Social media has become an important place for online marketing like never before. Businesses use various techniques to identify and reach potential customers across multiple platforms and deliver a message to grab their attention. A notable post could attract potential customers to the product landing page. However, the acquisition is only the beginning. The landing page should respond to the visitor's need for persuasion to increase conversion rates. Showing every visitor the same page is far from that goal. Even if the product meets everyone's needs, their priorities may differ. In this study, we propose a pipeline that includes gathering and identifying potential customers from Twitter, determining their priorities by understanding the context of their message, and creating a coherent paragraph that addresses the issue to display on the landing page.

In this paper, we propose and test three content-based hypotheses that significantly increase message virality. We measure virality as the retweet counts of messages in a pair of real-world Twitter datasets A large dataset - UK Brexit with 51 million tweets from 2.8 million users between June 1, 2015 and May 12, 2019 and a smaller dataset - Nord Stream 2 with 516,000 tweets from 250,000 users between October 1, 2019 and October 15, 2019. We hypothesize, test and conclude that messages incorporating “negativity bias”, “causal arguments” and “threats to personal or societal core values of target audiences” singularly and jointly increase message virality on social media.

Modeling and exploring high-order connectivity patterns, also called network motifs, are essential for understanding the fundamental structures that control and mediate the behavior of many complex systems. For example, in social networks, triangles have been proven to play the fundamental role in understanding social network communities; in online transaction networks, detecting directed looped transactions helps identify money laundering activities; in personally identifiable information networks, the star-shaped structures may correspond to a set of synthetic identities. Despite the ubiquity of such high-order structures, many existing graph clustering methods are either not designed for the high-order connectivity patterns, or suffer from the prohibitive computational cost when modeling high-order structures in the large-scale networks. This article generalizes the challenges in multiple dimensions. First ( Model ), we introduce the notion of high-order conductance, and define the high-order diffusion core, which is based on a high-order random walk induced by the user-specified high-order network structure. Second ( Algorithm ), we propose a novel high-order structure-preserving graph clustering framework named HOSGRAP , which partitions the graph into structure-rich clusters in polylogarithmic time with respect to the number of edges in the graph. Third ( Generalization ), we generalize our proposed algorithm to solve the real-world problems on various types of graphs, such as signed graphs, bipartite graphs, and multi-partite graphs. Experimental results on both synthetic and real graphs demonstrate the effectiveness and efficiency of the proposed algorithms.

Community detection on social media has attracted considerable attention for many years. However, existing methods do not reveal the relations between communities. Communities can form alliances or engage in antagonisms due to various factors, e.g., shared or conflicting goals and values. Uncovering such relations can provide better insights to understand communities and the structure of social media. According to social science findings, the attitudes that members from different communities express towards each other are largely shaped by their community membership. Hence, we hypothesize that intercommunity attitudes expressed among users in social media have the potential to reflect their inter-community relations. Therefore, we first validate this hypothesis in the context of social media. Then, inspired by the hypothesis, we develop a framework to detect communities and their relations by jointly modeling users' attitudes and social interactions. We present experimental results using three real-world social media datasets to demonstrate the efficacy of our framework.

Modeling and characterizing high-order connectivity patterns are essential for understanding many complex systems, ranging from social networks to collaboration networks, from finance to neuroscience. However, existing works on high-order graph clustering assume that the input networks are static. Consequently, they fail to explore the rich high-order connectivity patterns embedded in the network evolutions, which may play fundamental roles in real applications. For example, in financial fraud detection, detecting loops formed by sequenced transactions helps identify money laundering activities; in emerging trend detection, star-shaped structures showing in a short burst may indicate novel research topics in citation networks. In this paper, we bridge this gap by proposing a local graph clustering framework that captures structure-rich subgraphs, taking into consideration the information of high-order structures in temporal networks. In particular, our motif-preserving dynamic local graph cut framework (MOTLOC) is able to model various user-defined temporal network structures and find clusters with minimum conductance in a polylogarithmic time complexity. Extensive empirical evaluations on synthetic and real networks demonstrate the effectiveness and efficiency of our MOTLOC framework.

Nowadays, large-scale graph data is being generated in a variety of real-world applications, from social networks to co-authorship networks, from protein-protein interaction networks to road traffic networks. Many existing works on graph mining focus on the vertices and edges, with the first-order Markov chain as the underlying model. They fail to explore the high-order network structures, which are of key importance in many high impact domains. For example, in bank customer personally identifiable information (PII) networks, the star structures often correspond to a set of synthetic identities; in financial transaction networks, the loop structures may indicate the existence of money laundering. In this paper, we focus on mining user-specified high-order network structures and aim to find a structure-rich subgraph which does not break many such structures by separating the subgraph from the rest.

