Current algorithms for Maximal Frequent Sub-graph (MFS) mining do not scale to large databases with more than 300k graphs. Frequency computation is commonly done using numerous sub graph isomorphism operations, which are computationally expensive. This paper explores a partition-based secondary memory algorithm, PMCS, that can make MFS mining for large databases viable. PMCS intelligently avoids redundant computations to perform frequency computation operations optimally. Its intermediate results are also small enough to fit in the main memory. Our results demonstrate that PM CS scales to databases of up to 1000k graphs with an average of 25–30 edges per graph.

The paper aims to identify cited text spans in the reference paper related to the given citance in the citing paper. We refer to it as cited text span retrieval (CTSR). Most current methods attempt this task by relying on pre-trained off-the-shelf deep learning models like SciBERT. Though these models are pre-trained on large datasets, they under-perform in out-of-domain settings. We introduce CitRet, a novel hybrid model for CTSR that leverages unique semantic and syntactic structural characteristics of scientific documents. This enables us to use significantly less data for finetuning. We use only 1040 documents for finetuning. Our model augments mildly-trained SBERT-based contextual embeddings with pre-trained non-contextual Word2Vec embeddings to calculate semantic textual similarity. We demonstrate the performance of our model on the CLSciSumm shared tasks. It improves the state-of-the-art results by over 15% on the F1 score evaluation.

Existing Math Word Problem (MWP) solvers have achieved high accuracy on benchmark datasets.However, prior works have shown that such solvers do not generalize well and rely on superficial cues to achieve high performance.In this paper, we first conduct experiments to showcase that this behaviour is mainly associated with the limited size and diversity present in existing MWP datasets.Next, we propose several data augmentation techniques broadly categorized into Substitution and Paraphrasing based methods.By deploying these methods we increase the size of existing datasets by five folds.Extensive experiments on two benchmark datasets across three state-of-the-art MWP solvers shows that proposed methods increase the generalization and robustness of existing solvers.On average, proposed methods significantly increase the state-of-the-art results by over five percentage points on benchmark datasets.Further, the solvers trained on the augmented dataset performs comparatively better on the challenge test set.We also show the effectiveness of proposed techniques through ablation studies and verify the quality of augmented samples through human evaluation. Original ProblemProblem: Nancy grew 8 potatoes.Sandy grew 5 potatoes.How many potatoes did they grow in total ?True Equation: X = 8+5Paraphrasing Method Problem: How many potatoes did they grow in all given that nancy grew 8 potatoes and sandy grew 5 potatoes.Equation Label: X = 8+5 Substitution Method Problem: Dwight grew 8 potatoes.Juliette grew 5 potatoes.How many potatoes did they grow together ?Equation Label: X = 8+5

We study an important and challenging task of attacking natural language processing models in a hard label black box setting. We propose a decision-based attack strategy that crafts high quality adversarial examples on text classification and entailment tasks. Our proposed attack strategy leverages population-based optimization algorithm to craft plausible and semantically similar adversarial examples by observing only the top label predicted by the target model. At each iteration, the optimization procedure allow word replacements that maximizes the overall semantic similarity between the original and the adversarial text. Further, our approach does not rely on using substitute models or any kind of training data. We demonstrate the efficacy of our proposed approach through extensive experimentation and ablation studies on five state-of-the-art target models across seven benchmark datasets. In comparison to attacks proposed in prior literature, we are able to achieve a higher success rate with lower word perturbation percentage that too in a highly restricted setting.

We study an important task of attacking natural language processing models in a black box setting. We propose an attack strategy that crafts semantically similar adversarial examples on text classification and entailment tasks. Our proposed attack finds candidate words by considering the information of both the original word and its surrounding context. It jointly leverages masked language modelling and next sentence prediction for context understanding. In comparison to attacks proposed in prior literature, we are able to generate high quality adversarial examples that do significantly better both in terms of success rate and word perturbation percentage.

Existing black box search methods have achieved high success rate in generating adversarial attacks against NLP models.However, such search methods are inefficient as they do not consider the amount of queries required to generate adversarial attacks.Also, prior attacks do not maintain a consistent search space while comparing different search methods.In this paper, we propose a query efficient attack strategy to generate plausible adversarial examples on text classification and entailment tasks.Our attack jointly leverages attention mechanism and locality sensitive hashing (LSH) to reduce the query count.We demonstrate the efficacy of our approach by comparing our attack with four baselines across three different search spaces.Further, we benchmark our results across the same search space used in prior attacks.In comparison to attacks proposed, on an average, we are able to reduce the query count by 75% across all datasets and target models.We also demonstrate that our attack achieves a higher success rate when compared to prior attacks in a limited query setting.

