Supporting the efficient discovery and use of Web APIs is increasingly important as their use and popularity grows. Yet, a simple task like finding potentially interesting APIs and their related documentation turns out to be hard and time consuming even when using the best resources currently available on the Web. In this paper we describe our research towards an automated Web API documentation crawler and search engine. This paper presents two main contributions. First, we have devised and exploited crowdsourcing techniques to generate a curated dataset of Web APIs documentation. Second, thanks to this dataset, we have devised an engine able to automatically detect documentation pages. Our preliminary experiments have shown that we obtain an accuracy of 80% and a precision increase of 15 points over a keyword-based heuristic we have used as baseline.

Learning algorithms normally assume that there is at most one annotation or label per data point. However, in some scenarios, such as medical diagnosis and on-line collaboration,multiple annotations may be available. In either case, obtaining labels for data points can be expensive and time-consuming (in some circumstances ground-truth may not exist). Semi-supervised learning approaches have shown that utilizing the unlabeled data is often beneficial in these cases. This paper presents a probabilistic semi-supervised model and algorithm that allows for learning from both unlabeled and labeled data in the presence of multiple annotators. We assume that it is known what annotator labeled which data points. The proposed approach produces annotator models that allow us to provide (1) estimates of the true label and (2) annotator variable expertise for both labeled and unlabeled data. We provide numerical comparisons under various scenarios and with respect to standard semi-supervised learning. Experiments showed that the presented approach provides clear advantages over multi-annotator methods that do not use the unlabeled data and over methods that do not use multi-labeler information.

Gaussian processes (GPs) provide a probabilistic nonparametric representation of functions in regression, classification, and other problems. Unfortunately, exact learning with GPs is intractable for large datasets. A variety of approximate GP methods have been proposed that essentially map the large dataset into a small set of basis points. Among them, two state-of-the-art methods are sparse pseudo-input Gaussian process (SPGP) (Snelson and Ghahramani, 2006) and variablesigma GP (VSGP) Walder et al. (2008), which generalizes SPGP and allows each basis point to have its own length scale. However, VSGP was only derived for regression. In this paper, we propose a new sparse GP framework that uses expectation propagation to directly approximate general GP likelihoods using a sparse and smooth basis. It includes both SPGP and VSGP for regression as special cases. Plus as an EP algorithm, it inherits the ability to process data online. As a particular choice of approximating family, we blur each basis point with a Gaussian distribution that has a full covariance matrix representing the data distribution around that basis point; as a result, we can summarize local data manifold information with a small set of basis points. Our experiments demonstrate that this framework outperforms previous GP classification methods on benchmark datasets in terms of minimizing divergence to the non-sparse GP solution as well as lower misclassification rate.

Imitation learning in robots, also called programing by demonstration, has made important advances in recent years, allowing humans to teach context dependant motor skills/tasks to robots. We propose to extend the usual contexts investigated to also include acoustic linguistic expressions that might denote a given motor skill, and thus we target joint learning of the motor skills and their potential acoustic linguistic name. In addition to this, a modification of a class of existing algorithms within the imitation learning framework is made so that they can handle the unlabeled demonstration of several tasks/motor primitives without having to inform the imitator of what task is being demonstrated or what the number of tasks are, which is a necessity for language learning, i.e; if one wants to teach naturally an open number of new motor skills together with their acoustic names. Finally, a mechanism for detecting whether or not linguistic input is relevant to the task is also proposed, and our architecture also allows the robot to find the right framing for a given identified motor primitive. With these additions it becomes possible to build an imitator that bridges the gap between imitation learning and language learning by being able to learn linguistic expressions using methods from the imitation learning community. In this sense the imitator can learn a word by guessing whether a certain speech pattern present in the context means that a specific task is to be executed. The imitator is however not assumed to know that speech is relevant and has to figure this out on its own by looking at the demonstrations: indeed, the architecture allows the robot to transparently also learn tasks which should not be triggered by an acoustic word, but for example by the color or position of an object or a gesture made by someone in the environment. To demonstrate this ability to find the ...

