Ashish Sureka

Potholes and bad road conditions: mining Twitter to extract information on killer roads

Research shows that Twitter is being used as a platform to not only share and disseminate the information but also collecting complaints from citizens. However, due to the presence of high volume and large stream data, real-time manual identification of those complaints is overwhelmingly impractical. In this paper, we identify the complaints and grievances posted on bad road conditions causing life risks, discomfort and poor road experience to the citizens. We formulate the problem of killer road complaints identification as a multiclass text classification problem. We address the challenge of keyword based flagging methods and identify several linguistic features that are unique for the killer road complaints tweets such as the issue reported in the complaint, pinpoint location of the issue, city or region location information. Our results reveal that not all complaint reports posted to Public agencies contain the sufficient information and are not useful. Therefore, we further propose a mechanism to enrich the nearly-useful tweets and convert them into useful reports. We present our results using information visualization and gain actionable insights from them. Our results show that the proposed features are discriminatory and able to classify killer roads complaints with an accuracy of 67% and a recall of 65%.

A Case Study on the Application of Case-Based Learning in Software Testing

Software testing is a popular mean of examining the adequacy of a developed product. However, in academic institutions more emphasis is given to software development than ensuring its quality. In order to address the gaps between existing university-level software testing education and the training standards used in industry, we experiment with employing a popular teaching method Case-Based Learning (CBL) for the first time to facilitate the training of selected software testing concepts at tertiary-level. The CBL exercise is conducted for undergraduate students of DAIICT, Gandhinagar (India) to cultivate the decision making skills in a self-learning environment. After the CBL execution we collect students responses through a short survey and perform an empirical analysis on the survey results. The outcome of this CBL practice is positive as a majority of students are able to achieve the five stated objectives of CBL. We examine that there is a statistically significant difference between students responses based on gender diversity. We also investigate the difference in students feedback to the two different CBL cases that we use for practicing some aspects of software testing. Moreover, we draw useful inferences from the opinions of TAs (Teaching Assistants) about the CBL sessions.

Teaching requirements engineering concepts using case-based learning

Requirements Engineering (RE) is known to be critical for the success of software projects, and hence forms an important part of any Software Engineering (SE) education curriculum offered at tertiary level. In this paper, we report the results of an exploratory pilot study conducted to assess the effectiveness of Case-Based Learning (CBL) methodology in facilitating the learning of several RE concepts. The evaluation was made on the basis of graduate students responses to a set of questions representing various key learning principles, collected after the execution of two CBL sessions at DA-IICT, Gandhinagar (India). We investigate the perceived effectiveness of CBL in students learning of various RE concepts, based on factors like case difference, gender diversity, and team size. Additionally, we collect and analyse the Teaching Assistants (TAs) opinions about the conducted CBL sessions. The outcome of this CBL exercise was positive as maximum students were able to achieve all the five stated learning objectives. The authors also report various challenges, recommendations, and lessons learned while experiencing CBL sessions.

Feature Selection Techniques to Counter Class Imbalance Problem for Aging Related Bug Prediction: Aging Related Bug Prediction

Aging-Related Bugs (ARBs) occur in long running systems due to error conditions caused because of accumulation of problems such as memory leakage or unreleased files and locks. Aging-Related Bugs are hard to discover during software testing and also challenging to replicate. Automatic identification and prediction of aging related fault-prone files and classes in an object oriented system can help the software quality assurance team to optimize their testing efforts. In this paper, we present a study on the application of static source code metrics and machine learning techniques to predict aging related bugs. We conduct a series of experiments on publicly available dataset from two large open-source software systems: Linux and MySQL. Class imbalance and high dimensionality are the two main technical challenges in building effective predictors for aging related bugs. We investigate the application of five different feature selection techniques (OneR, Information Gain, Gain Ratio, RELEIF and Symmetric Uncertainty) for dimensionality reduction and five different strategies (Random Under-sampling, Random Oversampling, SMOTE, SMOTEBoost and RUSBoost) to counter the effect of class imbalance in our proposed machine learning based solution approach. Experimental results reveal that the random under-sampling approach performs best followed by RUSBoost in-terms of the mean AUC metric. Statistical significance test demonstrates that there is a significant difference between the performance of the various feature selection techniques. Experimental results shows that Gain Ratio and RELEIF performs best in comparison to other strategies to address the class imbalance problem. We infer from the statistical significance test that there is no difference between the performances of the five different learning algorithms.

Automatic email response suggestion for support departments within a university

9 Background. Office of Academic Affairs (OAA), Office of Student Life (OSL) and Information Technology Helpdesk (ITD) are support functions within a university which receives hundreds of email messages on the daily basis. A large percentage of emails received by these departments are frequent and commonly used queries or request for information. Responding to every query by manually typing is a tedious and time consuming task and an automated approach for email response suggestion can save lot of time. 10

Deep learning for conflicting statements detection in text

9 Background. Automatic contradiction detection or conflicting statements detection in text consists of identifying discrepancy, inconsistency and defiance in text and has several real world applications in questions and answering systems, multi-document summarization, dispute detection and finder in news, and detection of contradictions in opinions and sentiments on social media. Automatic contradiction detection is a technically challenging natural language processing problem. Contradiction detection between sources of text or two sentence pairs can be framed as a classification problem. 10

An empirical analysis of machine learning models for automated essay grading

9 Background. Automated Essay Scoring (AES) is an area which falls at the intersection of computing and linguistics. AES systems conduct a linguistic analysis of a given essay or prose and then estimates the writing skill or the essay quality in the form a numeric score or a letter grade. AES systems are useful for the school, university and testing company community for efficiently and effectively scaling the task of grading a large number of essays. 10

An Experience Report on Workshop on Emerging Software Engineering Education

This experience report presents the summary, outcomes and experiences of the 2nd Workshop on Emerging Software Engineering Education (WESEE), co-located with Innovations in Software Engineering Con...

Investigating the Role of Twitter in E-Governance by Extracting Information on Citizen Complaints and Grievances Reports

Open Source Social Media Intelligence (OSSMInt) is a field that focuses on extracting useful information and actionable insights from publicly available and overt sources of data on social media platforms. There are several applications that can be built by applying OSSMInt techniques on this human-sensor data. In this paper, we present some of the use-cases of OSSMInt that are useful for the public sector agencies for e-governance. E-governance on social media include the identification of complaints and grievances reported online by the public citizens for the government authorities and facilitate public agencies to response those complaints, provide better services and improve their connections with public citizens. We present the basic Natural Language Processing and Machine Learning based framework, tools and techniques within the context of OSSMInt and E-governance. The focus of this paper is on mining user-generated content on Twitter (the most popular social media and microblogging website) to identify public citizens complaints and grievances. In particular, we focus on two important applications: (1) complaints which are reported to spread awareness among other citizens and to bring government’s attention to the issues reported in the complaint, and (2) complaints which seek for immediate action and response from the concerned authorities. In addition to the basic introduction and motivation, we will discuss the unique challenges to these applications, open research problems, important literature, proposed approach, experimental results, and future directions.