Variational autoencoders were proven successful in domains such as computer vision and speech processing. Their adoption for modeling user preferences is still unexplored, although recently it is starting to gain attention in the current literature. In this work, we propose a model which extends variational autoencoders by exploiting the rich information present in the past preference history. We introduce a recurrent version of the VAE, where instead of passing a subset of the whole history regardless of temporal dependencies, we rather pass the consumption sequence subset through a recurrent neural network. At each time-step of the RNN, the sequence is fed through a series of fully-connected layers, the output of which models the probability distribution of the most likely future preferences. We show that handling temporal information is crucial for improving the accuracy of the VAE: In fact, our model beats the current state-of-the-art by valuable margins because of its ability to capture temporal dependencies among the user-consumption sequence using the recurrent encoder still keeping the fundamentals of variational autoencoders intact.

Recommender Systems are an integral part of music sharing platforms. Often the aim of these systems is to increase the time, the user spends on the platform and hence having a high commercial value. The systems which aim at increasing the average time a user spends on the platform often need to recommend songs which the user might want to listen to next at each point in time. This is different from recommendation systems which try to predict the item which might be of interest to the user at some point in the user lifetime but not necessarily in the very near future. Prediction of next song the user might like requires some kind of modeling of the user interests at the given point of time. Attentive neural networks have been exploiting the sequence in which the items were selected by the user to model the implicit short-term interests of the user for the task of next item prediction, however we feel that features of the songs occurring in the sequence could also convey some important information about the short-term user interest which only the items cannot. In this direction we propose a novel attentive neural architecture which in addition to the sequence of items selected by the user, uses the features of these items to better learn the user short-term preferences and recommend next song to the user.

In recent years, the importance of feature engineering has been confirmed by the exceptional performance of deep learning techniques, that automate this task for some applications. For others, feature engineering requires substantial manual effort in designing and selecting features and is often tedious and non-scalable. We present AutoLearn, a regression-based feature learning algorithm. Being data-driven, it requires no domain knowledge and is hence generic. Such a representation is learnt by mining pairwise feature associations, identifying the linear or non-linear relationship between each pair, applying regression and selecting those relationships that are stable and improve the prediction performance. Our experimental evaluation on 18 UC Irvine and 7 Gene expression datasets, across different domains, provides evidence that the features learnt through our model can improve the overall prediction accuracy by 13.28%, compared to original feature space and 5.87% over other top performing models, across 8 different classifiers without using any domain knowledge.

Extracting structured information from unstructured text is a critical problem. Over the past few years, various clustering algorithms have been proposed to solve this problem. In addition, various algorithms based on probabilistic topic models have been developed to find the hidden thematic structure from various corpora (i.e publications, blogs etc). Both types of algorithms have been transferred to the domain of scientific literature to extract structured information to solve problems like data exploration, expert detection etc. In order to remain domain-agnostic, these algorithms do not exploit the structure present in a scientific publication. Majority of researchers interpret a scientific publication as research conducted to report progress in solving some research problems. Following this interpretation, in this paper we present a different outlook to the same problem by modelling scientific publications around research problems. By associating a scientific publication with a research problem, exploring the scientific literature becomes more intuitive. In this paper, we propose an unsupervised framework to mine research problems from titles and abstracts of scientific literature. Our framework uses weighted frequent phrase mining to generate phrases and filters them to obtain high-quality phrases. These high-quality phrases are then used to segment the scientific publication into meaningful semantic units. After segmenting publications, we apply a number of heuristics to score the phrases and sentences to identify the research problems. In a postprocessing step we use a neighborhood based algorithm to merge different representations of the same problems. Experiments conducted on parts of DBLP dataset show promising results.

We present here a data mining approach for part-of-speech (POS) tagging, an important Natural language processing (NLP) classification task. We propose a semi-supervised associative classification method for POS tagging. Existing methods for building POS taggers require extensive domain and linguistic knowledge and resources. Our method uses a combination of a small POS tagged corpus and untagged text data as training data to build the classifier model using association rules. Our tagger works well with very little training data also. The use of semi-supervised learning provides the advantage of not requiring a large high quality tagged corpus. These properties make it especially suitable for resource poor languages. Our experiments on various resource-rich, resource-moderate and resource-poor languages show good performance without using any language specific linguistic information. We note that inclusion of such features in our method may further improve the performance. Results also show that for smaller training data sizes our tagger performs better than state-of-the-art CRF tagger using same features as our tagger.

