User-Generated text is a rich source of user insights and experiences that can be very helpful in many different daily life situations, such as when deciding what product to buy, what hotel to stay, what company to apply for a job, what region to buy a house, etc. This kind of text also plays a very relevant role in current research efforts in academic research groups, technology companies, as well as big publishers, telecommunications players, recruiting and job-market focuses organizations, etc. The goal of this new workshop is to bring together researchers interested in the application of novel techniques in AI/ML/NLP and Knowledge Discovery to address challenges around harnessing text-heavy user-generated data that is available to organizations and over the Web. The workshop program contains invited speakers, contributed talks, poster sessions and a discussion panel

Creating and collecting labeled data is one of the major bottlenecks in machine learning pipelines and the emergence of automated feature generation techniques such as deep learning, which typically requires a lot of training data, has further exacerbated the problem. While weak-supervision techniques have circumvented this bottleneck, existing frameworks either require users to write a set of diverse, high-quality rules to label data (e.g., Snorkel), or require a labeled subset of the data to automatically mine rules (e.g., Snuba). The process of manually writing rules can be tedious and time consuming. At the same time, creating a labeled subset of the data can be costly and even infeasible in imbalanced settings.

We present Emu, a system that semantically enhances multilingual sentence embeddings. Our framework fine-tunes pre-trained multilingual sentence embeddings using two main components: a semantic classifier and a language discriminator. The semantic classifier improves the semantic similarity of related sentences, whereas the language discriminator enhances the multilinguality of the embeddings via multilingual adversarial training. Our experimental results based on several language pairs show that our specialized embeddings outperform the state-of-the-art multilingual sentence embedding model on the task of cross-lingual intent classification using only monolingual labeled data.

Subjectivity is the expression of internal opinions or beliefs which cannot be objectively observed or verified, and has been shown to be important for sentiment analysis and wordsense disambiguation.Furthermore, subjectivity is an important aspect of user-generated data.In spite of this, subjectivity has not been investigated in contexts where such data is widespread, such as in question answering (QA).We develop a new dataset which allows us to investigate this relationship.We find that subjectivity is an important feature in the case of QA, albeit with more intricate interactions between subjectivity and QA performance than found in previous work on sentiment analysis.For instance, a subjective question may or may not be associated with a subjective answer.We release an English QA dataset (SUBJQA) based on customer reviews, containing subjectivity annotations for questions and answer spans across 6 domains.

Traditional recommender systems aim to satisfy individual users by providing them with recommendations that match their preferences. Such recommender systems don't take into consideration how the number of users recommended to use a particular item affects the users' experience. For example, a highly-recommended restaurant may match the preferences of many users. However, increasing its popularity via recommendations may make the experience unsatisfactory due to high volume of customers, long lines and inevitably slow service. In this paper, we develop a new recommendation-system paradigm that we call collective recommendations. Collective recommendations take into consideration not only the user preferences, but also the effect of the popularity of a venue to the overall user experience. We formally define the algorithmic problems motivated by collective recommendations and develop an algorithmic framework for solving them effectively. Our experiments with real data demonstrate the effectiveness of our methods in practice. Nobody goes there any more. It's too crowded — Yogi Berra

We present the Koko system that takes declarative information extraction to a new level by incorporating advances in natural language processing techniques in its extraction language. K oko is novel in that its extraction language simultaneously supports conditions on the surface of the text and on the structure of the dependency parse tree of sentences, thereby allowing for more refined extractions. K oko also supports conditions that are forgiving to linguistic variation of expressing concepts and allows to aggregate evidence from the entire document in order to filter extractions. To scale up, K oko exploits a multi-indexing scheme and heuristics for efficient extractions. We extensively evaluate K oko over publicly available text corpora. We show that K oko indices take up the smallest amount of space, are notably faster and more effective than a number of prior indexing schemes. Finally, we demonstrate K oko 's scalability on a corpus of 5 million Wikipedia articles.

K oko is a declarative information extraction system that incorporates advances in natural language processing techniques in its extraction language. K oko 's extraction language supports simultaneous specification of conditions over the surface syntax and on the structure of the dependency parse tree of sentences, thereby allowing for more refined extractions. Furthermore, the K oko extraction language allows for aggregating evidence from an input document and supports conditions that are tolerant of linguistic variation of expressing concepts. In this demo, we outline the design of K oko , a system for extracting information and understanding the results of the extraction. K oko provides an interactive interface that allows participants to write queries, understand the input and results of the queries. In particular, the user can customize the input text, visualize the input text's dependency parse trees, and understand the correspondences between query components, dependency tree nodes, text tokens, and the computation and associated scores that led to an extraction.

