http://repository.iiitd.edu.in/xmlui/handle/123456789/19 2026-04-11T15:39:21Z 2026-04-11T15:39:21Z Correa, Denzil Sureka, Ashish (Advisor) http://repository.iiitd.edu.in/xmlui/handle/123456789/209 2017-07-24T17:16:12Z 2014-12-24T04:19:50Z

Content quality in web 2.0 services : analysis, detection, systems and enhancement Correa, Denzil; Sureka, Ashish (Advisor) There has been a proliferation of Web 2.0 sites on the Internet. Contemporary Web 2.0 sites like Facebook and Twitter are primarily driven by user generated content (UGC). On the other hand, the volume of content by such user contributions has been increasing rapidly. However, user generated content may not conform to the set of guidelines and rules of the websites. Sub-par content can severely a ffect user engagement, retention and also have an adverse impact on information retrieval systems. Therefore, there is an impending need to manage and enhance content quality on Web 2.0 sites. In this thesis, we investigate three broad objectives – (1) Low quality content, (2) Content Quality Systems and (3) Information Retrieval Enhancement. In order to address these objectives, first - we look at low quality questions on a popular programming based community based question answering (CQA) website called Stackoverflow. We analyze user behavior, content patterns and also build supervised machine learning based predictive systems to detect low quality questions. In context to the second objective, we look at enhancing content quality on Issue Tracking Systems - a popular artifact used by developers during the software maintenance lifecycle. We conduct surveys from software practitioners to understand the needs of the community and discover that developers frequently use the Internet for their daily tasks. In order to reduce the context switch for software maintenance professionals, we develop two systems – (i) CQA integration with Issue Tracking Systems and (ii) Web Reference Management Browser Plugin. We develop both these systems to reduce cognitive load on software maintenance professionals during their daily tasks. In context to the third objective, we look at quality enhancement on social media to help information retrieval systems. Concretely, we propose a new algorithm to utilize social interactions to discover homogeneous topic-based communities on a social network. To address the challenge of scalability, our algorithm only visits required portions of the network based on an expectation-maximization approach. Further, we also propose an algorithm for tag recommendation on social media. Specifically, we utilize Twitter to suggest tags to external linked media like Flickr, Youtube and Soundcloud. In conclusion, we look at diff errant perspectives for quality analysis, systems, detection and enhancement of content on Web 2.0 sites.

2014-12-24T04:19:50Z Bhatt, Himanshu S Singh, Richa (Advisor) Vatsa, Mayank (Advisor) http://repository.iiitd.edu.in/xmlui/handle/123456789/120 2017-07-24T17:14:03Z 2014-04-14T04:51:59Z

Emerging covariates of face recognition Bhatt, Himanshu S; Singh, Richa (Advisor); Vatsa, Mayank (Advisor) A covariate in face recognition can be defined as an effect that independently increases the intra-class variability or decreases the inter-class variability or both. Covariates such as pose, illumination, expression, aging, and disguise are established and extensively studied in literature and are categorized as existing covariates of face recognition. However, ever increasing applications of face recognition have instigated many new and exciting scenarios such as matching forensic sketches to mug-shot photos, faces altered due to plastic surgery, low resolution surveillance images, and individual from videos. These covariates are categorized as emerging covariates of face recognition, which is the primary emphasis of this dissertation. One of the important cues in solving crimes and apprehending criminals is matching forensic sketches with digital face images. The first contribution of this dissertation is a memetically optimized multi-scale circular Weber’s local descriptor (MCWLD) for matching forensic sketches with digital face images. This dissertation presents an automated algorithm to extract discriminative information from local regions of both sketches and digital images using MCWLD. An evolutionary memetic optimization is proposed to assign optimal weights to every local facial region to boost the identification performance. Since, forensic sketches and digital images can be of poor quality, a pre-processing technique is also used to enhance the quality of images. Results on different sketch databases, including forensic sketch database, illustrate the efficacy of the proposed algorithm. Widespread acceptability and use of biometrics for person authentication has instigated several techniques for evading identification such as altering facial appearance using surgical procedures. These procedures modify both the shape and texture of facial features to varying degrees and thus degrade the performance of face recognition when matching pre- and post-surgery images. The second contribution of this dissertation is a multi-objective evolutionary granular algorithm for matching face images altered due to plastic surgery procedures. The algorithm first generates non-disjoint face granules at multiple levels of granularity. The granular information is assimilated using a multi-objective genetic algorithm that simultaneously optimizes the selection of feature extractor for each face granule along with the weights of individual granules. On IIIT-D plastic surgery database, the proposed algorithm yields the state-of-the-art performance. Face recognition performance degrades when a low resolution face image captured in unconstrained settings, such as surveillance, is matched with high resolution gallery images. The primary challenge is to extract discriminative features from the limited biometric content in low resolution images and match it with information-rich high resolution face images. The problem of cross-resolution face matching is further alleviated when there is limited labeled low resolution training data. The third contribution of this dissertation is co-transfer learning framework, a cross pollination of transfer learning and co-training paradigms, for enhancing the performance of cross-resolution face recognition. The transfer learning component transfers the knowledge that is learned while matching high resolution face images during training for matching low resolution probe images with high resolution gallery during testing. On the other hand, co-training component facilitates this knowledge transfer by assigning pseudo labels to unlabeled probe instances in the target domain. Experiments on a synthetic, three low resolution surveillance quality face databases, and real world examples show the efficacy of the proposed co-transfer learning algorithm as compared to other approaches. Due to prevalent applications and availability of large intra-personal variations, videos have gained significant attention for face recognition. Unlike still face images, videos provide abundant information that can be leveraged to compensate for variations in intra-personal variations and enhance face recognition performance. The fourth contribution of this dissertation is a video based face recognition algorithm which computes a discriminative video signature as an ordered (ranked) list of still face images from a large dictionary. A three stage approach is developed for optimizing ranked lists across multiple video frames and fusing them into a single composite ordered list to compute the video signature. The signature embeds diverse intra-personal variations and facilitates in matching two videos across large variations. Results obtained on Youtube and MBGC v2 video databases show the effectiveness of the proposed algorithm.