In this paper we present a novel associative classification algorithm for visual concept detection using interest points (color descriptors). Our algorithm relies on fuzzy (soft) partitions to generate a per-descriptor record model which is used for training associative classifiers. In this paper, we use the first 10000 images of the MIR Flickr dataset (with 38 visual concepts — 14 regular annotation topics and 24 pre-annotations) for evaluating our algorithm. Because of its fuzzy nature, our approach also handles polysemy and synonymy (common problems in most crisp (non-fuzzy) image classifiers) very well. As associative classification leverages frequent patterns mined from a given dataset, its performance as adjudged from average area under curve (AUC) values over three-fold cross validation is very good, especially for regular annotation topics and for “scenic” concepts. Our approach has the added advantage that the rules used for classification have clear semantics, and can be comprehended easily, unlike other classifiers, such as SVM, which act as black-boxes. From an empirical perspective (on the MIR Flickr dataset), the performance of our algorithm is much better, than that of either bag-of-words (Linear SVM and SVM RBF Kernel) or Linear SVM based on fuzzy (soft) partitions—both benchmark approaches are based on interest points (color descriptors).

Fuzzy Association Rule Mining (ARM) has been extensively used in relational or transactional datasets having less-to-medium number of attributes/dimensions. The mined fuzzy association rules (patterns) are not only used for manual analysis by domain experts, but are also leveraged to drive further mining tasks like classification and clustering which automate decision-making. Such fuzzy association rules can also be derived from high-dimensional numerical datasets, like image datasets, in order to train fuzzy associative classifiers or clustering algorithms. Traditional Fuzzy ARM algorithms are not able to mine rules from them efficiently, since such algorithms are meant to deal with datasets with relatively much less number of attributes/dimensions. Hence, in this paper we propose FAR-HD which is a Fuzzy ARM algorithm designed specifically for large high-dimensional datasets. FAR-HD processes fuzzy frequent itemsets in a DFS manner using a two-phased multiple-partition tidlist-based strategy. It also uses a byte-vector representation of tidlists, with the tidlists stored in the main memory in a compressed form (using a fast generic compression method). Additionally, FAR-HD uses Fuzzy Clustering to convert each numerical vector of the original input dataset to a fuzzy-cluster-based representation, which is ultimately used for the actual Fuzzy ARM process. FAR-HD has been compared experimentally with Fuzzy Apriori (7-15 times faster), which is the most popular Fuzzy ARM algorithm, and a Fuzzy ARM algorithm (1.1-4 times faster) which we proposed earlier and which is designed to work with very large but traditional (with fewer attributes) datasets.

This paper first presents an efficient quantum-dot cellular automata (QCA) design for the Ladner-Fischer prefix adder. We then present an efficient QCA design of a hybrid adder that combines the Ladner-Fischer adder with a ripple carry adder. We show that the hybrid adder has better performance (in terms of latency) in QCA than a Ladner-Fischer or a ripple carry adder. We also show that the hybrid adder has a smaller area-delay product than existing adder designs in QCA. Results of simulation in QCADesigner are also given and they support the theory presented.

All associative classifiers developed till now are crisp in nature, and thus use sharp partitioning to transform numerical attributes to binary ones like “Income = [100K and above]”. On the other hand, the novel fuzzy associative classification algorithm called FACISME, which we propose in this paper, uses fuzzy logic to convert numerical attributes to fuzzy attributes, like “Income = High”, thus maintaining the integrity of information conveyed by such numerical attributes. Moreover, FACISME is based on maximum entropy, and uses iterative scaling, both of which lend a very strong theoretical foundation to the algorithm. Entropy is one of the best measures of information, and maximum-entropy-based algorithms do not assume independence of parameters in the classification process. Thus, FACISME provides very good accuracy, and can work with all types of datasets (irrespective of size and type of attributes - numerical or binary) and domains.

