In many classification problems labels are relatively scarce. One context in which this occurs is where we have labels for groups of instances but not for the instances themselves, as in multi-instance learning. Past work on this problem has typically focused on learning classifiers to make predictions at the group level. In this paper we focus on the problem of learning classifiers to make predictions at the instance level. To achieve this we propose a new objective function that encourages smoothness of inferred instance-level labels based on instance-level similarity, while at the same time respecting group-level label constraints. We apply this approach to the problem of predicting labels for sentences given labels for reviews, using a convolutional neural network to infer sentence similarity. The approach is evaluated using three large review data sets from IMDB, Yelp, and Amazon, and we demonstrate the proposed approach is both accurate and scalable compared to various alternatives.

No abstract

Objective Amid electronic health records, laboratory tests, and other technology, office-based patient and provider communication is still the heart of primary medical care. Patients typically present multiple complaints, requiring physicians to decide how to balance competing demands. How this time is allocated has implications for patient satisfaction, payments, and quality of care. We investigate the effectiveness of machine learning methods for automated annotation of medical topics in patient-provider dialog transcripts. Materials and Methods We used dialog transcripts from 279 primary care visits to predict talk-turn topic labels. Different machine learning models were trained to operate on single or multiple local talk-turns (logistic classifiers, support vector machines, gated recurrent units) as well as sequential models that integrate information across talk-turn sequences (conditional random fields, hidden Markov models, and hierarchical gated recurrent units). Results Evaluation was performed using cross-validation to measure 1) classification accuracy for talk-turns and 2) precision, recall, and F1 scores at the visit level. Experimental results showed that sequential models had higher classification accuracy at the talk-turn level and higher precision at the visit level. Independent models had higher recall scores at the visit level compared with sequential models. Conclusions Incorporating sequential information across talk-turns improves the accuracy of topic prediction in patient-provider dialog by smoothing out noisy information from talk-turns. Although the results are promising, more advanced prediction techniques and larger labeled datasets will likely be required to achieve prediction performance appropriate for real-world clinical applications.

A large number of mainstream applications, like temporal search, event detection, and trend identification, assume knowledge of the timestamp of every document in a given textual collection. In many cases, however, the required timestamps are either unavailable or ambiguous. A characteristic instance of this problem emerges in the context of large repositories of old digitized documents. For such documents, the timestamp may be corrupted during the digitization process, or may simply be unavailable. In this paper, we study the task of approximating the timestamp of a document, so-called document dating. We propose a contentbased method and use recent advances in the domain of term burstiness, which allow it to overcome the drawbacks of previous document dating methods, e.g. the fix time partition strategy. We use an extensive experimental evaluation on different datasets to validate the efficacy and advantages of our methodology, showing that our method outperforms the state of the art methods on document dating.

Micro-blogging services such as Twitter have gained enormous popularity over the last few years leading to massive volumes of user generated content. A portion of this content is shared via geoaware mobile devices, such as smartphones. Pieces of information shared on such a device can be tagged with the user's location, conditional on the user's settings. These geostamps enable a number of mainstream applications, such as emergency response, disease tracking, news reporting, and advertising. Unfortunately, informative geostamps are typically sparse, since content is often shared via devices that do not support geo-tagging, such as desktop or laptop computers. In addition, even if a mobile device is used, a flawed geo-location service can lead to missing geostamps, or geostamps that are too general to be informative. In this work, we address this sparsity issue via a new approach that identifies users attached to a given location of interest, such as a city. We then focus on retrieving specific tweets at a finer granularity within the given location, such as specific blocks within a city. Our approach leverages the correlation between strong connectivity in the social graph and proximity in the real world, while utilizing both textual tweet content and Twitter's underlying social graph. Previous relevant work assumes that all required Twitter data is available without access restrictions. This is an unrealistic assumption, since Twitter limits the number of data requests per user and charges a subscription fee for unrestricted access. Therefore, in order to increase the number of practitioners and applications that can benefit from our work, we optimize our method to work with the minimum amount of queries to the Twitter API. Finally, our experiments demonstrate the efficacy of our work via both a quantitative and qualitative evaluation.