Dense subgraphs are fundamental patterns in graphs, and dense subgraph detection is often the key step of numerous graph mining applications. Most of the existing methods aim to find a single subgraph with a high density. However, dense subgraphs at different granularities could reveal more intriguing patterns in the underlying graph. In this paper, we propose to hierarchically detect dense subgraphs. The key idea of our method (HiDDen) is to envision the density of subgraphs as a relative measure to its background (i.e., the subgraph at the coarse granularity). Given that the hierarchical dense subgraph detection problem is essentially a nonconvex quadratic programming problem, we propose effective and efficient alternative projected gradient based algorithms to solve it. The experimental evaluations on real graphs demonstrate that (1) our proposed algorithms find subgraphs with an up to 40% higher density in almost every hierarchy; (2) the densities of different hierarchies exhibit a desirable variety across different granularities; (3) our projected gradient descent based algorithm scales linearly w.r.t the number of edges of the input graph; and (4) our methods are able to reveal interesting patterns in the underlying graphs (e.g., synthetic ID in financial fraud detection).

In this paper, we propose a new algorithm, called STRICLUSTER, to find tri-clusters from signed 3-partite graphs. The dataset contains three different types of nodes. Hyperedges connecting three nodes from three different partitions represent either positive or negative relations among those nodes. The aim of our algorithm is to find clusters with strong positive relations among its nodes. Moreover, negative relations up to a certain threshold is also allowed. Also, the clusters can have no overlapping hyperedges. We show the effectiveness of our algorithm via several experiments.

Community detection is a fundamental task in social network analysis. In this paper, first we develop an endorsement filtered user connectivity network by utilizing Heider's structural balance theory and certain Twitter triad patterns. Next, we develop three Nonnegative Matrix Factorization frameworks to investigate the contributions of different types of user connectivity and content information in community detection. We show that user content and endorsement filtered connectivity information are complementary to each other in clustering politically motivated users into pure political communities. Word usage is the strongest indicator of users' political orientation among all content categories. Incorporating user-word matrix and word similarity regularizer provides the missing link in connectivity-only methods which suffer from detection of artificially large number of clusters for Twitter networks.

In many real world applications such as image annotation, gene function prediction, and insider threat detection, the data collected from heterogeneous sources often exhibit multiple types of heterogeneity, such as task heterogeneity, view heterogeneity, and label heterogeneity. To address this problem, we propose a Hierarchical Multi-Latent Space (HiMLS) learning framework to jointly model the triple types of heterogeneity. The basic idea is to learn a hierarchical multi-latent space by which we can simultaneously leverage the task relatedness, view consistency and the label correlations to improve the learning performance. We first propose a multi-latent space approach to model the complex heterogeneity, which is then used as a building block to stack up a multi-layer structure in order to learn the hierarchical multi-latent space. In such a way, we can gradually learn the more abstract concepts in the higher level. We present two instantiated models of the generalized framework using different divergence measures. The two-phase learning algorithms are used to train the multi-layer models. We drive the multiplicative update rules for pre-training and fine-tuning in each model, and prove the convergence and correctness of the update methods. The effectiveness of the proposed approach is verified on various data sets.

In this work, a new fast dynamic community detection algorithm for large scale networks is presented. Most of the previous community detection algorithms are designed for static networks. However, large scale social networks are dynamic and evolve frequently over time. To quickly detect communities in dynamic large scale networks, we proposed dynamic modularity optimizer framework (DMO) that is constructed by modifying well-known static modularity based community detection algorithm. The proposed framework is tested using several different datasets. According to our results, community detection algorithms in the proposed framework perform better than static algorithms when large scale dynamic networks are considered.

A major challenge in automated text analysis is that different words are used for related concepts. Analyzing text at the surface level would treat related concepts (i.e. actors, actions, targets, and victims) as different objects, potentially missing common narrative patterns. Shallow parsers reveal semantic roles of words leading to subject-verb-object triplets. We developed a novel algorithm to extract information from triplets by clustering them into generalized concepts by utilizing syntactic criteria based on common contexts and semantic corpus-based statistical criteria based on "contextual synonyms". We show that generalized concepts representation of text (1) overcomes surface level differences (which arise when different keywords are used for related concepts) without drift, (2) leads to a higher-level semantic network representation of related stories, and (3) when used as features, they yield a significant 36% boost in performance for the story detection task.

We propose a multi-scale text mining methodology and develop a visual intelligence platform for tracking the diffusion of online social movements. The algorithms utilize large amounts of text collected from a wide variety of organizations' media outlets to discover their hotly debated topics, and their discriminative perspectives voiced by opposing camps organized into multiple scales. We utilize discriminating perspectives to classify and map individual Tweeter's message content to social movements based on the perspectives expressed in their weekly tweets. We developed a visual intelligence platform, named LookingGlass, to track the geographical footprint, shifting positions and flows of individuals, topics and perspectives between groups.