Conventional Association Rule Mining (ARM) algorithms usually deal with datasets with binary values, and expect any numerical values to be converted to binary ones using sharp partitions, like Age = 25 to 60. In order to mitigate this constraint, Fuzzy logic is used to convert quantitative values of attributes to binary ones, so as to eliminate any loss of information arising due to sharp partitioning, especially at partition boundaries, and then generate fuzzy association rules. But, before any fuzzy ARM algorithm can be used, the original dataset (with crisp attributes) needs to be transformed into a form with fuzzy attributes. This paper describes a methodology, called FPrep, to do this pre-processing, which first involves using fuzzy clustering to generate fuzzy partitions, and then uses these partitions to get a fuzzy version (with fuzzy records) of the original dataset. Ultimately, the fuzzy data (fuzzy records) are represented in a standard manner such that they can be used as input to any kind of fuzzy ARM algorithm, irrespective of how it works and processes fuzzy data. We also show that FPrep is much faster than other such comparable transformation techniques, which in turn depend on non-fuzzy techniques, like hard clustering (CLARANS and CURE). Moreover, we illustrate the quality of the fuzzy partitions generated using FPrep, and the number of frequent itemsets generated by a fuzzy ARM algorithm when preceded by FPrep.

This paper presents Perturbed Frequent Itemset based Classification Technique (PERFICT), a novel associative classification approach based on perturbed frequent itemsets. Most of the existing associative classifiers work well on transactional data where each record contains a set of boolean items. They are not very effective in general for relational data that typically contains real valued attributes. In PERFICT, we handle real attributes by treating items as (attribute, value) pairs, where the value is not the original one, but is perturbed by a small amount and is a range based value. We also propose our own similarity measure which captures the nature of real valued attributes and provide effective weights for the itemsets. The probabilistic contributions of different itemsets is taken into considerations during classification. Some of the applications where such a technique is useful are in signal classification, medical diagnosis and handwriting recognition. Experiments conducted on the UCI Repository datasets show that PERFICT is highly competitive in terms of accuracy in comparison with popular associative classification methods.

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 Apriori and its different variations are the only popular fuzzy association rule mining (ARM) algorithms available today. Like the crisp version of Apriori, fuzzy Apriori 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 Apriori, 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.

In this paper, we present RBNBC, a repeat based Naive Bayes classifier of bio-sequences that uses maximal frequent subsequences as features. RBNBC's design is based on generic ideas that can apply to other domains where the data is organized as collections of sequences. Specifically, RBNBC uses a novel formulation of Naive Bayes that incorporates repeated occurrences of subsequences within each sequence. Our extensive experiments on two collections of protein families show that it performs as well as existing state-of-the-art probabilistic classifiers for bio-sequences. This is surprising as it is a pure data mining based generic classifier that does not require domain-specific background knowledge. We note that domain-specific ideas could further increase its performance.

The output of Boolean association rule mining algorithms is often too large for manual examination. For dense datasets, it is often impractical to even generate all frequent itemsets. The closed itemset approach handles this information overload by pruning "uninteresting" rules following the observation that most rules can be derived from other rules. We propose a new framework, namely, the generalized closed (or g-closed) itemset framework. By allowing for a small tolerance in the accuracy of itemset supports, we show that the number of such redundant rules is far more than what was previously estimated. Our scheme can be integrated into both levelwise algorithms (Apriori) and two-pass algorithms (ARMOR). We evaluate its performance by measuring the reduction in output size as well as in response time. Our experiments show that incorporating g-closed itemsets provides significant performance improvements on a variety of databases.

Addresses the question of how much space remains for performance improvement over current association rule mining algorithms. Our approach is to compare their performance against an "Oracle algorithm" that knows in advance the identities of all frequent item sets in the database and only needs to gather the actual supports of these item sets, in one scan over the database, to complete the mining process. Clearly, any practical algorithm has to do at least this much work in order to generate mining rules. While the notion of the Oracle is conceptually simple, its construction is not equally straightforward. In particular, it is critically dependent on the choice of data structures and database organizations used during the counting process. We present a carefully engineered implementation of Oracle that makes the best choices for these design parameters at each stage of the counting process. We also present anew mining algorithm, called ARMOR (Association Rule Mining based on ORacle), whose structure is derived by making minimal changes to Oracle, and is guaranteed to complete in two passes over the database. This is in marked contrast to the earlier approaches which designed new algorithms by trying to address the limitations of previous online algorithms. Although ARMOR is derived from Oracle, it shares the positive features of a variety of previous algorithms such as PARTITION, CARMA, AS-CPA, VIPER and DELTA. Our empirical study shows that ARMOR consistently performs within a factor of two of Oracle, over both real and synthetic databases.