. It has been widely realized that this imbalance raises issues that are either nonexistent or less severe compared to balanced class cases and often results in a classifier's suboptimal performance. This is even more true when the imbalanced data are also high dimensional. In such cases, feature selection methods are critical to achieve optimal performance. In this paper, we propose a new feature selection method, Feature Assessment by Sliding Thresholds (FAST), which is based on the area under a ROC curve generated by moving the decision boundary of a single feature classifier with thresholds placed using an even-bin distribution. FAST is compared to two commonly-used feature selection methods, correlation coefficient and RELevance In Estimating Features (RELIEF), for imbalanced data classification. The experimental results obtained on text mining, mass spectrometry, and microarray data sets showed that the proposed method outperformed both RELIEF and correlation methods on skewed data sets and was comparable on balanced data sets; when small number of features is preferred, the classification performance of the proposed method was significantly improved compared to correlation and RELIEF-based methods.

Extracting shared subspace for multi-label classification

Abstract: Multi-label problems arise in various domains such as multi-topic document categorization and protein function prediction. One natural way to deal with such problems is to construct a binary classifier for each label, resulting in a set of independent binary classification problems. Since the multiple labels share the same input space, and the semantics conveyed by different labels are usually correlated, it is essential to exploit the correlation information contained in different labels. In this paper, we consider a general framework for extracting shared structures in multi-label classification. In this framework, a common subspace is assumed to be shared among multiple labels. We show that the optimal solution to the proposed formulation can be obtained by solving a generalized eigenvalue problem, though the problem is non-convex. For high-dimensional problems, direct computation of the solution is expensive, and we develop an efficient algorithm for this case. One appealing feature of the proposed framework is that it includes several well-known algorithms as special cases, thus elucidating their intrinsic relationships. We have conducted extensive experiments on eleven multi-topic web page categorization tasks, and results demonstrate the effectiveness of the proposed formulation in comparison with several representative algorithms.

Anonymizing transaction databases for publication

Abstract: This paper considers the problem of publishing "transaction data" for research purposes. Each transaction is an arbitrary set of items chosen from a large universe. Detailed transaction data provides an electronic image of one's life. This has two implications. One, transaction data are excellent candidates for data mining research. Two, use of transaction data would raise serious concerns over individual privacy. Therefore, before transaction data is released for data mining, it must be made anonymous so that data subjects cannot be re-identified. The challenge is that transaction data has no structure and can be extremely high dimensional. Traditional anonymization methods lose too much information on such data. To date, there has been no satisfactory privacy notion and solution proposed for anonymizing transaction data. This paper proposes one way to address this issue.

Classification with partial labels

Abstract: In this paper, we address the problem of learning when some cases are fully labeled while other cases are only partially labeled, in the form of partial labels. Partial labels are represented as a set of possible labels for each training example, one of which is the correct label. We introduce a discriminative learning approach that incorporates partial label information into the conventional margin-based learning framework. The partial label learning problem is formulated as a convex quadratic optimization minimizing the L2-norm regularized empirical risk using hinge loss. We also present an efficient algorithm for classification in the presence of partial labels. Experiments with different data sets show that partial label information improves the performance of classification when there is traditional fully-labeled data, and also yields reasonable performance in the absence of any fully labeled data.