A story is defined as “an actor(s) taking action(s) that culminates in a resolution(s).” In this paper, we investigate the utility of standard keyword based features, statistical features based on shallow-parsing (such as density of POS tags and named entities), and a new set of semantic features to develop a story classifier. This classifier is trained to identify a paragraph as a “story,” if the paragraph contains mostly story(ies). Training data is a collection of expert-coded story and non-story paragraphs from RSS feeds from a list of extremist web sites. Our proposed semantic features are based on suitable aggregation and generalization of; triplets that can be extracted using a parser. Experimental results show that a model of statistical features alongside memory-based semantic linguistic features achieves the best accuracy with a Support Vector Machine (SVM) classifier.

Internet and social media devices created a new public space for online debate on political and social topics. A debate is defined as a formal discussion on a set of related topics in a public meeting, in which opposing perspectives and arguments are put forward. In this paper, we develop automated perspective discovery techniques which would contribute to the understanding of features (i.e. social, political, cultural, religious beliefs, goals, and practices) shared by each side of the debate. Secondly, we show that, compared to a semi-automated process, our perspective discovery algorithms not only identify larger number of relevant features, but they also yield a higher accuracy scaling of moderate to extreme organizations on both sides of a debate.

In this paper, we develop a model for representing term dependence based on Markov Random Fields and present an approach based on Markov Chain Monte Carlo technique for generating phrase definitions. This approach can use a small corpus of keyword matching and a random sample of other product descriptions for an advertiser's search-phrase to effectively mine and rank alternative but highly relevantsearch-phrasedefinitions. These definitions, which are search-phrases themselves, can then be provided as alternative phrases to an advertiser.

In this paper, our focus will be on weakly annotated data (WAD) which is typically generated by a (semi) automated information extraction system from the Web documents. The extracted information has a certain level of accuracy which can be surpassed by using statistical models that are capable of contextual reasoning such as Bayesian models. Our contribution is an EM algorithm that operates on simple Bayesian models to re-annotate WAD. EM estimates the parameters, i.e., the prior and conditional probabilities by iterating Bayesian model on the given Web data. In the expectation step, Bayesian classifier is trained from current annotations, and in the maximization step, the roles of all the labels are re-annotated to find the best fitting annotation with the current model then the probabilities are re-adjusted from the new annotations. Our experiments show that EM increases the Web data annotation accuracies up to 8%. We use Baum-Welch methodology in our EM approach.

Most keyword search engines return directly matching keyword phrases. However, publishers cannot anticipate all possible ways in which users would search for the items in their documents. In fact, many times, there may be no direct keyword match between a keyword search phrase and descriptions of relevant items that are perfect matches for the search. We present an automated, high precision-based information retrieval solution to boost item find-ability by bridging the semantic gap between item information and popular keyword search phrases. Our solution achieves an average of 80% F-measure for various boosted matches for keyword search phrases in various categories.

Over the past few years World Wide Web has established as a vital resource for news. With the continuous growth in the number of available news Web sites and the diversity in their presentation of content, there is an increasing need to organize the news related information on the Web and keep track of it. In this paper, we present automated techniques for extracting metadata instance information by organizing and mining a set of news Web sites. We develop algorithms that detect and utilize HTML regularities in the Web documents to turn them into hierarchical semantic structures encoded as XML. The tree-mining algorithms that we present identify key domain concepts and their taxonomical relationships. We also extract semi-structured concept instances annotated with their labels whenever they are available. We report experimental evaluation for the news domain to demonstrate the efficacy of our algorithms.

The Web has established itself as the dominant medium for doing electronic commerce. Consequently the number of service providers, both large and small, advertising their services on the web continues to proliferate. In this paper we describe new extraction algorithms for mining service directories from web pages. We develop a novel propagation technique for identifying and accumulating all of the attributes related to a service entity in a web page. We provide experimental results of the effectiveness of our extraction techniques by mining a database of veterinarian service providers from web sources.

Considerable quantities of valuable data about product information and financial statements is often available in sources and formats that are not amenable for querying using traditional database techniques.One such important source is text documents.In such documents these kinds of data often appear in tabular form.A data item in these text tables may span several words (e.g.product description).Furthermore items supposedly within the same column do not necessarily begin or end at the same position.Thus the absence of any regularity in column separators makes it difficult to automatically mine, i.e. extract data items from text tables.Nevertheless an interesting characteristic often exhibited by these tables is that intra-column items are "closer" to each other than inter-column items.We exploit this observation to develop a clustering-based technique to extract data items from these tables.In contrast to previous appproaches, a unique and important aspect of using clustering is that it makes the technique robust in the presence of misalignments.We provide a characterization theorem for text tables on which this technique will always produce a correct extraction.We discuss the design and implementation of a system for extracting tabular data based on this clustering technique.We present experimental evidence of its effectiveness and usability on real industrial data.