In large organizations (e.g., companies, universities, etc.) individual experts with different work habits are asked to work together in order to complete projects or tasks. Oftentimes, the differences in the inherent work habits of these experts causes tension among them, which can prove detrimental for the organization's performance and functioning. The question we consider in this paper is the following: "can this tension be reduced by providing incentives to individuals to change their work habits?" We formalize this question in the definition of the k- AlterHabit problem. To the best of our knowledge we are the first to define this problem and analyze its properties. Although we show that k- AlterHabit is NP-hard, we devise polynomial-time algorithms for solving it in practice. Our algorithms are based on interesting connections that we draw between our problem and other combinatorial problems. Our experimental results demonstrate both the efficiency and the efficacy of our algorithmic techniques on a collection of real data.

The field of data integration has expanded significantly over the years, from providing a uniform query and update interface to structured databases within an enterprise to the ability to search, ex- change, and even update, structured or unstructured data that are within or external to the enterprise. This paper describes the evolution in the landscape of data integration since the work on rewriting queries using views in the mid-1990's. In addition, we describe two important challenges for the field going forward. The first challenge is to develop good open-source tools for different components of data integration pipelines. The second challenge is to provide practitioners with viable solutions for the long-standing problem of systematically combining structured and unstructured data.

Motivated by applications that arise in online social media and collaboration networks, there has been a lot of work on community-search. In this class of problems, the goal is to find a subgraph that satisfies a certain connectivity requirement and contains a given collection of seed nodes. In this paper, we extend the community-search problem by associating each individual with a profile. The profile is a numeric score that quantifies the position of an individual with respect to a topic. We adopt a model where each individual starts with a latent profile and arrives to a conformed profile through a dynamic conformation process, which takes into account the individual's social interaction and the tendency to conform with one's social environment. In this framework, social tension arises from the differences between the conformed profiles of neighboring individuals as well as from the differences between individuals' conformed and latent profiles. Given a network of individuals, their latent profiles and this conformation process, we extend the community-search problem by requiring the output subgraphs to have low social tension. From the technical point of view, we study the complexity of this problem and propose algorithms for solving it effectively. Our experimental evaluation in a number of social networks reveals the efficacy and efficiency of our methods.

Access to diverse perspectives nurtures an informed citizenry. Google and Bing have emerged as the duopoly that largely arbitrates which English-language documents are seen by web searchers. We present our empirical study over the search results produced by Google and Bing that shows a large overlap. Thus, citizens may not gain different perspectives by simultaneously probing them for the same query. Fortunately, our study also shows that by mining Twitter data, one can obtain search results that are quite distinct from those produced by Google, Bing, and Bing News. Additionally, the users found those results to be quite informative. We also present two novel tools we designed for this study. One uses tensor analysis to derive low-dimensional compact representation of search results and study their behavior over time. The other uses machine learning and quantifies the similarity of results between two search engines by framing it as a prediction problem. Although these tools have different underpinnings, the analytical results obtained using them corroborate each other, which reinforces the confidence one can place in them for finding meaningful insights from big data.

Access to diverse perspectives nurtures an informed citizenry. Google and Bing have emerged as the duopoly that largely arbitrates which English language documents are seen by web searchers. A recent study shows that there is now a large overlap in the top organic search results produced by them. Thus, citizens may no longer be able to gain different perspectives by using different search engines.

Given a class of large number of students, each exhibiting a different ability level, how can we group them into sections so that the overall gain for students is maximized? This question has been a topic of central concern and debate amongst social scientists and policy makers for a long time. We propose a framework for rigorously studying this question, taking a computational perspective. We present a formal definition of the grouping problem and investigate some of its variants. Such variants are determined by the desired number of groups as well as the definition of the gain for each student in the group. We focus on two natural instantiations of the gain function and we show that for both of them the problem of identifying a single group of students that maximizes the gain among its members can be solved in polynomial time. The corresponding partitioning problem, where the goal is to partition the students into non-overlapping groups appear to be much harder. However, the algorithms for the single-group version can be leveraged for solving the more complex partitioning problem. Our experiments with generated data coming from different distributions demonstrate that our algorithm is significantly better than the current strategies in vogue for dividing students in a class into sections.

