Arti Ramesh

Understanding MOOC Discussion Forums using Seeded LDA

Discussion forums serve as a platform for student discussions in massive open online courses (MOOCs). Analyzing content in these forums can uncover useful information for improving student retention and help in initiating instructor intervention. In this work, we explore the use of topic models, particularly seeded topic models toward this goal. We demonstrate that features derived from topic analysis help in predicting student survival.

Weakly Supervised Models of Aspect-Sentiment for Online Course Discussion Forums

Massive open online courses (MOOCs) are redefining the education system and transcending boundaries posed by traditional courses. With the increase in popularity of online courses, there is a corresponding increase in the need to understand and interpret the communications of the course participants. Identifying topics or aspects of conversation and inferring sentiment in online course forum posts can enable instructor interventions to meet the needs of the students, rapidly address course-related issues, and increase student retention. Labeled aspect-sentiment data for MOOCs are expensive to obtain and may not be transferable between courses, suggesting the need for approaches that do not require labeled data. We develop a weakly supervised joint model for aspectsentiment in online courses, modeling the dependencies between various aspects and sentiment using a recently developed scalable class of statistical relational models called hinge-loss Markov random fields. We validate our models on posts sampled from twelve online courses, each containing an average of 10,000 posts, and demonstrate that jointly modeling aspect with sentiment improves the prediction accuracy for both aspect and sentiment.

Uncovering hidden engagement patterns for predicting learner performance in MOOCs

Maintaining and cultivating student engagement is a prerequisite for MOOCs to have broad educational impact. Understanding student engagement as a course progresses helps characterize student learning patterns and can aid in minimizing dropout rates, initiating instructor intervention. In this paper, we construct a probabilistic model connecting student behavior and class performance, formulating student engagement types as latent variables. We show that our model identifies course success indicators that can be used by instructors to initiate interventions and assist students.

Topic Evolution Models for Long-Running MOOCs

Massive open online courses (MOOCs) have emerged as a powerful platform for imparting education in the last few years. Discussion forums in online courses connect various geographically separated MOOC participants and serve as the primary means of communication between them. The text in the forums reflects many important aspects of the course such as the student population and their changing interests, parts of the course that were well received and parts needing attention, and common misconceptions faced by students. In order to improve the quality of online courses and students’ interaction and learning experience, instructors need to actively monitor and discern patterns in previous iterations of the course and mold the course to suit the needs of the ever-changing student population. To enable this, in this work, we perform a systematic detailed analysis of the evolution of fine-grained topics in online course discussion forums across repeated MOOC offerings using seeded topic models and draw important insights on the nature of students, types of issues, and student satisfaction. We present topic evolution results on two successful long-running MOOCs: (i) a business course, and (ii) a computer science course. Our models uncover interesting topic trends in both courses including the decline of logistic issues in both courses as iterations unfold, decline in grading related issues when automatic grading is adopted in the business course, and prevalence of technical issues in the computer science course in comparison to the business course. Our models throw light on the different ways students interact on MOOCs and their changing needs, and are useful for instructors to understand the progression of courses and accordingly fine-tune courses to meet student expectations.

A Structured Approach to Understanding Recovery and Relapse in AA

Alcoholism, also known as Alcohol Use Disorder (AUD), is a serious problem affecting millions of people worldwide. Recovery from AUD is known to be challenging and often leads to relapse at various points after enrolling in a rehabilitation program such as Alcoholics Anonymous (AA). In this work, we take a structured approach to understand recovery and relapse from AUD using social media data. To do so, we combine linguistic and psychological attributes of users with relational features that capture useful structure in the user interaction network. We evaluate our models on AA-attending users extracted from the Twitter social network and predict recovery at two different points---90 days and 1 year after the user joins AA, respectively. Our experiments reveal that our structured approach is helpful in predicting recovery in these users. We perform extensive quantitative analysis of different groups of features and dependencies among them. Our analysis sheds light on the role of each feature group and how they combine to predict recovery and relapse. Finally, we present a qualitative analysis of the different reasons behind users relapsing to AUD. Our models and analysis are helpful in making meaningful predictions in scenarios where only a subset of features are available and can potentially be helpful in identifying and preventing relapse early.

