- You are here:
- Q-Labs Research Projects
Q-Labs Research Projects
A technique to measure comprehension burden:
- We live in a world where knowledge is expanding exponentially in many fields of study. It is very important for college students to interact with textbooks of high quality and increasing complexity as they prepare for a career. However, the textbooks that are used in the educational institutes fall way below the standards that are required. Thereby creating a gap between what a student learns and what is expected. To address this gap, measuring text complexity plays a vital role. Text mining techniques can help improve the quality of the text books and alleviate the comprehension burden among the students. There has been a fair amount of literature in the field of on measuring the text complexity however these works are limited to the quantitative measures such as word length, sentence length etc. Although there is literature which applies machine learning algorithms such as SVM, statistical probability models, they represent a document as bag of words (unigram) or as n-grams either of the two representations doesn’t capture the complete context to which the terms are related, capturing the context will be useful to identify the semantics of the text. In this work, we propose a novel approach to measure text complexity/comprehension burden from context of occurrence of terms using network analysis
Automatic evaluation of subjective answers:
- Objective-type answers can be evaluated with far greater ease than subjective or essay-type answers. Hence most of the popular competitive exams, such as GRE, CAT, GMAT, IIT-JEE and AIEEE, are based on multiple choice questions. However, this is only an arrangement of convenience. Essay answers can provide far much better assessment of a student’s knowledge and understanding. If only ways can be devised to assess essay answers with less trouble, subjective questions would be back in vogue. Our education system as such would stand to benefit greatly. Towards this end we propose a fairly new approach to scoring subjective answers automatically. In this paper, we demonstrate methods to detect similarity between the correct answer and the student’s answer by complex comparing of the contexts. A context is extracted from the text and represented as a lexical co-occurrence graph, similarity is measured by comparing the lexical co-occurrence graphs of the two documents.
Analytics in E-learning:
- E-learning has moved to a new level with the spreading impact of MOOCs (Massively Open Online Courses) in the recent years. But while the notion of offering free online courses from the top universities of the world to the general public is very appealing, it involves many challenges. Assessing students’ knowledge and performance on such a large scale with geographical distribution across the globe is a formidable problem. However there are various data-mining techniques being developed that can detect interesting and meaningful patterns in the inputs coming from the students, which can then help in making more reliable automatic assessments. . In this work, we make an attempt to address, improving user experience, building a virtual class room by improving the class room experience, personalization to replicate a teacher in the virtual class room.
Auto conversion of textbooks to interactive digital versions:
- With the growing popularity of MOOCs and other forms of e-learning, there is a great need for developing digital versions of widely used textbooks. While it may not be a difficult thing to make a good pdf copy of a textbook, there is scope for making these books more interesting to the learners by making them more dynamic and interactive. In this paper, we make an attempt to make a digital book smart by auto locating relevant images, videos and audio, scoring the complexity of the text, auto suggesting alternate less complex web articles for better understanding.
A technique to characterize revenue leakage in Customs, through automatic detection of wrongly-entered consignment data:
- Customs clearance of import and export consignments is a highly complex process. Any wrong-entering of data, intentional or otherwise, can lead to substantial revenue leakage for the country. Consequently, it is critical to detect wrongly-entered data, and to characterize and prevent revenue leakage. In this work, we present a novel approach using data mining and text mining approaches for automatic detection of wrongly-entered consignment data. Data mining is widely used in financial fraud detection. Data of consignments such as its description, value, HS Code and various other fields are analyzed, to build a model. That model can be used to predict if a consignment, and what properties of that consignment, can eventually lead to revenue leakage.