Abstract

Online advertising offers significantly finer granularity, which has been leveraged in state-of-the-art targeting methods, like Behavioral Targeting (BT). Such methods have been fur-ther complemented by recent work in Look-alike Modeling(LAM) which helps in creating models which are customized according to each advertiser’s requirements and each cam-paign’s characteristics, and which show ads to users who are most likely to convert on them, not just click them. In Look-a like Modeling given data about converters and non-converters, obtained from advertisers, we would like to train models automatically for each ad campaign. Such custom models would help target more users who are similar to the set of converters the advertiser provides. The advertisers get more freedom to define their preferred sets of users which should be used as a basis to build custom targeting models.In behavioral data, the number of conversions (positive class) per campaign is very small (conversions per impression for the advertisers in our data set are much less than 10−4),giving rise to a highly skewed training dataset, which has most records pertaining to the negative class. Campaigns with very few conversions are called as tail campaigns, and those with many conversions are called head campaigns.Creation of Look-alike Models for tail campaigns is very challenging and tricky using popular classifiers like Linear SVM and GBDT, because of the very few number of posi-tive class examples such campaigns contain. In this paper,we present an Associative Classification (AC) approach to LAM for tail campaigns. Pairs of features are used to deriverules to build a Rule-based Associative Classifier, with the rules being sorted by frequency-weighted log-likelihood ratio(F-LLR). The top krules, sorted by F-LLR, are then applied to any test record to score it. Individual features can also form rules by themselves, though the number of such rules in the top krules and the whole rule-set is very small. Our algorithm is based on Hadoop, and is thus very efficient in terms of speed.

Abstract

The concept of similarity is fundamentally important in al-most every scientific field. Clustering, distance-based outlier detection,classification, regression and search are major data mining techniques which compute the similarities between instances and hence the choice of a particular similarity measure can turn out to be a major cause of success or failure of the algorithm. The notion of similarity or distance for categorical data is not as straightforward as for continuous data and hence, is a major challenge. This is due to the fact that different values taken by a categorical attribute are not inherently ordered and hence anotion of direct comparison between two categorical values is not pos-sible. In addition, the notion of similarity can differ depending on the particular domain, dataset, or task at hand. In this paper we present anew similarity measure for categorical data DISC - Data-Intensive Simi-larity Measure for Categorical Data. DISC captures the semantics of the data without any help from domain expert for defining the similarity. In addition to these, it is generic and simple to implement. These desirable features make it a very attractive alternative to existing approaches. Our experimental study compares it with 14 other similarity measures on 24standard real datasets, out of which 12 are used for classification and 12for regression, and shows that it is more accurate than all its competitors.

Abstract

In this paper we propose an efficient method to rank the research papers from various fields of research published in various conferences over the years. This ranking method is based on citation network. The importance of a research paper is captured well by the peer vote, which in this case is the research paper being cited in other research papers. Using a modified version of the PageRank algorithm, we rank the research papers, assigning each of them an authoritative score. Using the scores of the research papers calculated by above mentioned method, we formulate scores for conferences and authors and rank them as well. We have introduced a new metric in the algorithm which takes into account the time factor in ranking the research papers to reduce the bias against the recent papers which get less time for being studied and consequently cited by the researchers as compared to the older papers. Often a researcher is more interested in finding the top conferences in a particular year rather than the overall conference ranking. Considering the year of publication of the papers, in addition to the paper scores we also calculated the year-wise score of each conference by slight improvisation of the above mentioned algorithm.

Abstract

In this paper we introduce a novel and efficient approach to detect topics in a large corpus of research papers. With rapidly growing size of academic literature, the problem of topic detection has become a very challenging task. We present a unique approach that uses closed frequent keyword-set to form topics. Our approach also provides a natural method to cluster the research papers into hierarchical, overlapping clusters using topic as similarity measure. To rank the research papers in the topic cluster, we devise a modified PageRank algorithm that assigns an authoritative score to each research paper by considering the sub-graph in which the research paper appears. We test our algorithms on the DBLP dataset and experimentally show that our algorithms are fast, effective and scalable.

Abstract

In this paper we introduce a novel and efficient approach to detect and rank topics in a large corpus of research papers. With rapidly growing size of academic literature, the problem of topic detection and topic ranking has become a challenging task. We present a unique approach that uses closed frequent keyword-set to form topics. We devise a modified time independent Page Rank algorithm that assigns an authoritative score to each topic by considering the sub-graph in which the topic appears, producing a ranked list of topics.The use of citation network and the introduction of time in variance in the topic ranking algorithm reveal very interesting results. Our approach also provides a clustering technique for the research papers using topics as similarity measure. We extend our algorithms to study various aspects of topic evolution which gives interesting insight into trends in research areas over time.Our algorithms also detect hot topics and landmark topics over the years. We test our algorithms on the DBLP dataset and show that our algorithms are fast, effective and scalable.