The web has established itself as the dominant medium for doing electronic commerce. Realizing that its global reach provides significant market and business opportunities, service providers, both large and small are advertising their services on the web. A number of them operate their own web sites promoting their services at length while others are merely listed in a referral site. Aggregating all of the providers into a queriable service directory makes it easy for customers to locate the one most suited for his/her needs.YellowPager is a tool for creating service directories by mining web sources. Service directories created by YellowPager have several merits compared to those generated by existing practices, which typically require participation by service providers (e.g. Verizon's SuperYellowPages.com). Firstly, the information content will be rich. Secondly since the process is automated and repeatable the content can always be kept current. Finally the same process can be readily adapted to different domains.YellowPager builds service directories by mining the web through a combination of keyword-based search engines,web agents, text classifiers and novel extraction algorithms.The extraction is driven by a services ontology consisting of a taxonomy of service concepts and their associated attributes (such as names and addresses) and type descriptions for the attributes. In addition the ontology also associates an extractor function with each attribute. Applying the function to a web page will identify all the occurrences of the attribute in that page.YellowPager's mining algorithm consists of a training step followed by classification and extraction steps. In the training step a classifier is trained to identify web pages relevant to the service of interest. The classification step proceeds by doing a search for the particular service of interest using a keyword based web search engine and retrieves all the matching web pages. From these pages the relevant ones are identified using the classifier. The final step is extraction of attribute values, associated with the service, from these pages. Each web page is parsed into a DOM tree and the extractor functions are applied. All of the attributes corresponding to a service provider are then correctly aggregated. This can pose difficulties especially in the presence of multiple service providers in a page. Using a novel concept of scoring and conflict resolution to prevent erroneous associations of attributes with service provider entities in the page, the algorithm aggregates all the attribute occurrences correctly. The extractor function may not be complete in the sense that it cannot always identify all the attributes in a page. By exploiting the regularity of the sequence in which attributes occurr in referral pages, the mining algorithm automatically learns generalized patterns to locate attributes that the extractor function misses. The distinguishing aspects of YellowPager's extraction algorithm are: (i) it is unsupervised, and (ii) the attribute values in the pages are extracted independent of any page-specific relationships that may exist among the markup tags.YellowPager has been used by a large pet food producer to build a directory of veterinarian service providers in the United States. The resulting database was found to be much larger and richer than that found in Vetquest, Vetworld, and the Super Yellow pages.YellowPager is implemented in JAVA and is interfaced to Rainbow, a library utility in C that is used for classification. The tool will demonstrate the creation of a service directory for any service domain by mining web sources.

A wealth of information relevant for e-commerce often appears in text form. This includes specification and performance data sheets of products, financial statements, product offerings etc. Typically these types of product and financial data are published in tabular form. The only separators between items in the table are white spaces and line separators. We will refer to such tables as text tables. Due to the lack of structure in such tables, the information present is not readily queriable using traditional database query languages like SQL. One way to make it amenable to standard database querying techniques is to extract the data items in the tables and create a database out of the extracted data. But extraction from text tables poses difficulties due to the irregularity of the data in the column.

Automatic data extraction from semistructured sources such as HTML pages is rapidly growing into a problem of significant importance, spurred by the growing popularity of the so called “shopbots” that enable end users to compare prices of goods and other services at various web sites without having to manually browse and fill out forms at each one of these sites.

The design of webbases, database systems for supporting Web-based applications, is currently an active area of research. In this paper, we propose a 3-year architecture for designing and implementing webbases for querying dynamic Web content(i.e., data that can only be extracted by filling out multiple forms). The lowest layer, virtual physical layer, provides navigation independence by shielding the user from the complexities associated with retrieving data from raw Web sources. Next, the traditional logical layer supports site independence. The top layer is analogous to the external schema layer in traditional databases.

WC propose Concurrent Transaction Logic (C7X) as the language for specifying, analyzing, and scheduling of workflows.We show that both local and global properties of worktlows can be naturally represented as C7X formulas and reasoning can be done with the use of the proof theory and the semantics of this logic, We describe a transformation that leads to an eilicicnt algorithm for scheduling worldlows in the presencc of global temporal constraints, which leads to decision proccdurcs for dealing with several safety related properties such as whether every valid execution of the workflow satisfits a particular property or whether a worlcfiow execution is consistent with some given global constraints on the ordering of events in a workflow.We also provide tight complexity results on the running times of these algorithms.