Understanding Types of Cyberbullying in an Anonymous Messaging Application

The possibility of anonymity and lack of effective ways to identify inappropriate messages have resulted in a significant amount of online interaction data that attempt to harass, bully, or offend the recipient. In this work, we perform a preliminary linguistic study on messages exchanged using one such popular web/smartphone application---Sarahah, that allows friends to exchange messages anonymously. Since messages exchanged via Sarahah are private, we collect them when the recipient shares it on Twitter. We then perform an analysis of the different kinds of messages exchanged through this application. Our linguistic analysis reveals that a significant number of these messages (~20%) include inappropriate, hurtful, or profane language intended to embarrass, offend, or bully the recipient. Our analysis helps in understanding the different ways in which anonymous message exchange platforms are used and the different types of bullying present in such exchanges.

On the goodput of flows in heterogeneous mobile networks

Abstract In practice heterogeneous networks comprising of diverse nodes need to operate efficiently under a wide range of node mobility and link quality regimes. In this paper, we propose algorithms to determine the goodput of flows in heterogeneous mobile networks. We consider a scenario where some network nodes operate as routers while others operate as flooders, based on the underlying forwarding policy. When a node operates as a router, it forwards packets based on the routing table as determined by the underlying routing algorithm and when it operates as a flooder, it broadcasts packets to all its neighbors. We begin with the case of a single network flow and demonstrate that the problem of determining the goodput is challenging even for this simple setting. We construct a Bayesian network, and propose an algorithm based on the sum-product algorithm to determine the exact goodput. We extend the proposed Bayesian network model for exact goodput calculation to feed forward networks with multiple flows. For a general network with multiple flows, the problem becomes more challenging. The difficulty of the problem stems from the fact that node pairs can forward traffic to one another, resulting in cyclical dependencies. We propose a fixed-point approximation to determine the goodput in this case. Finally, we present an application scenario, where we leverage the fixed-point approximation to design a forwarding strategy adaptive-flood that adapts seamlessly to varying networking conditions. We perform simulations and show that adaptive-flood can effectively classify individual nodes as routers/flooders, achieving performance equivalent to, and in some cases significantly better than that of network-wide routing or flooding alone.

Multi-relational influence models for online professional networks

Professional networks are a specialized class of social networks that are particularly aimed at forming and strengthening professional connections and have become a vital component of professional success and growth. In this paper, we present a holistic model to jointly represent different heterogenous relationships between pairs of individuals, user actions and their respective propagations to characterize influence in online professional networks. Previous work on influence in social networks typically only consider a single action type in characterizing influence. Our model is capable of representing and combining different kinds of information users assimilate in the network and compute pairwise values of influence taking the different types of actions into account. We evaluate our models on data from the largest professional network, LinkedIn and show the effectiveness of the inferred influence scores in predicting user actions. We further demonstrate that modeling different user actions, node features, and edge relationships between users leads to around 20% increase in precision at top k in predicting user actions, when compared to the current state-of-the-art model.

A Probabilistic Approach to Modeling Socio-Behavioral Interactions

The vast growth and reach of internet and social media have led to a tremendous increase in socio-behavioral interaction content on the web. The ever-increasing number of online interactions have led to a growing interest to understand and interpret online communications to enhance user experience. This includes personalization, user retention, predicting user interests, and product recommendations. In this thesis, I address how to use machine learning methods to model sociobehavioral interactions and predict user behavior patterns in online networks. In the first part of this proposal, I focus on one such emerging online interaction platform—online courses (MOOCs). Structured data from these courses contain behavioral, and interaction data and provide opportunity to design machine learning methods for understanding user interaction. The data also contains unstructured data, such as natural language text from forum posts and other online discussions. I present a family of probabilistic models that I have developed for: 1) modeling student engagement, 2) predicting student completion and dropouts, 3) modeling student sentiment toward various course aspects (e.g., content vs. logistics), and 4) detecting coarse and fine-grained course aspects (e.g., grading, video, content) in online courses. These methods have the potential to improve student experience and focus limited instructor resources in ways that will have the most impact. In the second part of the proposal, I describe how I plan to extend the above-mentioned models to model socio-behavioral interactions at multiple scales in networks. I plan to test the effectiveness of this model via experimentation on different types of platforms such as MOOCs and professional networks (e.g., LinkedIn).