Abstract

Measuring inter-document similarity is one of the most essential steps in text document clustering. Traditional methods rely on representing text documents using the simple Bag-of-Words (BOW) model which assumes that terms of a text document are independent of each other. Such single term analysis of the text completely ignores the underlying (semantic) structure of a document. In the literature, sufficient efforts have been made to enrich BOW representation using phrases and n-grams like bi-grams and tri-grams. These approaches take into account dependency only between adjacent terms or a continuous sequence of terms. However, while some of the dependencies exist between adjacent words, others are more distant. In this paper, we make an effort to enrich traditional document vector by adding the notion of term-pair features. A Term-Pair feature is a pair of two terms of the same document such that they may be adjacent to each other or distant. We investigate the process of term-pair selection and propose a methodology to select potential term-pairs fromthe given document. Utilizing term proximity between distant terms also allows some flexibility for two documents to be similar if they are about similar topics but with varied writing styles. Experimental results on standard web document data set show that the clustering performance is substantially improved by adding term-pair features.

Abstract

Fuzzy association rules use fuzzy logic to convert numerical attributes to fuzzy attributes, like ldquoIncome = Highrdquo, thus maintaining the integrity of information conveyed by such numerical attributes. On the other hand, crisp association rules use sharp partitioning to transform numerical attributes to binary ones like ldquoIncome = [100 K and above]rdquo, and can potentially introduce loss of information due to these sharp ranges. Fuzzy A priori and its different variations are the only popular fuzzy association rule mining (ARM) algorithms available today. Like the crisp version of Apriori, fuzzy A priori is a very slow and inefficient algorithm for very large datasets (in the order of millions of transactions). Hence, we have come up with a new fuzzy ARM algorithm meant for fast and efficient performance on very large datasets. As compared to fuzzy A priori, our algorithm is 8-19 times faster for the very large standard real-life dataset we have used for testing with various mining workloads, both typical and extreme ones. A novel combination of features like two-phased multiple-partition tidlist-style processing, byte-vector representation of tidlists, and fast compression of tidlists contribute a lot to the efficiency in performance. In addition, unlike most two-phased ARM algorithms, the second phase is totally different from the first one in the method of processing (individual itemset processing as opposed to simultaneous itemset processing at each k-level), and is also many times faster. Our algorithm also includes an effective preprocessing technique for converting a crisp dataset to a fuzzy dataset

Abstract

Regression algorithms are used for prediction (including forecasting of time-series data), inference, hypothesis testing, and modeling of causal relationships. Statistical approaches although popular, are not generic in that they require the user to make an intelligent guess about the form of the regression equation. In this paper we present a new regression algorithm SEAR – Scalable, Efficient, Accurate kNN-based Regression. In addition to this, SEAR is simple and outlier-resilient. These desirable features make SEAR a very attractive alternative to existing approaches. Our experimental study compares SEAR with fourteen other algorithms on five standard real datasets, and shows that SEAR is more accurate than all its competitors.

Abstract

Active learning is a rapidly growing field of machine learning which aims at reducing the labeling effortof the oracle (human expert) in acquiring informa-tive training samples in domains where the costof labeling is high. Associative classification is a well established prediction method which possesses the advantages of high accuracy and faster learning rates in classification. In this paper, we propose a novel algorithm which unifies associative classification with active learning. The algorithm has two major procedures of Rule generation and rule pruning.The algorithm selects unlabeled instances from the pool of available samples and uses a unique dynamic rule sampling procedure for updating the model.The rules are dynamically sampled class association rules (CAR) which are generated using the mined Minimal infrequent itemsets. The results derived over 10 datasets from the UCI-ML repository for our approach have been compared with those from the ACTIVE-DECORATE algorithm. We also analyze our sampling method against the state of art sampling frameworks and show that our method performs better.

Abstract

In recent years, the number of electronic journal articles is growing faster than ever before; information is generated faster than people can deal with it. In order to handle this problem, many electronic periodical databases have proposed keyword search methods to decrease the effort and time spent by users in searching the journal's archives. However, the users still have to deal with a huge number of search results. In this paper, we present the problem of mining landmark papers. We treat papers that introduce important key phrases for the first time as landmark papers. Our approach combines simple ideas from text mining, information extraction and information retrieval to identify landmark papers. We show that existing related techniques such as first story detection, mining hot topics and theme mining do not effectively handle the landmark paper mining problem. Our approach is simpler and more direct for this task. We experimentally evaluate our approach on a large dataset of papers in the database or data mining areas downloaded using DBLP.