Team formation has been long recognized as a natural way to acquire a diverse pool of useful skills, by combining experts with complementary talents. This allows organizations to effectively complete beneficial projects from different domains, while also helping individual experts position themselves and succeed in highly competitive job markets. Here, we assume a collection of projects ensuremath{P}, where each project requires a certain set of skills, and yields a different benefit upon completion. We are further presented with a pool of experts ensuremath{X}, where each expert has his own skillset and compensation demands. Then, we study the problem of hiring a cluster of experts T ⊆ X, so that the overall compensation (cost) does not exceed a given budget B, and the total benefit of the projects that this team can collectively cover is maximized. We refer to this as the ClusterHire problem. Our work presents a detailed analysis of the computational complexity and hardness of approximation of the problem, as well as heuristic, yet effective, algorithms for solving it in practice. We demonstrate the efficacy of our approaches through experiments on real datasets of experts, and demonstrate their advantage over intuitive baselines. We also explore additional variants of the fundamental problem formulation, in order to account for constraints and considerations that emerge in realistic cluster-hiring scenarios. All variants considered in this paper have immediate applications in the cluster hiring process, as it emerges in the context of different organizational settings.

Given a class of large number of students, each exhibiting a different ability level, how can we form teams of students so that the expected performance of team members improves due to team participation? We take a computational perspective and formally define two versions of such team-formation problem: the MAXTEAM and the MAXPARTITION problems. The first asks for the identification of a single team of students that improves the performance of most of the participating team members. The second asks for a partitioning of students into non-overlapping teams that also maximizes the benefit of the participating students. We show that the first problem can be solved optimally in polynomial time, while the second is NP-complete. For the MAXPARTITION problem, we also design an efficient approximate algorithm for solving it. Our experiments with generated data coming from different distributions demonstrate that our algorithm is significantly better than any of the popular strategies for dividing students in a class into sections.

Motivated by emerging applications including vehicular traffic networks and crowdsourcing systems, we introduce the Unveil problem defined as follows: given a system of linear equations with less equations than unknowns (i.e., an underdetermined system), identify a set of k variables to query their values so that the number of variables that can be uniquely deduced in the new system is maximized. In this paper, we study the complexity of the Unveil problem and several of its variants and we obtain the following results. In Unveil, when the variables of the linear system take real values, then the problem is NP-hard to solve or even approximate. These inapproximability results carry over to the case where the queries may involve linear combinations of variables. When the variables in the linear equations are binary, then the corresponding binary version of the UNVEIL problem becomes coNP-hard to solve or even approximate. In order to explore the practical significance of our formulation we also develop a set of heuristics for the basic Unveil problem and demonstrate that choosing the queried variables using heuristic algorithms is much more effective than random querying.

Numerous datasets ranging from group memberships within social networks to purchase histories on e-commerce sites are represented by binary matrices. While this data is often either proprietary or sensitive, aggregated data, notably row and column marginals, is often viewed as much less sensitive, and may be furnished for analysis. Here, we investigate how these data can be exploited to make inferences about the underlying matrix H. Instead of assuming a generative model for H, we view the input marginals as constraints on the dataspace of possible realizations of H and compute the probability density function of particular entries H(i,j) of interest. We do this, for all the cells of H simultaneously, without generating realizations but rather via implicitly sampling the datasets that satisfy the input marginals. The end result is an efficient algorithm with running time equal to the time required by standard sampling techniques to generate a single dataset from the same dataspace. Our experimental evaluation demonstrates the efficiency and the efficacy of our framework in multiple settings.

When working on a new project, researchers need to devote a significant amount of time and effort to surveying the relevant literature. This is required in order to gain expertise, evaluate the significance of their work and gain useful insights about a particular scientific domain. While necessary, relevant-work search is also a time-consuming and arduous process, requiring the continuous participation of the user. In this work, we introduce Sofia Search, a tool that fully automates the search and retrieval of the literature related to a topic. Given a seed of papers submitted by the user, Sofia Search searches the Web for candidate related papers, evaluates their relevance to the seed and downloads them for the user. The tool also provides modules for the evaluation and ranking of authors and papers, in the context of the retrieved papers. In the demo, we will demonstrate the functionality of our tool, by allowing users to use it via a simple and intuitive interface.