2014-04-14T04:51:59Z Yadav, Kuldeep Naik, Vinayak (Advisor) Singh, Pushpendra (Advisor) Singh, Amarjeet (Advisor) http://repository.iiitd.edu.in/xmlui/handle/123456789/114 2018-04-26T10:43:01Z 2014-01-21T11:26:00Z

Geo-localization and location-aware opportunistic communication for mobile phones Yadav, Kuldeep; Naik, Vinayak (Advisor); Singh, Pushpendra (Advisor); Singh, Amarjeet (Advisor) Location-based mobile applications are steadily gaining popularity across the world. These applications require information about user's current location to access different kind of services. However, location-based applications have diverse set of requirements, some of them require location information intermittently such as local search, whereas other applications require continuous access to location information i.e. ones which need to infer high level information such as places and routes. Additionally, localization accuracy requirements are di erent across various locationbased services. For instance, navigation applications require high level of accuracy (¤ 10 meters) whereas sharing location with online social networks may su ce with an accuracy of hundreds of meters. There are mainly three di erent localization approaches which are used to estimate current user location using a mobile phone, i.e. Global Positioning System (GPS), WiFi-based, and GSM-based. These three di erent approaches di er in terms of localization accuracy, availability, and energy consumption. GPS and WiFi-based approaches provide ne grained localization accuracy but there are many phones, which do not have GPS and WiFi sensors (i.e. feature phones). It is predicted that for the at least next ve years, over 50% of the phones will not have GPS. Apart from limited availability, GPS and WiFi-based approaches result in high energy consumption specially for the services which require continuous tracking of location information. Further, many cities in the world do not have a large scale Wi-Fi infrastructure, which is a sole requirement for all vi WiFi-based approaches. GSM-based approaches (Cell ID-based) work on both feature phones as well as smartphones and energy-e cient as compared to GPS/WiFi. However, they require access to a comprehensive database of Cell IDs created using war-driving. Such a database either does not exist or have limited coverage in developing countries. In this thesis, we make the following contributions to enable energy-e cient geolocalization and location-aware communication on mobile phones: (1) We propose a novel Cell Broadcast (CBS) based localization system, which removes the necessity of war-driving or building a Cell ID database for GSM-based localization. Evaluation using self-collected real world traces show that the proposed approach provide good accuracy (nearly 400 - 500 meters), which is su cient for enabling many locationbased services on feature phones. We have developed several location-aware applications using CBS-based approach and combined it with existing techniques such as Cell ID and GPS for improving localization availability while minimizing energy consumption on smartphones. (2) We propose PlaceMap, a system to discover places and routes visited by mobile users based on only Cell ID information. Our system employs a novel graph-based clustering algorithm, which handles challenges such as uctuating among Cell IDs on same place and segregate Cell IDs according to physical places. To provide better accuracy in place discovery, we design algorithm that uses an initial training of WiFi/GPS data to learn places and later use Cell ID data only. Our evaluations on two large scale mobility dataset collected in India and Switzerland show that PlaceMap can correctly discover nearly 80% of places as compared to baseline (GPS/WiFi). (3) We build and evaluate designs of two Cloud-enabled mobile systems, which facilitate opportunistic communication among co-located phones. These system are designed speci cally for bandwidth constrained settings. One of them, MobiShare uses the Cloud for scalable content search and an encounter prediction framework to predict encounter time between content source vii and requestor based on their mobility history. Second system, Unity nds social groups, who have similar interests and have frequent encounters to enable collaborative download of mutually interested content from the Internet. (4) We discover aggregated mobility and place visiting patterns of people in developing countries using one CDR (Call Detail Records) dataset collected in Ivory Coast and two negrained location information datasets collected in India and Switzerland. We have compared these mobility patterns with existing studies for developed countries (US and Switzerland) and found several di erences. One of the di erence is that people in developing countries are less likely to travel long distance on weekends as compared to developed countries. With the fast evolution of hardware and software technologies for mobile phones, there has been a large gap created between capabilities of feature phones and smartphones. This thesis tries to ll that gap and provide practical and promising solutions to enable location-based services on both feature phones and smartphones using low energy location interfaces.

2014-01-21T11:26:00Z