Matthew D. Schmill

Language models for financial news recommendation

ABSTRACT We present a unique approa h to identifying news stories that in uen e the behavior of nan ial markets. Spe i ally, we des ribe the design and implementation of nalyst, a system that an re ommend interesting news stories { stories that are likely to a e t market behavior. nalyst operates by orrelating the ontent of news stories with trends in nan ial time series. We identify trends in time series using pie ewise linear tting and then assign labels to the trends a ording to an automated binning pro edure. We use language models to represent patterns of language that are highly asso iated with parti ular labeled trends. nalyst an then identify and re ommend news stories that are highly indi ative of future trends. We evaluate the system in terms of its ability to re ommend the stories that will a e t the behavior of the sto k market. We demonstrate that stories re ommended by nalyst ould be used to pro tably predi t forth oming trends in sto k pri es.

Parallel and Distributed Search for Structure in Multivariate Time Series

Efficient data mining algorithms are crucial for effective knowledge discovery. We present the Multi-Stream Dependency Detection (Msdd) data mining algorithm that performs a systematic search for structure in multivariate time series of categorical data. The systematicity of Msdds search makes implementation of both parallel and distributed versions straightforward. Distributing the search for structure over multiple processors or networked machines makes mining of large numbers of databases or very large databases feasible. We present results showing that msdd efficiently finds complex structure in multivariate time series, and that the distributed version finds the same structure in approximately

Detecting Complex Dependencies in Categorical Data

1/n

Learning what is relevant to the effects of actions for a mobile robot

of the time required by Msdd, where

Ambiguous Geographies: Connecting Case Study Knowledge with Global Change Science

n

Identifying qualitatively different outcomes of actions: gaining autonomy through learning

is the number of machines across which the search is distributed. msdd differs from other data mining algorithms in the complexity of the structure that it can find. msdd also requires no domain knowledge to focus or limit its search, although such knowledge is easily incorporated when it is available.

A motivational system that drives the development of activity

Locating and evaluating relationships among values in multiple streams of data is a difficult and important task. Consider the data flowing from monitors in an intensive care unit. Readings from various subsets of the monitors are indicative and predictive of certain aspects of the patient’s state. We present an algorithm that facilitates discovery and assessment of the strength of such predictive relationships called Multi-stream Dependency Detection (MSDD). We use heuristic search to guide our exploration of the space of potentially interesting dependencies to uncover those that are significant. We begin by reviewing the dependency detection technique described in [3], and extend it to the multiple stream case, describing in detail our heuristic search over the space of possible dependencies. Quantitative evidence for the utility of our approach is provided through a series of experiments with artificially-generated data. In addition, we present results from the application of our algorithm to two real problem domains: feature-based classification and prediction of pathologies in a simulated shipping network.

Contextualizing the global relevance of local land change observations

We have developed a learning mechanism that allows robots to discover the conditional effects of their actions. Based on sansorimotor experience, this mechanism permits a robot to explore its environment and observe effects of its actions, These observations are used to learn a con&t operalor diflcrcncc taldc, a structure that relates circumstances (context) and actions (operators) to effects on the environmom, From the context operator difference table, one can extract a relatively small set of state variables, which simplifies tho problem of learning policies for complex activities. We demonstrate results with the Pioneer 1 mobile robot.

Managing Uncertainty in Text-to-Sketch Tracking Problems

Case studies have long been a gold standard for investigating causal mechanisms in human–environment interactions. Yet it remains a challenge to generalize across case studies to produce knowledge at broader regional and global scales even as the effort to do so, mostly using metastudy methods, has accelerated. One major obstacle is that the geographic context of case study knowledge is often presented in a vague and incomplete form, making it difficult to reuse and link with the regional and global contexts within which it was produced and is therefore most relevant. Here we assess the degree to which the quality of geographic description in published land change case studies limits their effective reuse in spatially explicit global and regional syntheses based on 437 spatially bounded cases derived from 261 case studies used in published land change metastudies. Common ambiguities in published representations of case geographic contexts were identified and scored using three indicators of geographic data quality for reuse in spatially explicit regional and global metastudy research. Statistically significant differences in the quality of case geographic descriptions were evident among the six major disciplinary categories examined, with the earth and planetary sciences evidencing greater clarity and conformance scores than other disciplines. The quality of case geography reporting showed no statistically significant improvement over the past fifty years. By following a few simple and readily implemented guidelines, case geographic context reporting could be radically improved, enabling more effective case study reuse in regional to global synthesis research, thereby yielding substantial benefits to both case study and synthesis researchers.

Tools for detecting dependencies in AI systems

If robotic agents are to act autonomously they must have the ability to construct and reason about models of their physical environment. In all but the simplest, static domains, such models must represent the dynamics of environmental change. For example, because the effects of actions are not instantaneous, planning to achieve goals requires knowledge of how the robot s actions affect the s tate of the world over time. The tradit ional approach of hand-coding this knowledge is often quite difficult, especially for robotic agents with rich sensing abilities tha t exist in dynamic and uncertain environments. Ideally, agents would acquire knowledge of their environment autonomously, and then use this knowledge to act autonomously. This paper presents an unsupervised method for learning models of environmental dynamics based on clustering mult ivariate t ime series. Experiments with a Pioneer-1 mobile robot demonst ra te the utility of the method and show tha t the models acquired by the robot correlate surprisingly well with human models of the environment. Individual t ime series are obtained by recording the output of a subset of an agent s sensors. We call these t ime series experiences. An example of a sensor subset on the Pioneer-1 robot is its array of seven sonars. Each sonar returns the distance to the closest object in the direction tha t it points. Recording of t ime series is usually triggered by events, such as the initiation of a particular action. Each t ime a given event occurs, the t ime series tha t was recorded is added to a bucket associated with tha t event. Once a sufficient number of experiences are recorded, clusters can be formed. Clustering requires a measure of similarity between mult ivariate t ime series. One such measure tha t is part icularly appropriate for this problem is Dynamic Time Warping