This paper presents a new approach to disambiguate company names in the Twitter social network. We have focused on making lighter the processing of comparing company profiles with tweets in order to obtain a competitive real-time system. With this aim, we only use the home page of each company as information source to create a unique profile. On the other hand, we compute the similarity of a tweet in connection to a profile by comparing the content of the tweet with the profile. Both steps do not use any other external information source and all the process is developed in an unsupervised way. We have tested our application with the test WePS-3 CLEF ORM corpus obtaining encouraging results.

Micro-blogging sites such as Facebook, Twitter, Google+ present a nice opportunity for targeting advertisements that are contextually related to the microblog content. By virtue of the sparse and noisy text makes identifying the microblogs suitable for advertising a very hard problem. In this work, we approach the problem of identifying the microblogs that could be targeted for advertisements as a two-step classification approach. In the first pass, microblogs suitable for advertising are identified. Next, in the second pass, we build a model to find the sentiment of the advertisable microblog. The systems use features derived from the Part-of-speech tags, the tweet content and uses external resources such as query logs and n-gram dictionaries from previously labeled data.This work aims at providing a thorough insight into the problem and analyzing various features to assess which features contribute the most towards identifying the tweets that can be targeted for advertisements.

Today, YouTube is the largest user-driven video content provider in the world; it has become a major platform for disseminating multimedia information. A major contribution to its success comes from the user-to-user social experience that differentiates it from traditional content broadcasters. This work examines the social network aspect of YouTube by measuring the full-scale YouTube subscription graph, comment graph, and video content corpus. We find YouTube to deviate significantly from network characteristics that mark traditional online social networks, such as homophily, reciprocative linking, and assortativity. However, comparing to reported characteristics of another content-driven online social network, Twitter, YouTube is remarkably similar. Examining the social and content facets of user popularity, we find a stronger correlation between a user's social popularity and his/her most popular content as opposed to typical content popularity. Finally, we demonstrate an application of our measurements for classifying YouTube Partners, who are selected users that share YouTube's advertisement revenue. Results are motivating despite the highly imbalanced nature of the classification problem.

Relational classification has been extensively studied recently due to its applications in social, biological, technological, and information networks. Much of the work in relational learning has focused on analyzing input data that comprise a single network. Although machine learning researchers have considered the issue of how to sample training and test sets from the input network (for evaluation), the mechanisms which are used to construct the input networks have largely been ignored. In most cases, the input network has itself been sampled from a larger target network (e.g., Facebook) and often the researcher is unaware of how the input network was constructed or what impact that may have on evaluation of the relational models. Since the goal in evaluating relational classification algorithms is to accurately assess their performance on the larger target network, it is critical to understand what impact the initial sampling method may have on our estimates of classification accuracy.In this paper, we present different sampling methods and systematically study their impact on evaluation of relational classification. Our results indicate that the choice of sampling method can impact classification performance, and thus consequently affects the accuracy of evaluation.

Social media profiles are telling examples of the everyday need for disclosure and concealment. The balance between concealment and disclosure varies across individuals, and personality traits might partly explain this variability. Experimental findings on the relationship between information disclosure and personality have been so far inconsistent. We thus study this relationship anew with 1,313 Facebook users in the United States using two personality tests: the big five personality test and the self-monitoring test. We model the process of information disclosure in a principled way using Item Response Theory and correlate the resulting user disclosure scores with personality traits. We find a correlation with the trait of Openness and observe gender effects, in that, men and women share equal amount of private information, but men tend to make it more publicly available, well beyond their social circles. Interestingly, geographic (e.g., residence, hometown) and work-related information is used as relationship currency, in that, it is selectively shared with social contacts and is rarely shared with the Facebook community at large.

Merger and Acquisition (M&A) prediction has been an interesting and challenging research topic in the past a few decades. However, past work has only adopted numerical features in building models, and yet the valuable textual information from the great variety of social media sites has not been touched at all. To fully explore this information, we used the profiles and news articles for companies and people on TechCrunch, the leading and largest public database for the tech world, which anybody can edit. Specifically, we explored topic features via topic modeling techniques, as well as a set of other novel features of our design within a machine learning framework. We conducted experiments of the largest scale in the literature, and achieved a high true positive rate (TP) between 60% to 79.8% with a false positive rate (FP) mostly between 0% and 8.3% over company categories with a small number of missing attributes in the CrunchBase profiles.